Image Classification

TFLite Runtime

No. Model Name Description

1 ResNeXt101 Wide residual network with grouped convolutions for high accuracy.

Steps to Run on QIMSDK

🟡 Float32 - Model Precision

Download Required Files

File	Save as
Model	resnext101_float.tflite
mobilenet.json	mobilenet.json
ai_demo_sample.mp4	ai_demo_sample.mp4

If any downloaded file is a .zip archive, extract it on your host machine before copying: unzip filename.zip

Copy Files to Device

# Replace $HOME to the appropriate device path before running the commands.
# For QLI:    /root
# For Ubuntu: /home/ubuntu
# Modify this based on your platform and ensure files are copied to the correct location on the device.

ssh <user>@<device-ip> "mkdir -p $HOME/{models,labels,media,media/output}"
scp resnext101_float.tflite             <user>@<device-ip>:$HOME/models/
scp mobilenet.json                      <user>@<device-ip>:$HOME/labels/
scp ai_demo_sample.mp4    <user>@<device-ip>:$HOME/media/

Connect to device

# Run from your host machine — replace <user> and <device-ip>
ssh <user>@<device-ip>

Set Environment Variables

Qualcomm Linux

export MODEL_NAME=resnext101_float.tflite
export LABELS_NAME=mobilenet.json
export SRC_VIDEO_NAME=ai_demo_sample.mp4

Run the Pipeline

gst-launch-1.0 -e --gst-debug=2 \
filesrc location=$HOME/media/$SRC_VIDEO_NAME ! qtdemux ! h264parse ! \
v4l2h264dec capture-io-mode=4 output-io-mode=4 ! video/x-raw,format=NV12 ! queue ! \
tee name=t ! qtimetamux name=class_mux ! qtivoverlay ! waylandsink fullscreen=true sync=true \
t. ! queue ! qtimlvconverter ! queue ! \
qtimltflite model=$HOME/models/$MODEL_NAME delegate=gpu ! queue ! \
qtimlpostprocess module=mobilenet-softmax labels=$HOME/labels/$LABELS_NAME settings="{\"confidence\": 31.0}" ! text/x-raw ! queue ! class_mux.

🟢 W8A8 — Model Precision

Download Required Files

File	Save as
Model	resnext101_w8a8.tflite
mobilenet.json	mobilenet.json
ai_demo_sample.mp4	ai_demo_sample.mp4

If any downloaded file is a .zip archive, extract it on your host machine before copying: unzip filename.zip

Copy Files to Device

# Replace $HOME to the appropriate device path before running the commands.
# For QLI:    /root
# For Ubuntu: /home/ubuntu
# Modify this based on your platform and ensure files are copied to the correct location on the device.

ssh <user>@<device-ip> "mkdir -p $HOME/{models,labels,media,media/output}"
scp resnext101_w8a8.tflite                <user>@<device-ip>:$HOME/models/
scp mobilenet.json                        <user>@<device-ip>:$HOME/labels/
scp ai_demo_sample.mp4      <user>@<device-ip>:$HOME/media/

Connect to device

# Run from your host machine — replace <user> and <device-ip>
ssh <user>@<device-ip>

Set Environment Variables

Qualcomm Linux

export MODEL_NAME=resnext101_w8a8.tflite
export LABELS_NAME=mobilenet.json
export SRC_VIDEO_NAME=ai_demo_sample.mp4

Run the Pipeline

gst-launch-1.0 -e --gst-debug=2 \
filesrc location=$HOME/media/$SRC_VIDEO_NAME ! qtdemux ! h264parse ! \
v4l2h264dec capture-io-mode=4 output-io-mode=4 ! video/x-raw,format=NV12 ! queue ! \
tee name=t ! qtimetamux name=class_mux ! qtivoverlay ! waylandsink fullscreen=true sync=true \
t. ! queue ! qtimlvconverter ! queue ! \
qtimltflite model=$HOME/models/$MODEL_NAME delegate=external external-delegate-path=libQnnTFLiteDelegate.so external-delegate-options="QNNExternalDelegate,backend_type=htp,log_level=(string)1;" ! queue ! \
qtimlpostprocess module=mobilenet-softmax labels=$HOME/labels/$LABELS_NAME settings="{\"confidence\": 35.0}" ! text/x-raw ! queue ! class_mux.

2 VIT Vision Transformer applying self-attention on image patches.

Steps to Run on QIMSDK

🟢 W8A8 — Model Precision

Download Required Files

File	Save as
Model	vit_w8a8.tflite
mobilenet.json	mobilenet.json
ai_demo_sample.mp4	ai_demo_sample.mp4

If any downloaded file is a .zip archive, extract it on your host machine before copying: unzip filename.zip

Copy Files to Device

# Replace $HOME to the appropriate device path before running the commands.
# For QLI:    /root
# For Ubuntu: /home/ubuntu
# Modify this based on your platform and ensure files are copied to the correct location on the device.

ssh <user>@<device-ip> "mkdir -p $HOME/{models,labels,media,media/output}"
scp vit_w8a8.tflite                    <user>@<device-ip>:$HOME/models/
scp mobilenet.json                     <user>@<device-ip>:$HOME/labels/
scp ai_demo_sample.mp4   <user>@<device-ip>:$HOME/media/

Connect to device

# Run from your host machine — replace <user> and <device-ip>
ssh <user>@<device-ip>

Set Environment Variables

Qualcomm Linux

export MODEL_NAME=vit_w8a8.tflite
export LABELS_NAME=mobilenet.json
export SRC_VIDEO_NAME=ai_demo_sample.mp4

Run the Pipeline

gst-launch-1.0 -e --gst-debug=2 \
filesrc location=$HOME/media/$SRC_VIDEO_NAME ! qtdemux ! h264parse ! \
v4l2h264dec capture-io-mode=4 output-io-mode=4 ! video/x-raw,format=NV12 ! queue ! \
tee name=t ! qtimetamux name=class_mux ! qtivoverlay ! waylandsink fullscreen=true sync=true \
t. ! queue ! qtimlvconverter ! queue ! \
qtimltflite model=$HOME/models/$MODEL_NAME delegate=external external-delegate-path=libQnnTFLiteDelegate.so external-delegate-options="QNNExternalDelegate,backend_type=htp,log_level=(string)1;" ! queue ! \
qtimlpostprocess module=mobilenet-softmax labels=$HOME/labels/$LABELS_NAME settings="{\"confidence\": 35.0}" ! text/x-raw ! queue ! class_mux.

3 EfficientViT-b2-cls Memory-efficient vision transformer (B2 classification).

Steps to Run on QIMSDK

🟡 Float32 - Model Precision

Download Required Files

File	Save as
Model	efficientvit_b2_cls_float.tflite
mobilenet.json	mobilenet.json
ai_demo_sample.mp4	ai_demo_sample.mp4

If any downloaded file is a .zip archive, extract it on your host machine before copying: unzip filename.zip

Copy Files to Device

# Replace $HOME to the appropriate device path before running the commands.
# For QLI:    /root
# For Ubuntu: /home/ubuntu
# Modify this based on your platform and ensure files are copied to the correct location on the device.

ssh <user>@<device-ip> "mkdir -p $HOME/{models,labels,media,media/output}"
scp efficientvit_b2_cls_float.tflite   <user>@<device-ip>:$HOME/models/
scp mobilenet.json                     <user>@<device-ip>:$HOME/labels/
scp ai_demo_sample.mp4   <user>@<device-ip>:$HOME/media/

Connect to device

# Run from your host machine — replace <user> and <device-ip>
ssh <user>@<device-ip>

Set Environment Variables

Qualcomm Linux

export MODEL_NAME=efficientvit_b2_cls_float.tflite
export LABELS_NAME=mobilenet.json
export SRC_VIDEO_NAME=ai_demo_sample.mp4

Run the Pipeline

gst-launch-1.0 -e --gst-debug=2 \
filesrc location=$HOME/media/$SRC_VIDEO_NAME ! qtdemux ! h264parse ! \
v4l2h264dec capture-io-mode=4 output-io-mode=4 ! video/x-raw,format=NV12 ! queue ! \
tee name=t ! qtimetamux name=class_mux ! qtivoverlay ! waylandsink fullscreen=true sync=true \
t. ! queue ! qtimlvconverter ! queue ! \
qtimltflite model=$HOME/models/$MODEL_NAME delegate=gpu ! queue ! \
qtimlpostprocess module=mobilenet-softmax labels=$HOME/labels/$LABELS_NAME settings="{\"confidence\": 51.0}" ! text/x-raw ! queue ! class_mux.

4 EfficientViT-l2-cls Larger memory-efficient vision transformer (L2 classification).

Steps to Run on QIMSDK

🟡 Float32 - Model Precision

Download Required Files

File	Save as
Model	efficientvit_l2_cls_float.tflite
mobilenet.json	mobilenet.json
ai_demo_sample.mp4	ai_demo_sample.mp4

If any downloaded file is a .zip archive, extract it on your host machine before copying: unzip filename.zip

Copy Files to Device

# Replace $HOME to the appropriate device path before running the commands.
# For QLI:    /root
# For Ubuntu: /home/ubuntu
# Modify this based on your platform and ensure files are copied to the correct location on the device.

ssh <user>@<device-ip> "mkdir -p $HOME/{models,labels,media,media/output}"
scp efficientvit_l2_cls_float.tflite   <user>@<device-ip>:$HOME/models/
scp mobilenet.json                     <user>@<device-ip>:$HOME/labels/
scp ai_demo_sample.mp4   <user>@<device-ip>:$HOME/media/

Connect to device

# Run from your host machine — replace <user> and <device-ip>
ssh <user>@<device-ip>

Set Environment Variables

Qualcomm Linux

export MODEL_NAME=efficientvit_l2_cls_float.tflite
export LABELS_NAME=mobilenet.json
export SRC_VIDEO_NAME=ai_demo_sample.mp4

Run the Pipeline

gst-launch-1.0 -e --gst-debug=2 \
filesrc location=$HOME/media/$SRC_VIDEO_NAME ! qtdemux ! h264parse ! \
v4l2h264dec capture-io-mode=4 output-io-mode=4 ! video/x-raw,format=NV12 ! queue ! \
tee name=t ! qtimetamux name=class_mux ! qtivoverlay ! waylandsink fullscreen=true sync=true \
t. ! queue ! qtimlvconverter ! queue ! \
qtimltflite model=$HOME/models/$MODEL_NAME delegate=gpu ! queue ! \
qtimlpostprocess module=mobilenet-softmax labels=$HOME/labels/$LABELS_NAME settings="{\"confidence\": 51.0}" ! text/x-raw ! queue ! class_mux.

5 EfficientNet-B0 Balanced accuracy/efficiency via compound scaling (B0 base).

Steps to Run on QIMSDK

🟡 Float32 - Model Precision

Download Required Files

File	Save as
Model	efficientnet_b0_float.tflite
mobilenet.json	mobilenet.json
ai_demo_sample.mp4	ai_demo_sample.mp4

If any downloaded file is a .zip archive, extract it on your host machine before copying: unzip filename.zip

Copy Files to Device

# Replace $HOME to the appropriate device path before running the commands.
# For QLI:    /root
# For Ubuntu: /home/ubuntu
# Modify this based on your platform and ensure files are copied to the correct location on the device.

ssh <user>@<device-ip> "mkdir -p $HOME/{models,labels,media,media/output}"
scp efficientnet_b0_float.tflite      <user>@<device-ip>:$HOME/models/
scp mobilenet.json                    <user>@<device-ip>:$HOME/labels/
scp ai_demo_sample.mp4  <user>@<device-ip>:$HOME/media/

Connect to device

# Run from your host machine — replace <user> and <device-ip>
ssh <user>@<device-ip>

Set Environment Variables

Qualcomm Linux

export MODEL_NAME=efficientnet_b0_float.tflite
export LABELS_NAME=mobilenet.json
export SRC_VIDEO_NAME=ai_demo_sample.mp4

Run the Pipeline

gst-launch-1.0 -e --gst-debug=2 \
filesrc location=$HOME/media/$SRC_VIDEO_NAME ! qtdemux ! h264parse ! \
v4l2h264dec capture-io-mode=4 output-io-mode=4 ! video/x-raw,format=NV12 ! queue ! \
tee name=t ! qtimetamux name=class_mux ! qtivoverlay ! waylandsink fullscreen=true sync=true \
t. ! queue ! qtimlvconverter ! queue ! \
qtimltflite model=$HOME/models/$MODEL_NAME delegate=gpu ! queue ! \
qtimlpostprocess module=mobilenet labels=$HOME/labels/$LABELS_NAME settings="{\"confidence\": 51.0}" ! text/x-raw ! queue ! class_mux.

6 ConvNext-Base Pure CNN architecture inspired by ViT design principles (Base).

Steps to Run on QIMSDK

🟡 Float32 - Model Precision

Download Required Files

File	Save as
Model	convnext_base_float.tflite
mobilenet.json	mobilenet.json
ai_demo_sample.mp4	ai_demo_sample.mp4

If any downloaded file is a .zip archive, extract it on your host machine before copying: unzip filename.zip

Copy Files to Device

# Replace $HOME to the appropriate device path before running the commands.
# For QLI:    /root
# For Ubuntu: /home/ubuntu
# Modify this based on your platform and ensure files are copied to the correct location on the device.

ssh <user>@<device-ip> "mkdir -p $HOME/{models,labels,media,media/output}"
scp convnext_base_float.tflite         <user>@<device-ip>:$HOME/models/
scp mobilenet.json                     <user>@<device-ip>:$HOME/labels/
scp ai_demo_sample.mp4   <user>@<device-ip>:$HOME/media/

Connect to device

# Run from your host machine — replace <user> and <device-ip>
ssh <user>@<device-ip>

Set Environment Variables

Qualcomm Linux

export MODEL_NAME=convnext_base_float.tflite
export LABELS_NAME=mobilenet.json
export SRC_VIDEO_NAME=ai_demo_sample.mp4

Run the Pipeline

gst-launch-1.0 -e --gst-debug=2 \
filesrc location=$HOME/media/$SRC_VIDEO_NAME ! qtdemux ! h264parse ! \
v4l2h264dec capture-io-mode=4 output-io-mode=4 ! video/x-raw,format=NV12 ! queue ! \
tee name=t ! qtimetamux name=class_mux ! qtivoverlay ! waylandsink fullscreen=true sync=true \
t. ! queue ! qtimlvconverter ! queue ! \
qtimltflite model=$HOME/models/$MODEL_NAME delegate=gpu ! queue ! \
qtimlpostprocess module=mobilenet labels=$HOME/labels/$LABELS_NAME settings="{\"confidence\": 51.0}" ! text/x-raw ! queue ! class_mux.

7 ConvNext-Tiny Compact ConvNeXt variant for efficient classification.

Steps to Run on QIMSDK

🟡 Float32 - Model Precision

Download Required Files

File	Save as
Model	convnext_tiny_float.tflite
mobilenet.json	mobilenet.json
ai_demo_sample.mp4	ai_demo_sample.mp4

If any downloaded file is a .zip archive, extract it on your host machine before copying: unzip filename.zip

Copy Files to Device

# Replace $HOME to the appropriate device path before running the commands.
# For QLI:    /root
# For Ubuntu: /home/ubuntu
# Modify this based on your platform and ensure files are copied to the correct location on the device.

ssh <user>@<device-ip> "mkdir -p $HOME/{models,labels,media,media/output}"
scp convnext_tiny_float.tflite        <user>@<device-ip>:$HOME/models/
scp mobilenet.json                    <user>@<device-ip>:$HOME/labels/
scp ai_demo_sample.mp4  <user>@<device-ip>:$HOME/media/

Connect to device

# Run from your host machine — replace <user> and <device-ip>
ssh <user>@<device-ip>

Set Environment Variables

Qualcomm Linux

export MODEL_NAME=convnext_tiny_float.tflite
export LABELS_NAME=mobilenet.json
export SRC_VIDEO_NAME=ai_demo_sample.mp4

Run the Pipeline

gst-launch-1.0 -e --gst-debug=2 \
filesrc location=$HOME/media/$SRC_VIDEO_NAME ! qtdemux ! h264parse ! \
v4l2h264dec capture-io-mode=4 output-io-mode=4 ! video/x-raw,format=NV12 ! queue ! \
tee name=t ! qtimetamux name=class_mux ! qtivoverlay ! waylandsink fullscreen=true sync=true \
t. ! queue ! qtimlvconverter ! queue ! \
qtimltflite model=$HOME/models/$MODEL_NAME delegate=gpu ! queue ! \
qtimlpostprocess module=mobilenet labels=$HOME/labels/$LABELS_NAME settings="{\"confidence\": 51.0}" ! text/x-raw ! queue ! class_mux.

8 EfficientNet-V2-s Faster training EfficientNet V2 small variant.

Steps to Run on QIMSDK

🟡 Float32 - Model Precision

Download Required Files

File	Save as
Model	efficientnet_v2_s_float.tflite
mobilenet.json	mobilenet.json
ai_demo_sample.mp4	ai_demo_sample.mp4

If any downloaded file is a .zip archive, extract it on your host machine before copying: unzip filename.zip

Copy Files to Device

# Replace $HOME to the appropriate device path before running the commands.
# For QLI:    /root
# For Ubuntu: /home/ubuntu
# Modify this based on your platform and ensure files are copied to the correct location on the device.

ssh <user>@<device-ip> "mkdir -p $HOME/{models,labels,media,media/output}"
scp efficientnet_v2_s_float.tflite     <user>@<device-ip>:$HOME/models/
scp mobilenet.json                     <user>@<device-ip>:$HOME/labels/
scp ai_demo_sample.mp4   <user>@<device-ip>:$HOME/media/

Connect to device

# Run from your host machine — replace <user> and <device-ip>
ssh <user>@<device-ip>

Set Environment Variables

Qualcomm Linux

export MODEL_NAME=efficientnet_v2_s_float.tflite
export LABELS_NAME=mobilenet.json
export SRC_VIDEO_NAME=ai_demo_sample.mp4

Run the Pipeline

gst-launch-1.0 -e --gst-debug=2 \
filesrc location=$HOME/media/$SRC_VIDEO_NAME ! qtdemux ! h264parse ! \
v4l2h264dec capture-io-mode=4 output-io-mode=4 ! video/x-raw,format=NV12 ! queue ! \
tee name=t ! qtimetamux name=class_mux ! qtivoverlay ! waylandsink fullscreen=true sync=true \
t. ! queue ! qtimlvconverter ! queue ! \
qtimltflite model=$HOME/models/$MODEL_NAME delegate=gpu ! queue ! \
qtimlpostprocess module=mobilenet labels=$HOME/labels/$LABELS_NAME settings="{\"confidence\": 51.0}" ! text/x-raw ! queue ! class_mux.

9 Beit BERT pre-training for image transformers.

Steps to Run on QIMSDK

🟡 Float32 - Model Precision

Download Required Files

File	Save as
Model	beit_float.tflite
mobilenet.json	mobilenet.json
ai_demo_sample.mp4	ai_demo_sample.mp4

If any downloaded file is a .zip archive, extract it on your host machine before copying: unzip filename.zip

Copy Files to Device

# Replace $HOME to the appropriate device path before running the commands.
# For QLI:    /root
# For Ubuntu: /home/ubuntu
# Modify this based on your platform and ensure files are copied to the correct location on the device.

ssh <user>@<device-ip> "mkdir -p $HOME/{models,labels,media,media/output}"
scp beit_float.tflite                  <user>@<device-ip>:$HOME/models/
scp mobilenet.json                     <user>@<device-ip>:$HOME/labels/
scp ai_demo_sample.mp4   <user>@<device-ip>:$HOME/media/

Connect to device

# Run from your host machine — replace <user> and <device-ip>
ssh <user>@<device-ip>

Set Environment Variables

Qualcomm Linux

export MODEL_NAME=beit_float.tflite
export LABELS_NAME=mobilenet.json
export SRC_VIDEO_NAME=ai_demo_sample.mp4

Run the Pipeline

gst-launch-1.0 -e --gst-debug=2 \
filesrc location=$HOME/media/$SRC_VIDEO_NAME ! qtdemux ! h264parse ! \
v4l2h264dec capture-io-mode=4 output-io-mode=4 ! video/x-raw,format=NV12 ! queue ! \
tee name=t ! qtimetamux name=class_mux ! qtivoverlay ! waylandsink fullscreen=true sync=true \
t. ! queue ! qtimlvconverter ! queue ! \
qtimltflite model=$HOME/models/$MODEL_NAME delegate=gpu ! queue ! \
qtimlpostprocess module=mobilenet labels=$HOME/labels/$LABELS_NAME settings="{\"confidence\": 51.0}" ! text/x-raw ! queue ! class_mux.

10 MNASNet05 Neural architecture search model for mobile efficiency.

Steps to Run on QIMSDK

🟡 Float32 - Model Precision

Download Required Files

File	Save as
Model	mnasnet05_float.tflite
mobilenet.json	mobilenet.json
ai_demo_sample.mp4	ai_demo_sample.mp4

If any downloaded file is a .zip archive, extract it on your host machine before copying: unzip filename.zip

Copy Files to Device

# Replace $HOME to the appropriate device path before running the commands.
# For QLI:    /root
# For Ubuntu: /home/ubuntu
# Modify this based on your platform and ensure files are copied to the correct location on the device.

ssh <user>@<device-ip> "mkdir -p $HOME/{models,labels,media,media/output}"
scp mnasnet05_float.tflite              <user>@<device-ip>:$HOME/models/
scp mobilenet.json                      <user>@<device-ip>:$HOME/labels/
scp ai_demo_sample.mp4    <user>@<device-ip>:$HOME/media/

Connect to device

# Run from your host machine — replace <user> and <device-ip>
ssh <user>@<device-ip>

Set Environment Variables

Qualcomm Linux

export MODEL_NAME=mnasnet05_float.tflite
export LABELS_NAME=mobilenet.json
export SRC_VIDEO_NAME=ai_demo_sample.mp4

Run the Pipeline

gst-launch-1.0 -e --gst-debug=2 \
filesrc location=$HOME/media/$SRC_VIDEO_NAME ! qtdemux ! h264parse ! \
v4l2h264dec capture-io-mode=4 output-io-mode=4 ! video/x-raw,format=NV12 ! queue ! \
tee name=t ! qtimetamux name=class_mux ! qtivoverlay ! waylandsink fullscreen=true sync=true \
t. ! queue ! qtimlvconverter ! queue ! \
qtimltflite model=$HOME/models/$MODEL_NAME delegate=gpu ! queue ! \
qtimlpostprocess module=mobilenet labels=$HOME/labels/$LABELS_NAME settings="{\"confidence\": 51.0}" ! text/x-raw ! queue ! class_mux.

11 DenseNet-121 Densely connected CNN with feature reuse across layers.

Steps to Run on QIMSDK

🟡 Float32 - Model Precision

Download Required Files

File	Save as
Model	densenet_121_float.tflite
mobilenet.json	mobilenet.json
ai_demo_sample.mp4	ai_demo_sample.mp4

If any downloaded file is a .zip archive, extract it on your host machine before copying: unzip filename.zip

Copy Files to Device

# Replace $HOME to the appropriate device path before running the commands.
# For QLI:    /root
# For Ubuntu: /home/ubuntu
# Modify this based on your platform and ensure files are copied to the correct location on the device.

ssh <user>@<device-ip> "mkdir -p $HOME/{models,labels,media,media/output}"
scp densenet_121_float.tflite           <user>@<device-ip>:$HOME/models/
scp mobilenet.json                      <user>@<device-ip>:$HOME/labels/
scp ai_demo_sample.mp4    <user>@<device-ip>:$HOME/media/

Connect to device

# Run from your host machine — replace <user> and <device-ip>
ssh <user>@<device-ip>

Set Environment Variables

Qualcomm Linux

export MODEL_NAME=densenet_121_float.tflite
export LABELS_NAME=mobilenet.json
export SRC_VIDEO_NAME=ai_demo_sample.mp4

Run the Pipeline

gst-launch-1.0 -e --gst-debug=2 \
filesrc location=$HOME/media/$SRC_VIDEO_NAME ! qtdemux ! h264parse ! \
v4l2h264dec capture-io-mode=4 output-io-mode=4 ! video/x-raw,format=NV12 ! queue ! \
tee name=t ! qtimetamux name=class_mux ! qtivoverlay ! waylandsink fullscreen=true sync=true \
t. ! queue ! qtimlvconverter ! queue ! \
qtimltflite model=$HOME/models/$MODEL_NAME delegate=gpu ! queue ! \
qtimlpostprocess module=mobilenet-softmax labels=$HOME/labels/$LABELS_NAME settings="{\"confidence\": 51.0}" ! text/x-raw ! queue ! class_mux.

12 GoogLeNet Inception-based deep CNN for efficient classification.

Steps to Run on QIMSDK

🟡 Float32 - Model Precision

Download Required Files

File	Save as
Model	googlenet_float.tflite
mobilenet.json	mobilenet.json
ai_demo_sample.mp4	ai_demo_sample.mp4

If any downloaded file is a .zip archive, extract it on your host machine before copying: unzip filename.zip

Copy Files to Device

# Replace $HOME to the appropriate device path before running the commands.
# For QLI:    /root
# For Ubuntu: /home/ubuntu
# Modify this based on your platform and ensure files are copied to the correct location on the device.

ssh <user>@<device-ip> "mkdir -p $HOME/{models,labels,media,media/output}"
scp googlenet_float.tflite              <user>@<device-ip>:$HOME/models/
scp mobilenet.json                      <user>@<device-ip>:$HOME/labels/
scp ai_demo_sample.mp4    <user>@<device-ip>:$HOME/media/

Connect to device

# Run from your host machine — replace <user> and <device-ip>
ssh <user>@<device-ip>

Set Environment Variables

Qualcomm Linux

export MODEL_NAME=googlenet_float.tflite
export LABELS_NAME=mobilenet.json
export SRC_VIDEO_NAME=ai_demo_sample.mp4

Run the Pipeline

gst-launch-1.0 -e --gst-debug=2 \
filesrc location=$HOME/media/$SRC_VIDEO_NAME ! qtdemux ! h264parse ! \
v4l2h264dec capture-io-mode=4 output-io-mode=4 ! video/x-raw,format=NV12 ! queue ! \
tee name=t ! qtimetamux name=class_mux ! qtivoverlay ! waylandsink fullscreen=true sync=true \
t. ! queue ! qtimlvconverter ! queue ! \
qtimltflite model=$HOME/models/$MODEL_NAME delegate=gpu ! queue ! \
qtimlpostprocess module=mobilenet-softmax labels=$HOME/labels/$LABELS_NAME settings="{\"confidence\": 51.0}" ! text/x-raw ! queue ! class_mux.

🟢 W8A8 — Model Precision

Download Required Files

File	Save as
Model	googlenet_w8a8.tflite
mobilenet.json	mobilenet.json
ai_demo_sample.mp4	ai_demo_sample.mp4

If any downloaded file is a .zip archive, extract it on your host machine before copying: unzip filename.zip

Copy Files to Device

# Replace $HOME to the appropriate device path before running the commands.
# For QLI:    /root
# For Ubuntu: /home/ubuntu
# Modify this based on your platform and ensure files are copied to the correct location on the device.

ssh <user>@<device-ip> "mkdir -p $HOME/{models,labels,media,media/output}"
scp googlenet_w8a8.tflite               <user>@<device-ip>:$HOME/models/
scp mobilenet.json                      <user>@<device-ip>:$HOME/labels/
scp ai_demo_sample.mp4    <user>@<device-ip>:$HOME/media/

Connect to device

# Run from your host machine — replace <user> and <device-ip>
ssh <user>@<device-ip>

Set Environment Variables

Qualcomm Linux

export MODEL_NAME=googlenet_w8a8.tflite
export LABELS_NAME=mobilenet.json
export SRC_VIDEO_NAME=ai_demo_sample.mp4

Run the Pipeline

gst-launch-1.0 -e --gst-debug=2 \
filesrc location=$HOME/media/$SRC_VIDEO_NAME ! qtdemux ! h264parse ! \
v4l2h264dec capture-io-mode=4 output-io-mode=4 ! video/x-raw,format=NV12 ! queue ! \
tee name=t ! qtimetamux name=class_mux ! qtivoverlay ! waylandsink fullscreen=true sync=true \
t. ! queue ! qtimlvconverter ! queue ! \
qtimltflite model=$HOME/models/$MODEL_NAME delegate=external external-delegate-path=libQnnTFLiteDelegate.so external-delegate-options="QNNExternalDelegate,backend_type=htp,log_level=(string)1;" ! queue ! \
qtimlpostprocess module=mobilenet labels=$HOME/labels/$LABELS_NAME settings="{\"confidence\": 51.0}" ! text/x-raw ! queue ! class_mux.

13 Resnet101 Deep residual network with 101 layers.

Steps to Run on QIMSDK

🟡 Float32 - Model Precision

Download Required Files

File	Save as
Model	resnet101_float.tflite
mobilenet.json	mobilenet.json
ai_demo_sample.mp4	ai_demo_sample.mp4

If any downloaded file is a .zip archive, extract it on your host machine before copying: unzip filename.zip

Copy Files to Device

# Replace $HOME to the appropriate device path before running the commands.
# For QLI:    /root
# For Ubuntu: /home/ubuntu
# Modify this based on your platform and ensure files are copied to the correct location on the device.

ssh <user>@<device-ip> "mkdir -p $HOME/{models,labels,media,media/output}"
scp resnet101_float.tflite                 <user>@<device-ip>:$HOME/models/
scp mobilenet.json                         <user>@<device-ip>:$HOME/labels/
scp ai_demo_sample.mp4       <user>@<device-ip>:$HOME/media/

Connect to device

# Run from your host machine — replace <user> and <device-ip>
ssh <user>@<device-ip>

Set Environment Variables

Qualcomm Linux

export MODEL_NAME=resnet101_float.tflite
export LABELS_NAME=mobilenet.json
export SRC_VIDEO_NAME=ai_demo_sample.mp4

Run the Pipeline

gst-launch-1.0 -e --gst-debug=2 \
filesrc location=$HOME/media/$SRC_VIDEO_NAME ! qtdemux ! h264parse ! \
v4l2h264dec capture-io-mode=4 output-io-mode=4 ! video/x-raw,format=NV12 ! queue ! \
tee name=t ! qtimetamux name=class_mux ! qtivoverlay ! waylandsink fullscreen=true sync=true \
t. ! queue ! qtimlvconverter ! queue ! \
qtimltflite model=$HOME/models/$MODEL_NAME delegate=gpu ! queue ! \
qtimlpostprocess module=mobilenet-softmax labels=$HOME/labels/$LABELS_NAME settings="{\"confidence\": 51.0}" ! text/x-raw ! queue ! class_mux.

🟢 W8A8 — Model Precision

Download Required Files

File	Save as
Model	resnet101_w8a8.tflite
mobilenet.json	mobilenet.json
ai_demo_sample.mp4	ai_demo_sample.mp4

If any downloaded file is a .zip archive, extract it on your host machine before copying: unzip filename.zip

Copy Files to Device

# Replace $HOME to the appropriate device path before running the commands.
# For QLI:    /root
# For Ubuntu: /home/ubuntu
# Modify this based on your platform and ensure files are copied to the correct location on the device.

ssh <user>@<device-ip> "mkdir -p $HOME/{models,labels,media,media/output}"
scp resnet101_w8a8.tflite               <user>@<device-ip>:$HOME/models/
scp mobilenet.json                      <user>@<device-ip>:$HOME/labels/
scp ai_demo_sample.mp4    <user>@<device-ip>:$HOME/media/

Connect to device

# Run from your host machine — replace <user> and <device-ip>
ssh <user>@<device-ip>

Set Environment Variables

Qualcomm Linux

export MODEL_NAME=resnet101_w8a8.tflite
export LABELS_NAME=mobilenet.json
export SRC_VIDEO_NAME=ai_demo_sample.mp4

Run the Pipeline

gst-launch-1.0 -e --gst-debug=2 \
filesrc location=$HOME/media/$SRC_VIDEO_NAME ! qtdemux ! h264parse ! \
v4l2h264dec capture-io-mode=4 output-io-mode=4 ! video/x-raw,format=NV12 ! queue ! \
tee name=t ! qtimetamux name=class_mux ! qtivoverlay ! waylandsink fullscreen=true sync=true \
t. ! queue ! qtimlvconverter ! queue ! \
qtimltflite model=$HOME/models/$MODEL_NAME delegate=external external-delegate-path=libQnnTFLiteDelegate.so external-delegate-options="QNNExternalDelegate,backend_type=htp,log_level=(string)1;" ! queue ! \
qtimlpostprocess module=mobilenet labels=$HOME/labels/$LABELS_NAME settings="{\"confidence\": 51.0}" ! text/x-raw ! queue ! class_mux.

14 Resnet50 Classic 50-layer residual network for image classification.

Steps to Run on QIMSDK

🟡 Float32 - Model Precision

Download Required Files

File	Save as
Model	resnet50_float.tflite
mobilenet.json	mobilenet.json
ai_demo_sample.mp4	ai_demo_sample.mp4

If any downloaded file is a .zip archive, extract it on your host machine before copying: unzip filename.zip

Copy Files to Device

# Replace $HOME to the appropriate device path before running the commands.
# For QLI:    /root
# For Ubuntu: /home/ubuntu
# Modify this based on your platform and ensure files are copied to the correct location on the device.

ssh <user>@<device-ip> "mkdir -p $HOME/{models,labels,media,media/output}"
scp resnet50_float.tflite               <user>@<device-ip>:$HOME/models/
scp mobilenet.json                      <user>@<device-ip>:$HOME/labels/
scp ai_demo_sample.mp4    <user>@<device-ip>:$HOME/media/

Connect to device

# Run from your host machine — replace <user> and <device-ip>
ssh <user>@<device-ip>

Set Environment Variables

Qualcomm Linux

export MODEL_NAME=resnet50_float.tflite
export LABELS_NAME=mobilenet.json
export SRC_VIDEO_NAME=ai_demo_sample.mp4

Run the Pipeline

gst-launch-1.0 -e --gst-debug=2 \
filesrc location=$HOME/media/$SRC_VIDEO_NAME ! qtdemux ! h264parse ! \
v4l2h264dec capture-io-mode=4 output-io-mode=4 ! video/x-raw,format=NV12 ! queue ! \
tee name=t ! qtimetamux name=class_mux ! qtivoverlay ! waylandsink fullscreen=true sync=true \
t. ! queue ! qtimlvconverter ! queue ! \
qtimltflite model=$HOME/models/$MODEL_NAME delegate=gpu ! queue ! \
qtimlpostprocess module=mobilenet-softmax labels=$HOME/labels/$LABELS_NAME settings="{\"confidence\": 51.0}" ! text/x-raw ! queue ! class_mux.

🟢 W8A8 — Model Precision

Download Required Files

File	Save as
Model	resnet50_w8a8.tflite
mobilenet.json	mobilenet.json
ai_demo_sample.mp4	ai_demo_sample.mp4

If any downloaded file is a .zip archive, extract it on your host machine before copying: unzip filename.zip

Copy Files to Device

# Replace $HOME to the appropriate device path before running the commands.
# For QLI:    /root
# For Ubuntu: /home/ubuntu
# Modify this based on your platform and ensure files are copied to the correct location on the device.

ssh <user>@<device-ip> "mkdir -p $HOME/{models,labels,media,media/output}"
scp resnet50_w8a8.tflite                <user>@<device-ip>:$HOME/models/
scp mobilenet.json                      <user>@<device-ip>:$HOME/labels/
scp ai_demo_sample.mp4    <user>@<device-ip>:$HOME/media/

Connect to device

# Run from your host machine — replace <user> and <device-ip>
ssh <user>@<device-ip>

Set Environment Variables

Qualcomm Linux

export MODEL_NAME=resnet50_w8a8.tflite
export LABELS_NAME=mobilenet.json
export SRC_VIDEO_NAME=ai_demo_sample.mp4

Run the Pipeline

gst-launch-1.0 -e --gst-debug=2 \
filesrc location=$HOME/media/$SRC_VIDEO_NAME ! qtdemux ! h264parse ! \
v4l2h264dec capture-io-mode=4 output-io-mode=4 ! video/x-raw,format=NV12 ! queue ! \
tee name=t ! qtimetamux name=class_mux ! qtivoverlay ! waylandsink fullscreen=true sync=true \
t. ! queue ! qtimlvconverter ! queue ! \
qtimltflite model=$HOME/models/$MODEL_NAME delegate=external external-delegate-path=libQnnTFLiteDelegate.so external-delegate-options="QNNExternalDelegate,backend_type=htp,log_level=(string)1;" ! queue ! \
qtimlpostprocess module=mobilenet labels=$HOME/labels/$LABELS_NAME settings="{\"confidence\": 51.0}" ! text/x-raw ! queue ! class_mux.

15 Inception-v3 Multi-scale feature extraction with factorized convolutions.

Steps to Run on QIMSDK

🟡 Float32 - Model Precision

Download Required Files

File	Save as
Model	inception_v3_float.tflite
mobilenet.json	mobilenet.json
ai_demo_sample.mp4	ai_demo_sample.mp4

If any downloaded file is a .zip archive, extract it on your host machine before copying: unzip filename.zip

Copy Files to Device

# Replace $HOME to the appropriate device path before running the commands.
# For QLI:    /root
# For Ubuntu: /home/ubuntu
# Modify this based on your platform and ensure files are copied to the correct location on the device.

ssh <user>@<device-ip> "mkdir -p $HOME/{models,labels,media,media/output}"
scp inception_v3_float.tflite           <user>@<device-ip>:$HOME/models/
scp mobilenet.json                      <user>@<device-ip>:$HOME/labels/
scp ai_demo_sample.mp4    <user>@<device-ip>:$HOME/media/

Connect to device

# Run from your host machine — replace <user> and <device-ip>
ssh <user>@<device-ip>

Set Environment Variables

Qualcomm Linux

export MODEL_NAME=inception_v3_float.tflite
export LABELS_NAME=mobilenet.json
export SRC_VIDEO_NAME=ai_demo_sample.mp4

Run the Pipeline

gst-launch-1.0 -e --gst-debug=2 \
filesrc location=$HOME/media/$SRC_VIDEO_NAME ! qtdemux ! h264parse ! \
v4l2h264dec capture-io-mode=4 output-io-mode=4 ! video/x-raw,format=NV12 ! queue ! \
tee name=t ! qtimetamux name=class_mux ! qtivoverlay ! waylandsink fullscreen=true sync=true \
t. ! queue ! qtimlvconverter ! queue ! \
qtimltflite model=$HOME/models/$MODEL_NAME delegate=gpu ! queue ! \
qtimlpostprocess module=mobilenet-softmax labels=$HOME/labels/$LABELS_NAME settings="{\"confidence\": 51.0}" ! text/x-raw ! queue ! class_mux.

🟢 W8A8 — Model Precision

Download Required Files

File	Save as
Model	inception_v3_w8a8.tflite
mobilenet.json	mobilenet.json
ai_demo_sample.mp4	ai_demo_sample.mp4

If any downloaded file is a .zip archive, extract it on your host machine before copying: unzip filename.zip

Copy Files to Device

# Replace $HOME to the appropriate device path before running the commands.
# For QLI:    /root
# For Ubuntu: /home/ubuntu
# Modify this based on your platform and ensure files are copied to the correct location on the device.

ssh <user>@<device-ip> "mkdir -p $HOME/{models,labels,media,media/output}"
scp inception_v3_w8a8.tflite            <user>@<device-ip>:$HOME/models/
scp mobilenet.json                      <user>@<device-ip>:$HOME/labels/
scp ai_demo_sample.mp4    <user>@<device-ip>:$HOME/media/

Connect to device

# Run from your host machine — replace <user> and <device-ip>
ssh <user>@<device-ip>

Set Environment Variables

Qualcomm Linux

export MODEL_NAME=inception_v3_w8a8.tflite
export LABELS_NAME=mobilenet.json
export SRC_VIDEO_NAME=ai_demo_sample.mp4

Run the Pipeline

gst-launch-1.0 -e --gst-debug=2 \
filesrc location=$HOME/media/$SRC_VIDEO_NAME ! qtdemux ! h264parse ! \
v4l2h264dec capture-io-mode=4 output-io-mode=4 ! video/x-raw,format=NV12 ! queue ! \
tee name=t ! qtimetamux name=class_mux ! qtivoverlay ! waylandsink fullscreen=true sync=true \
t. ! queue ! qtimlvconverter ! queue ! \
qtimltflite model=$HOME/models/$MODEL_NAME delegate=external external-delegate-path=libQnnTFLiteDelegate.so external-delegate-options="QNNExternalDelegate,backend_type=htp,log_level=(string)1;" ! queue ! \
qtimlpostprocess module=mobilenet labels=$HOME/labels/$LABELS_NAME settings="{\"confidence\": 51.0}" ! text/x-raw ! queue ! class_mux.

16 MobileNet-v2 Inverted residual blocks for mobile image classification.

Steps to Run on QIMSDK

🟡 Float32 - Model Precision

Download Required Files

File	Save as
Model	mobilenet_v2_float.tflite
mobilenet.json	mobilenet.json
ai_demo_sample.mp4	ai_demo_sample.mp4

If any downloaded file is a .zip archive, extract it on your host machine before copying: unzip filename.zip

Copy Files to Device

# Replace $HOME to the appropriate device path before running the commands.
# For QLI:    /root
# For Ubuntu: /home/ubuntu
# Modify this based on your platform and ensure files are copied to the correct location on the device.

ssh <user>@<device-ip> "mkdir -p $HOME/{models,labels,media,media/output}"
scp mobilenet_v2_float.tflite           <user>@<device-ip>:$HOME/models/
scp mobilenet.json                      <user>@<device-ip>:$HOME/labels/
scp ai_demo_sample.mp4    <user>@<device-ip>:$HOME/media/

Connect to device

# Run from your host machine — replace <user> and <device-ip>
ssh <user>@<device-ip>

Set Environment Variables

Qualcomm Linux

export MODEL_NAME=mobilenet_v2_float.tflite
export LABELS_NAME=mobilenet.json
export SRC_VIDEO_NAME=ai_demo_sample.mp4

Run the Pipeline

gst-launch-1.0 -e --gst-debug=2 \
filesrc location=$HOME/media/$SRC_VIDEO_NAME ! qtdemux ! h264parse ! \
v4l2h264dec capture-io-mode=4 output-io-mode=4 ! video/x-raw,format=NV12 ! queue ! \
tee name=t ! qtimetamux name=class_mux ! qtivoverlay ! waylandsink fullscreen=true sync=true \
t. ! queue ! qtimlvconverter ! queue ! \
qtimltflite model=$HOME/models/$MODEL_NAME delegate=gpu ! queue ! \
qtimlpostprocess module=mobilenet-softmax labels=$HOME/labels/$LABELS_NAME settings="{\"confidence\": 51.0}" ! text/x-raw ! queue ! class_mux.

🟢 W8A8 — Model Precision

Download Required Files

File	Save as
Model	mobilenet_v2_w8a8.tflite
mobilenet.json	mobilenet.json
ai_demo_sample.mp4	ai_demo_sample.mp4

If any downloaded file is a .zip archive, extract it on your host machine before copying: unzip filename.zip

Copy Files to Device

# Replace $HOME to the appropriate device path before running the commands.
# For QLI:    /root
# For Ubuntu: /home/ubuntu
# Modify this based on your platform and ensure files are copied to the correct location on the device.

ssh <user>@<device-ip> "mkdir -p $HOME/{models,labels,media,media/output}"
scp mobilenet_v2_w8a8.tflite            <user>@<device-ip>:$HOME/models/
scp mobilenet.json                      <user>@<device-ip>:$HOME/labels/
scp ai_demo_sample.mp4    <user>@<device-ip>:$HOME/media/

Connect to device

# Run from your host machine — replace <user> and <device-ip>
ssh <user>@<device-ip>

Set Environment Variables

Qualcomm Linux

export MODEL_NAME=mobilenet_v2_w8a8.tflite
export LABELS_NAME=mobilenet.json
export SRC_VIDEO_NAME=ai_demo_sample.mp4

Run the Pipeline

gst-launch-1.0 -e --gst-debug=2 \
filesrc location=$HOME/media/$SRC_VIDEO_NAME ! qtdemux ! h264parse ! \
v4l2h264dec capture-io-mode=4 output-io-mode=4 ! video/x-raw,format=NV12 ! queue ! \
tee name=t ! qtimetamux name=class_mux ! qtivoverlay ! waylandsink fullscreen=true sync=true \
t. ! queue ! qtimlvconverter ! queue ! \
qtimltflite model=$HOME/models/$MODEL_NAME delegate=external external-delegate-path=libQnnTFLiteDelegate.so external-delegate-options="QNNExternalDelegate,backend_type=htp,log_level=(string)1;" ! queue ! \
qtimlpostprocess module=mobilenet labels=$HOME/labels/$LABELS_NAME settings="{\"confidence\": 51.0}" ! text/x-raw ! queue ! class_mux.

17 MobileNet-v3-Large Optimized mobile CNN with SE modules (Large variant).

Steps to Run on QIMSDK

🟡 Float32 - Model Precision

Download Required Files

File	Save as
Model	mobilenet_v3_large_float.tflite
mobilenet.json	mobilenet.json
ai_demo_sample.mp4	ai_demo_sample.mp4

If any downloaded file is a .zip archive, extract it on your host machine before copying: unzip filename.zip

Copy Files to Device

# Replace $HOME to the appropriate device path before running the commands.
# For QLI:    /root
# For Ubuntu: /home/ubuntu
# Modify this based on your platform and ensure files are copied to the correct location on the device.

ssh <user>@<device-ip> "mkdir -p $HOME/{models,labels,media,media/output}"
scp mobilenet_v3_large_float.tflite     <user>@<device-ip>:$HOME/models/
scp mobilenet.json                      <user>@<device-ip>:$HOME/labels/
scp ai_demo_sample.mp4    <user>@<device-ip>:$HOME/media/

Connect to device

# Run from your host machine — replace <user> and <device-ip>
ssh <user>@<device-ip>

Set Environment Variables

Qualcomm Linux

export MODEL_NAME=mobilenet_v3_large_float.tflite
export LABELS_NAME=mobilenet.json
export SRC_VIDEO_NAME=ai_demo_sample.mp4

Run the Pipeline

gst-launch-1.0 -e --gst-debug=2 \
filesrc location=$HOME/media/$SRC_VIDEO_NAME ! qtdemux ! h264parse ! \
v4l2h264dec capture-io-mode=4 output-io-mode=4 ! video/x-raw,format=NV12 ! queue ! \
tee name=t ! qtimetamux name=class_mux ! qtivoverlay ! waylandsink fullscreen=true sync=true \
t. ! queue ! qtimlvconverter ! queue ! \
qtimltflite model=$HOME/models/$MODEL_NAME delegate=gpu ! queue ! \
qtimlpostprocess module=mobilenet-softmax labels=$HOME/labels/$LABELS_NAME settings="{\"confidence\": 51.0}" ! text/x-raw ! queue ! class_mux.

18 RegNet Regularized network design with quantized widths and depths.

Steps to Run on QIMSDK

🟡 Float32 - Model Precision

Download Required Files

File	Save as
Model	regnet_float.tflite
mobilenet.json	mobilenet.json
ai_demo_sample.mp4	ai_demo_sample.mp4

If any downloaded file is a .zip archive, extract it on your host machine before copying: unzip filename.zip

Copy Files to Device

# Replace $HOME to the appropriate device path before running the commands.
# For QLI:    /root
# For Ubuntu: /home/ubuntu
# Modify this based on your platform and ensure files are copied to the correct location on the device.

ssh <user>@<device-ip> "mkdir -p $HOME/{models,labels,media,media/output}"
scp regnet_float.tflite                 <user>@<device-ip>:$HOME/models/
scp mobilenet.json                      <user>@<device-ip>:$HOME/labels/
scp ai_demo_sample.mp4    <user>@<device-ip>:$HOME/media/

Connect to device

# Run from your host machine — replace <user> and <device-ip>
ssh <user>@<device-ip>

Set Environment Variables

Qualcomm Linux

export MODEL_NAME=regnet_float.tflite
export LABELS_NAME=mobilenet.json
export SRC_VIDEO_NAME=ai_demo_sample.mp4

Run the Pipeline

gst-launch-1.0 -e --gst-debug=2 \
filesrc location=$HOME/media/$SRC_VIDEO_NAME ! qtdemux ! h264parse ! \
v4l2h264dec capture-io-mode=4 output-io-mode=4 ! video/x-raw,format=NV12 ! queue ! \
tee name=t ! qtimetamux name=class_mux ! qtivoverlay ! waylandsink fullscreen=true sync=true \
t. ! queue ! qtimlvconverter ! queue ! \
qtimltflite model=$HOME/models/$MODEL_NAME delegate=gpu ! queue ! \
qtimlpostprocess module=mobilenet-softmax labels=$HOME/labels/$LABELS_NAME settings="{\"confidence\": 51.0}" ! text/x-raw ! queue ! class_mux.

🟢 W8A8 — Model Precision

Download Required Files

File	Save as
Model	regnet_w8a8.tflite
mobilenet.json	mobilenet.json
ai_demo_sample.mp4	ai_demo_sample.mp4

If any downloaded file is a .zip archive, extract it on your host machine before copying: unzip filename.zip

Copy Files to Device

# Replace $HOME to the appropriate device path before running the commands.
# For QLI:    /root
# For Ubuntu: /home/ubuntu
# Modify this based on your platform and ensure files are copied to the correct location on the device.

ssh <user>@<device-ip> "mkdir -p $HOME/{models,labels,media,media/output}"
scp regnet_w8a8.tflite                  <user>@<device-ip>:$HOME/models/
scp mobilenet.json                      <user>@<device-ip>:$HOME/labels/
scp ai_demo_sample.mp4    <user>@<device-ip>:$HOME/media/

Connect to device

# Run from your host machine — replace <user> and <device-ip>
ssh <user>@<device-ip>

Set Environment Variables

Qualcomm Linux

export MODEL_NAME=regnet_w8a8.tflite
export LABELS_NAME=mobilenet.json
export SRC_VIDEO_NAME=ai_demo_sample.mp4

Run the Pipeline

gst-launch-1.0 -e --gst-debug=2 \
filesrc location=$HOME/media/$SRC_VIDEO_NAME ! qtdemux ! h264parse ! \
v4l2h264dec capture-io-mode=4 output-io-mode=4 ! video/x-raw,format=NV12 ! queue ! \
tee name=t ! qtimetamux name=class_mux ! qtivoverlay ! waylandsink fullscreen=true sync=true \
t. ! queue ! qtimlvconverter ! queue ! \
qtimltflite model=$HOME/models/$MODEL_NAME delegate=external external-delegate-path=libQnnTFLiteDelegate.so external-delegate-options="QNNExternalDelegate,backend_type=htp,log_level=(string)1;" ! queue ! \
qtimlpostprocess module=mobilenet labels=$HOME/labels/$LABELS_NAME settings="{\"confidence\": 51.0}" ! text/x-raw ! queue ! class_mux.

19 ResNet18 Lightweight 18-layer residual network.

Steps to Run on QIMSDK

🟡 Float32 - Model Precision

Download Required Files

File	Save as
Model	resnet18_float.tflite
mobilenet.json	mobilenet.json
ai_demo_sample.mp4	ai_demo_sample.mp4

If any downloaded file is a .zip archive, extract it on your host machine before copying: unzip filename.zip

Copy Files to Device

# Replace $HOME to the appropriate device path before running the commands.
# For QLI:    /root
# For Ubuntu: /home/ubuntu
# Modify this based on your platform and ensure files are copied to the correct location on the device.

ssh <user>@<device-ip> "mkdir -p $HOME/{models,labels,media,media/output}"
scp resnet18_float.tflite               <user>@<device-ip>:$HOME/models/
scp mobilenet.json                      <user>@<device-ip>:$HOME/labels/
scp ai_demo_sample.mp4    <user>@<device-ip>:$HOME/media/

Connect to device

# Run from your host machine — replace <user> and <device-ip>
ssh <user>@<device-ip>

Set Environment Variables

Qualcomm Linux

export MODEL_NAME=resnet18_float.tflite
export LABELS_NAME=mobilenet.json
export SRC_VIDEO_NAME=ai_demo_sample.mp4

Run the Pipeline

gst-launch-1.0 -e --gst-debug=2 \
filesrc location=$HOME/media/$SRC_VIDEO_NAME ! qtdemux ! h264parse ! \
v4l2h264dec capture-io-mode=4 output-io-mode=4 ! video/x-raw,format=NV12 ! queue ! \
tee name=t ! qtimetamux name=class_mux ! qtivoverlay ! waylandsink fullscreen=true sync=true \
t. ! queue ! qtimlvconverter ! queue ! \
qtimltflite model=$HOME/models/$MODEL_NAME delegate=gpu ! queue ! \
qtimlpostprocess module=mobilenet-softmax labels=$HOME/labels/$LABELS_NAME settings="{\"confidence\": 51.0}" ! text/x-raw ! queue ! class_mux.

🟢 W8A8 — Model Precision

Download Required Files

File	Save as
Model	resnet18_w8a8.tflite
mobilenet.json	mobilenet.json
ai_demo_sample.mp4	ai_demo_sample.mp4

If any downloaded file is a .zip archive, extract it on your host machine before copying: unzip filename.zip

Copy Files to Device

# Replace $HOME to the appropriate device path before running the commands.
# For QLI:    /root
# For Ubuntu: /home/ubuntu
# Modify this based on your platform and ensure files are copied to the correct location on the device.

ssh <user>@<device-ip> "mkdir -p $HOME/{models,labels,media,media/output}"
scp resnet18_w8a8.tflite                <user>@<device-ip>:$HOME/models/
scp mobilenet.json                      <user>@<device-ip>:$HOME/labels/
scp ai_demo_sample.mp4    <user>@<device-ip>:$HOME/media/

Connect to device

# Run from your host machine — replace <user> and <device-ip>
ssh <user>@<device-ip>

Set Environment Variables

Qualcomm Linux

export MODEL_NAME=resnet18_w8a8.tflite
export LABELS_NAME=mobilenet.json
export SRC_VIDEO_NAME=ai_demo_sample.mp4

Run the Pipeline

gst-launch-1.0 -e --gst-debug=2 \
filesrc location=$HOME/media/$SRC_VIDEO_NAME ! qtdemux ! h264parse ! \
v4l2h264dec capture-io-mode=4 output-io-mode=4 ! video/x-raw,format=NV12 ! queue ! \
tee name=t ! qtimetamux name=class_mux ! qtivoverlay ! waylandsink fullscreen=true sync=true \
t. ! queue ! qtimlvconverter ! queue ! \
qtimltflite model=$HOME/models/$MODEL_NAME delegate=external external-delegate-path=libQnnTFLiteDelegate.so external-delegate-options="QNNExternalDelegate,backend_type=htp,log_level=(string)1;" ! queue ! \
qtimlpostprocess module=mobilenet labels=$HOME/labels/$LABELS_NAME settings="{\"confidence\": 51.0}" ! text/x-raw ! queue ! class_mux.

20 ResNeXt50 Compact grouped-convolution ResNet variant.

Steps to Run on QIMSDK

🟡 Float32 - Model Precision

Download Required Files

File	Save as
Model	resnext50_float.tflite
mobilenet.json	mobilenet.json
ai_demo_sample.mp4	ai_demo_sample.mp4

If any downloaded file is a .zip archive, extract it on your host machine before copying: unzip filename.zip

Copy Files to Device

# Replace $HOME to the appropriate device path before running the commands.
# For QLI:    /root
# For Ubuntu: /home/ubuntu
# Modify this based on your platform and ensure files are copied to the correct location on the device.

ssh <user>@<device-ip> "mkdir -p $HOME/{models,labels,media,media/output}"
scp resnext50_float.tflite              <user>@<device-ip>:$HOME/models/
scp mobilenet.json                      <user>@<device-ip>:$HOME/labels/
scp ai_demo_sample.mp4    <user>@<device-ip>:$HOME/media/

Connect to device

# Run from your host machine — replace <user> and <device-ip>
ssh <user>@<device-ip>

Set Environment Variables

Qualcomm Linux

export MODEL_NAME=resnext50_float.tflite
export LABELS_NAME=mobilenet.json
export SRC_VIDEO_NAME=ai_demo_sample.mp4

Run the Pipeline

gst-launch-1.0 -e --gst-debug=2 \
filesrc location=$HOME/media/$SRC_VIDEO_NAME ! qtdemux ! h264parse ! \
v4l2h264dec capture-io-mode=4 output-io-mode=4 ! video/x-raw,format=NV12 ! queue ! \
tee name=t ! qtimetamux name=class_mux ! qtivoverlay ! waylandsink fullscreen=true sync=true \
t. ! queue ! qtimlvconverter ! queue ! \
qtimltflite model=$HOME/models/$MODEL_NAME delegate=gpu ! queue ! \
qtimlpostprocess module=mobilenet-softmax labels=$HOME/labels/$LABELS_NAME settings="{\"confidence\": 51.0}" ! text/x-raw ! queue ! class_mux.

🟢 W8A8 — Model Precision

Download Required Files

File	Save as
Model	resnext50_w8a8.tflite
mobilenet.json	mobilenet.json
ai_demo_sample.mp4	ai_demo_sample.mp4

If any downloaded file is a .zip archive, extract it on your host machine before copying: unzip filename.zip

Copy Files to Device

# Replace $HOME to the appropriate device path before running the commands.
# For QLI:    /root
# For Ubuntu: /home/ubuntu
# Modify this based on your platform and ensure files are copied to the correct location on the device.

ssh <user>@<device-ip> "mkdir -p $HOME/{models,labels,media,media/output}"
scp resnext50_w8a8.tflite               <user>@<device-ip>:$HOME/models/
scp mobilenet.json                      <user>@<device-ip>:$HOME/labels/
scp ai_demo_sample.mp4    <user>@<device-ip>:$HOME/media/

Connect to device

# Run from your host machine — replace <user> and <device-ip>
ssh <user>@<device-ip>

Set Environment Variables

Qualcomm Linux

export MODEL_NAME=resnext50_w8a8.tflite
export LABELS_NAME=mobilenet.json
export SRC_VIDEO_NAME=ai_demo_sample.mp4

Run the Pipeline

gst-launch-1.0 -e --gst-debug=2 \
filesrc location=$HOME/media/$SRC_VIDEO_NAME ! qtdemux ! h264parse ! \
v4l2h264dec capture-io-mode=4 output-io-mode=4 ! video/x-raw,format=NV12 ! queue ! \
tee name=t ! qtimetamux name=class_mux ! qtivoverlay ! waylandsink fullscreen=true sync=true \
t. ! queue ! qtimlvconverter ! queue ! \
qtimltflite model=$HOME/models/$MODEL_NAME delegate=external external-delegate-path=libQnnTFLiteDelegate.so external-delegate-options="QNNExternalDelegate,backend_type=htp,log_level=(string)1;" ! queue ! \
qtimlpostprocess module=mobilenet labels=$HOME/labels/$LABELS_NAME settings="{\"confidence\": 51.0}" ! text/x-raw ! queue ! class_mux.

21 Shufflenet-v2 Channel-split shuffle network for efficient mobile inference.

Steps to Run on QIMSDK

🟡 Float32 - Model Precision

Download Required Files

File	Save as
Model	shufflenet_v2_float.tflite
mobilenet.json	mobilenet.json
ai_demo_sample.mp4	ai_demo_sample.mp4

If any downloaded file is a .zip archive, extract it on your host machine before copying: unzip filename.zip

Copy Files to Device

# Replace $HOME to the appropriate device path before running the commands.
# For QLI:    /root
# For Ubuntu: /home/ubuntu
# Modify this based on your platform and ensure files are copied to the correct location on the device.

ssh <user>@<device-ip> "mkdir -p $HOME/{models,labels,media,media/output}"
scp shufflenet_v2_float.tflite          <user>@<device-ip>:$HOME/models/
scp mobilenet.json                      <user>@<device-ip>:$HOME/labels/
scp ai_demo_sample.mp4    <user>@<device-ip>:$HOME/media/

Connect to device

# Run from your host machine — replace <user> and <device-ip>
ssh <user>@<device-ip>

Set Environment Variables

Qualcomm Linux

export MODEL_NAME=shufflenet_v2_float.tflite
export LABELS_NAME=mobilenet.json
export SRC_VIDEO_NAME=ai_demo_sample.mp4

Run the Pipeline

gst-launch-1.0 -e --gst-debug=2 \
filesrc location=$HOME/media/$SRC_VIDEO_NAME ! qtdemux ! h264parse ! \
v4l2h264dec capture-io-mode=4 output-io-mode=4 ! video/x-raw,format=NV12 ! queue ! \
tee name=t ! qtimetamux name=class_mux ! qtivoverlay ! waylandsink fullscreen=true sync=true \
t. ! queue ! qtimlvconverter ! queue ! \
qtimltflite model=$HOME/models/$MODEL_NAME delegate=gpu ! queue ! \
qtimlpostprocess module=mobilenet-softmax labels=$HOME/labels/$LABELS_NAME settings="{\"confidence\": 51.0}" ! text/x-raw ! queue ! class_mux.

🟢 W8A8 — Model Precision

Download Required Files

File	Save as
Model	shufflenet_v2_w8a8.tflite
mobilenet.json	mobilenet.json
ai_demo_sample.mp4	ai_demo_sample.mp4

If any downloaded file is a .zip archive, extract it on your host machine before copying: unzip filename.zip

Copy Files to Device

# Replace $HOME to the appropriate device path before running the commands.
# For QLI:    /root
# For Ubuntu: /home/ubuntu
# Modify this based on your platform and ensure files are copied to the correct location on the device.

ssh <user>@<device-ip> "mkdir -p $HOME/{models,labels,media,media/output}"
scp shufflenet_v2_w8a8.tflite           <user>@<device-ip>:$HOME/models/
scp mobilenet.json                      <user>@<device-ip>:$HOME/labels/
scp ai_demo_sample.mp4    <user>@<device-ip>:$HOME/media/

Connect to device

# Run from your host machine — replace <user> and <device-ip>
ssh <user>@<device-ip>

Set Environment Variables

Qualcomm Linux

export MODEL_NAME=shufflenet_v2_w8a8.tflite
export LABELS_NAME=mobilenet.json
export SRC_VIDEO_NAME=ai_demo_sample.mp4

Run the Pipeline

gst-launch-1.0 -e --gst-debug=2 \
filesrc location=$HOME/media/$SRC_VIDEO_NAME ! qtdemux ! h264parse ! \
v4l2h264dec capture-io-mode=4 output-io-mode=4 ! video/x-raw,format=NV12 ! queue ! \
tee name=t ! qtimetamux name=class_mux ! qtivoverlay ! waylandsink fullscreen=true sync=true \
t. ! queue ! qtimlvconverter ! queue ! \
qtimltflite model=$HOME/models/$MODEL_NAME delegate=external external-delegate-path=libQnnTFLiteDelegate.so external-delegate-options="QNNExternalDelegate,backend_type=htp,log_level=(string)1;" ! queue ! \
qtimlpostprocess module=mobilenet labels=$HOME/labels/$LABELS_NAME settings="{\"confidence\": 51.0}" ! text/x-raw ! queue ! class_mux.

22 SqueezeNet-1.1 AlexNet-level accuracy at 50x fewer parameters.

Steps to Run on QIMSDK

🟡 Float32 - Model Precision

Download Required Files

File	Save as
Model	squeezenet_1_1_float.tflite
mobilenet.json	mobilenet.json
ai_demo_sample.mp4	ai_demo_sample.mp4

If any downloaded file is a .zip archive, extract it on your host machine before copying: unzip filename.zip

Copy Files to Device

# Replace $HOME to the appropriate device path before running the commands.
# For QLI:    /root
# For Ubuntu: /home/ubuntu
# Modify this based on your platform and ensure files are copied to the correct location on the device.

ssh <user>@<device-ip> "mkdir -p $HOME/{models,labels,media,media/output}"
scp squeezenet_1_1_float.tflite         <user>@<device-ip>:$HOME/models/
scp mobilenet.json                      <user>@<device-ip>:$HOME/labels/
scp ai_demo_sample.mp4    <user>@<device-ip>:$HOME/media/

Connect to device

# Run from your host machine — replace <user> and <device-ip>
ssh <user>@<device-ip>

Set Environment Variables

Qualcomm Linux

export MODEL_NAME=squeezenet_1_1_float.tflite
export LABELS_NAME=mobilenet.json
export SRC_VIDEO_NAME=ai_demo_sample.mp4

Run the Pipeline

gst-launch-1.0 -e --gst-debug=2 \
filesrc location=$HOME/media/$SRC_VIDEO_NAME ! qtdemux ! h264parse ! \
v4l2h264dec capture-io-mode=4 output-io-mode=4 ! video/x-raw,format=NV12 ! queue ! \
tee name=t ! qtimetamux name=class_mux ! qtivoverlay ! waylandsink fullscreen=true sync=true \
t. ! queue ! qtimlvconverter ! queue ! \
qtimltflite model=$HOME/models/$MODEL_NAME delegate=gpu ! queue ! \
qtimlpostprocess module=mobilenet-softmax labels=$HOME/labels/$LABELS_NAME settings="{\"confidence\": 51.0}" ! text/x-raw ! queue ! class_mux.

🟢 W8A8 — Model Precision

Download Required Files

File	Save as
Model	squeezenet_1_1_w8a8.tflite
mobilenet.json	mobilenet.json
ai_demo_sample.mp4	ai_demo_sample.mp4

If any downloaded file is a .zip archive, extract it on your host machine before copying: unzip filename.zip

Copy Files to Device

# Replace $HOME to the appropriate device path before running the commands.
# For QLI:    /root
# For Ubuntu: /home/ubuntu
# Modify this based on your platform and ensure files are copied to the correct location on the device.

ssh <user>@<device-ip> "mkdir -p $HOME/{models,labels,media,media/output}"
scp squeezenet_1_1_w8a8.tflite          <user>@<device-ip>:$HOME/models/
scp mobilenet.json                      <user>@<device-ip>:$HOME/labels/
scp ai_demo_sample.mp4    <user>@<device-ip>:$HOME/media/

Connect to device

# Run from your host machine — replace <user> and <device-ip>
ssh <user>@<device-ip>

Set Environment Variables

Qualcomm Linux

export MODEL_NAME=squeezenet_1_1_w8a8.tflite
export LABELS_NAME=mobilenet.json
export SRC_VIDEO_NAME=ai_demo_sample.mp4

Run the Pipeline

gst-launch-1.0 -e --gst-debug=2 \
filesrc location=$HOME/media/$SRC_VIDEO_NAME ! qtdemux ! h264parse ! \
v4l2h264dec capture-io-mode=4 output-io-mode=4 ! video/x-raw,format=NV12 ! queue ! \
tee name=t ! qtimetamux name=class_mux ! qtivoverlay ! waylandsink fullscreen=true sync=true \
t. ! queue ! qtimlvconverter ! queue ! \
qtimltflite model=$HOME/models/$MODEL_NAME delegate=external external-delegate-path=libQnnTFLiteDelegate.so external-delegate-options="QNNExternalDelegate,backend_type=htp,log_level=(string)1;" ! queue ! \
qtimlpostprocess module=mobilenet labels=$HOME/labels/$LABELS_NAME settings="{\"confidence\": 51.0}" ! text/x-raw ! queue ! class_mux.

23 WideResNet50 Wide residual network with increased channel width.

Steps to Run on QIMSDK

🟡 Float32 - Model Precision

Download Required Files

File	Save as
Model	wideresnet50_float.tflite
mobilenet.json	mobilenet.json
ai_demo_sample.mp4	ai_demo_sample.mp4

If any downloaded file is a .zip archive, extract it on your host machine before copying: unzip filename.zip

Copy Files to Device

# Replace $HOME to the appropriate device path before running the commands.
# For QLI:    /root
# For Ubuntu: /home/ubuntu
# Modify this based on your platform and ensure files are copied to the correct location on the device.

ssh <user>@<device-ip> "mkdir -p $HOME/{models,labels,media,media/output}"
scp wideresnet50_float.tflite           <user>@<device-ip>:$HOME/models/
scp mobilenet.json                      <user>@<device-ip>:$HOME/labels/
scp ai_demo_sample.mp4    <user>@<device-ip>:$HOME/media/

Connect to device

# Run from your host machine — replace <user> and <device-ip>
ssh <user>@<device-ip>

Set Environment Variables

Qualcomm Linux

export MODEL_NAME=wideresnet50_float.tflite
export LABELS_NAME=mobilenet.json
export SRC_VIDEO_NAME=ai_demo_sample.mp4

Run the Pipeline

gst-launch-1.0 -e --gst-debug=2 \
filesrc location=$HOME/media/$SRC_VIDEO_NAME ! qtdemux ! h264parse ! \
v4l2h264dec capture-io-mode=4 output-io-mode=4 ! video/x-raw,format=NV12 ! queue ! \
tee name=t ! qtimetamux name=class_mux ! qtivoverlay ! waylandsink fullscreen=true sync=true \
t. ! queue ! qtimlvconverter ! queue ! \
qtimltflite model=$HOME/models/$MODEL_NAME delegate=gpu ! queue ! \
qtimlpostprocess module=mobilenet-softmax labels=$HOME/labels/$LABELS_NAME settings="{\"confidence\": 51.0}" ! text/x-raw ! queue ! class_mux.

🟢 W8A8 — Model Precision

Download Required Files

File	Save as
Model	wideresnet50_w8a8.tflite
mobilenet.json	mobilenet.json
ai_demo_sample.mp4	ai_demo_sample.mp4

If any downloaded file is a .zip archive, extract it on your host machine before copying: unzip filename.zip

Copy Files to Device

# Replace $HOME to the appropriate device path before running the commands.
# For QLI:    /root
# For Ubuntu: /home/ubuntu
# Modify this based on your platform and ensure files are copied to the correct location on the device.

ssh <user>@<device-ip> "mkdir -p $HOME/{models,labels,media,media/output}"
scp wideresnet50_w8a8.tflite            <user>@<device-ip>:$HOME/models/
scp mobilenet.json                      <user>@<device-ip>:$HOME/labels/
scp ai_demo_sample.mp4    <user>@<device-ip>:$HOME/media/

Connect to device

# Run from your host machine — replace <user> and <device-ip>
ssh <user>@<device-ip>

Set Environment Variables

Qualcomm Linux

export MODEL_NAME=wideresnet50_w8a8.tflite
export LABELS_NAME=mobilenet.json
export SRC_VIDEO_NAME=ai_demo_sample.mp4

Run the Pipeline

gst-launch-1.0 -e --gst-debug=2 \
filesrc location=$HOME/media/$SRC_VIDEO_NAME ! qtdemux ! h264parse ! \
v4l2h264dec capture-io-mode=4 output-io-mode=4 ! video/x-raw,format=NV12 ! queue ! \
tee name=t ! qtimetamux name=class_mux ! qtivoverlay ! waylandsink fullscreen=true sync=true \
t. ! queue ! qtimlvconverter ! queue ! \
qtimltflite model=$HOME/models/$MODEL_NAME delegate=external external-delegate-path=libQnnTFLiteDelegate.so external-delegate-options="QNNExternalDelegate,backend_type=htp,log_level=(string)1;" ! queue ! \
qtimlpostprocess module=mobilenet labels=$HOME/labels/$LABELS_NAME settings="{\"confidence\": 51.0}" ! text/x-raw ! queue ! class_mux.

24 EfficientNet-B4 Higher-capacity EfficientNet for improved accuracy.

Steps to Run on QIMSDK

🟡 Float32 - Model Precision

Download Required Files

File	Save as
Model	efficientnet_b4_float.tflite
mobilenet.json	mobilenet.json
ai_demo_sample.mp4	ai_demo_sample.mp4

If any downloaded file is a .zip archive, extract it on your host machine before copying: unzip filename.zip

Copy Files to Device

# Replace $HOME to the appropriate device path before running the commands.
# For QLI:    /root
# For Ubuntu: /home/ubuntu
# Modify this based on your platform and ensure files are copied to the correct location on the device.

ssh <user>@<device-ip> "mkdir -p $HOME/{models,labels,media,media/output}"
scp efficientnet_b4_float.tflite        <user>@<device-ip>:$HOME/models/
scp mobilenet.json                      <user>@<device-ip>:$HOME/labels/
scp ai_demo_sample.mp4    <user>@<device-ip>:$HOME/media/

Connect to device

# Run from your host machine — replace <user> and <device-ip>
ssh <user>@<device-ip>

Set Environment Variables

Qualcomm Linux

export MODEL_NAME=efficientnet_b4_float.tflite
export LABELS_NAME=mobilenet.json
export SRC_VIDEO_NAME=ai_demo_sample.mp4

Run the Pipeline

gst-launch-1.0 -e --gst-debug=2 \
filesrc location=$HOME/media/$SRC_VIDEO_NAME ! qtdemux ! h264parse ! \
v4l2h264dec capture-io-mode=4 output-io-mode=4 ! video/x-raw,format=NV12 ! queue ! \
tee name=t ! qtimetamux name=class_mux ! qtivoverlay ! waylandsink fullscreen=true sync=true \
t. ! queue ! qtimlvconverter ! queue ! \
qtimltflite model=$HOME/models/$MODEL_NAME delegate=gpu ! queue ! \
qtimlpostprocess module=mobilenet labels=$HOME/labels/$LABELS_NAME settings="{\"confidence\": 51.0}" ! text/x-raw ! queue ! class_mux.

25 DLA-102-X Deep layer aggregation for hierarchical feature fusion.

Steps to Run on QIMSDK

🟡 Float32 - Model Precision

Download Required Files

File	Save as
Model	dla_102_x_float.tflite
mobilenet.json	mobilenet.json
ai_demo_sample.mp4	ai_demo_sample.mp4

If any downloaded file is a .zip archive, extract it on your host machine before copying: unzip filename.zip

Copy Files to Device

# Replace $HOME to the appropriate device path before running the commands.
# For QLI:    /root
# For Ubuntu: /home/ubuntu
# Modify this based on your platform and ensure files are copied to the correct location on the device.

ssh <user>@<device-ip> "mkdir -p $HOME/{models,labels,media,media/output}"
scp dla_102_x_float.tflite              <user>@<device-ip>:$HOME/models/
scp mobilenet.json                      <user>@<device-ip>:$HOME/labels/
scp ai_demo_sample.mp4    <user>@<device-ip>:$HOME/media/

Connect to device

# Run from your host machine — replace <user> and <device-ip>
ssh <user>@<device-ip>

Set Environment Variables

Qualcomm Linux

export MODEL_NAME=dla_102_x_float.tflite
export LABELS_NAME=mobilenet.json
export SRC_VIDEO_NAME=ai_demo_sample.mp4

Run the Pipeline

gst-launch-1.0 -e --gst-debug=2 \
filesrc location=$HOME/media/$SRC_VIDEO_NAME ! qtdemux ! h264parse ! \
v4l2h264dec capture-io-mode=4 output-io-mode=4 ! video/x-raw,format=NV12 ! queue ! \
tee name=t ! qtimetamux name=class_mux ! qtivoverlay ! waylandsink fullscreen=true sync=true \
t. ! queue ! qtimlvconverter ! queue ! \
qtimltflite model=$HOME/models/$MODEL_NAME delegate=gpu ! queue ! \
qtimlpostprocess module=mobilenet-softmax labels=$HOME/labels/$LABELS_NAME settings="{\"confidence\": 51.0}" ! text/x-raw ! queue ! class_mux.

🟢 W8A8 — Model Precision

Download Required Files

File	Save as
Model	dla_102_x_w8a8.tflite
mobilenet.json	mobilenet.json
ai_demo_sample.mp4	ai_demo_sample.mp4

If any downloaded file is a .zip archive, extract it on your host machine before copying: unzip filename.zip

Copy Files to Device

# Replace $HOME to the appropriate device path before running the commands.
# For QLI:    /root
# For Ubuntu: /home/ubuntu
# Modify this based on your platform and ensure files are copied to the correct location on the device.

ssh <user>@<device-ip> "mkdir -p $HOME/{models,labels,media,media/output}"
scp dla_102_x_w8a8.tflite               <user>@<device-ip>:$HOME/models/
scp mobilenet.json                      <user>@<device-ip>:$HOME/labels/
scp ai_demo_sample.mp4    <user>@<device-ip>:$HOME/media/

Connect to device

# Run from your host machine — replace <user> and <device-ip>
ssh <user>@<device-ip>

Set Environment Variables

Qualcomm Linux

export MODEL_NAME=dla_102_x_w8a8.tflite
export LABELS_NAME=mobilenet.json
export SRC_VIDEO_NAME=ai_demo_sample.mp4

Run the Pipeline

gst-launch-1.0 -e --gst-debug=2 \
filesrc location=$HOME/media/$SRC_VIDEO_NAME ! qtdemux ! h264parse ! \
v4l2h264dec capture-io-mode=4 output-io-mode=4 ! video/x-raw,format=NV12 ! queue ! \
tee name=t ! qtimetamux name=class_mux ! qtivoverlay ! waylandsink fullscreen=true sync=true \
t. ! queue ! qtimlvconverter ! queue ! \
qtimltflite model=$HOME/models/$MODEL_NAME delegate=external external-delegate-path=libQnnTFLiteDelegate.so external-delegate-options="QNNExternalDelegate,backend_type=htp,log_level=(string)1;" ! queue ! \
qtimlpostprocess module=mobilenet labels=$HOME/labels/$LABELS_NAME settings="{\"confidence\": 51.0}" ! text/x-raw ! queue ! class_mux.

26 MobileNet-v3-Small Ultra-compact mobile CNN (Small variant).

Steps to Run on QIMSDK

🟡 Float32 - Model Precision

Download Required Files

File	Save as
Model	mobilenet_v3_small_float.tflite
mobilenet.json	mobilenet.json
ai_demo_sample.mp4	ai_demo_sample.mp4

If any downloaded file is a .zip archive, extract it on your host machine before copying: unzip filename.zip

Copy Files to Device

# Replace $HOME to the appropriate device path before running the commands.
# For QLI:    /root
# For Ubuntu: /home/ubuntu
# Modify this based on your platform and ensure files are copied to the correct location on the device.

ssh <user>@<device-ip> "mkdir -p $HOME/{models,labels,media,media/output}"
scp mobilenet_v3_small_float.tflite                 <user>@<device-ip>:$HOME/models/
scp mobilenet.json                                  <user>@<device-ip>:$HOME/models/
scp ai_demo_sample.mp4                <user>@<device-ip>:$HOME/media/

Connect to device

# Run from your host machine — replace <user> and <device-ip>
ssh <user>@<device-ip>

Set Environment Variables

Qualcomm Linux

export MODEL_NAME=mobilenet_v3_small_float.tflite
export LABELS_NAME=mobilenet.json
export SRC_VIDEO_NAME=ai_demo_sample.mp4

Run the Pipeline

gst-launch-1.0 -e --gst-debug=2 \
filesrc location=$HOME/media/$SRC_VIDEO_NAME ! qtdemux ! h264parse ! \
v4l2h264dec capture-io-mode=4 output-io-mode=4 ! video/x-raw,format=NV12 ! queue ! \
tee name=t ! qtimetamux name=class_mux ! qtivoverlay ! waylandsink fullscreen=true sync=true \
t. ! queue ! qtimlvconverter ! queue ! \
qtimltflite model=$HOME/models/$MODEL_NAME delegate=gpu ! queue ! \
qtimlpostprocess module=mobilenet-softmax labels=$HOME/labels/$LABELS_NAME settings="{\"confidence\": 51.0}" ! text/x-raw ! queue ! class_mux.

27 EfficientFormer Ultra-fast transformer-style ViT for edge deployment.

Steps to Run on QIMSDK

🟡 Float32 - Model Precision

Download Required Files

File	Save as
Model	efficientformer_float.tflite
mobilenet.json	mobilenet.json
ai_demo_sample.mp4	ai_demo_sample.mp4

If any downloaded file is a .zip archive, extract it on your host machine before copying: unzip filename.zip

Copy Files to Device

# Replace $HOME to the appropriate device path before running the commands.
# For QLI:    /root
# For Ubuntu: /home/ubuntu
# Modify this based on your platform and ensure files are copied to the correct location on the device.

ssh <user>@<device-ip> "mkdir -p $HOME/{models,labels,media,media/output}"
scp efficientformer_float.tflite        <user>@<device-ip>:$HOME/models/
scp mobilenet.json                      <user>@<device-ip>:$HOME/models/
scp ai_demo_sample.mp4    <user>@<device-ip>:$HOME/media/

Connect to device

# Run from your host machine — replace <user> and <device-ip>
ssh <user>@<device-ip>

Set Environment Variables

Qualcomm Linux

export MODEL_NAME=efficientformer_float.tflite
export LABELS_NAME=mobilenet.json
export SRC_VIDEO_NAME=ai_demo_sample.mp4

Run the Pipeline

gst-launch-1.0 -e --gst-debug=2 \
filesrc location=$HOME/media/$SRC_VIDEO_NAME ! qtdemux ! h264parse ! \
v4l2h264dec capture-io-mode=4 output-io-mode=4 ! video/x-raw,format=NV12 ! queue ! \
tee name=t ! qtimetamux name=class_mux ! qtivoverlay ! waylandsink fullscreen=true sync=true \
t. ! queue ! qtimlvconverter ! queue ! \
qtimltflite model=$HOME/models/$MODEL_NAME delegate=gpu ! queue ! \
qtimlpostprocess module=mobilenet-softmax labels=$HOME/labels/$LABELS_NAME settings="{\"confidence\": 51.0}" ! text/x-raw ! queue ! class_mux.

28 GPUNet GPU-optimized network via neural architecture search.

Steps to Run on QIMSDK

🟡 Float32 - Model Precision

Download Required Files

File	Save as
Model	gpunet_float.tflite
mobilenet.json	mobilenet.json
Input video	ai_demo_sample.mp4

If any downloaded file is a .zip archive, extract it on your host machine before copying: unzip filename.zip

Copy Files to Device

# Replace $HOME to the appropriate device path before running the commands.
# For QLI:    /root
# For Ubuntu: /home/ubuntu
# Modify this based on your platform and ensure files are copied to the correct location on the device.

ssh <user>@<device-ip> "mkdir -p $HOME/{models,labels,media,media/output}"
scp gpunet_float.tflite                     <user>@<device-ip>:$HOME/models/
scp mobilenet.json                          <user>@<device-ip>:$HOME/models/
scp ai_demo_sample.mp4        <user>@<device-ip>:$HOME/media/

Connect to device

# Run from your host machine — replace <user> and <device-ip>
ssh <user>@<device-ip>

Set Environment Variables

Qualcomm Linux

export MODEL_NAME=gpunet_float.tflite
export LABELS_NAME=mobilenet.json
export SRC_VIDEO_NAME=ai_demo_sample.mp4

Run the Pipeline

gst-launch-1.0 -e --gst-debug=2 \
filesrc location=$HOME/media/$SRC_VIDEO_NAME ! qtdemux ! h264parse ! \
v4l2h264dec capture-io-mode=4 output-io-mode=4 ! video/x-raw,format=NV12 ! queue ! \
tee name=t ! qtimetamux name=class_mux ! qtivoverlay ! waylandsink fullscreen=true sync=true \
t. ! queue ! qtimlvconverter ! queue ! \
qtimltflite model=$HOME/models/$MODEL_NAME delegate=gpu ! queue ! \
qtimlpostprocess module=mobilenet labels=$HOME/labels/$LABELS_NAME settings="{\"confidence\": 51.0}" ! text/x-raw ! queue ! class_mux.

29 Sequencer2D 2D LSTM-based alternative to attention for image classification.

Steps to Run on QIMSDK

🟡 Float32 - Model Precision

Download Required Files

File	Save as
Model	sequencer2d_float.tflite
mobilenet.json	mobilenet.json
ai_demo_sample.mp4	ai_demo_sample.mp4

If any downloaded file is a .zip archive, extract it on your host machine before copying: unzip filename.zip

Copy Files to Device

# Replace $HOME to the appropriate device path before running the commands.
# For QLI:    /root
# For Ubuntu: /home/ubuntu
# Modify this based on your platform and ensure files are copied to the correct location on the device.

ssh <user>@<device-ip> "mkdir -p $HOME/{models,labels,media,media/output}"
scp sequencer2d_float.tflite                    <user>@<device-ip>:$HOME/models/
scp mobilenet.json                              <user>@<device-ip>:$HOME/models/
scp ai_demo_sample.mp4            <user>@<device-ip>:$HOME/media/

Connect to device

# Run from your host machine — replace <user> and <device-ip>
ssh <user>@<device-ip>

Set Environment Variables

Qualcomm Linux

export MODEL_NAME=sequencer2d_float.tflite
export LABELS_NAME=mobilenet.json
export SRC_VIDEO_NAME=ai_demo_sample.mp4

Run the Pipeline

gst-launch-1.0 -e --gst-debug=2 \
filesrc location=$HOME/media/$SRC_VIDEO_NAME ! qtdemux ! h264parse ! \
v4l2h264dec capture-io-mode=4 output-io-mode=4 ! video/x-raw,format=NV12 ! queue ! \
tee name=t ! qtimetamux name=class_mux ! qtivoverlay ! waylandsink fullscreen=true sync=true \
t. ! queue ! qtimlvconverter ! queue ! \
qtimltflite model=$HOME/models/$MODEL_NAME delegate=gpu ! queue ! \
qtimlpostprocess module=mobilenet labels=$HOME/labels/$LABELS_NAME settings="{\"confidence\": 51.0}" ! text/x-raw ! queue ! class_mux.

30 LeViT Hybrid convolutional-attention ViT optimized for fast inference.

Steps to Run on QIMSDK

🟡 Float32 - Model Precision

Download Required Files

File	Save as
Model	levit_float.tflite
mobilenet.json	mobilenet.json
Input video	ai_demo_sample.mp4

If any downloaded file is a .zip archive, extract it on your host machine before copying: unzip filename.zip

Copy Files to Device

# Replace $HOME to the appropriate device path before running the commands.
# For QLI:    /root
# For Ubuntu: /home/ubuntu
# Modify this based on your platform and ensure files are copied to the correct location on the device.

ssh <user>@<device-ip> "mkdir -p $HOME/{models,labels,media,media/output}"
scp levit_float.tflite                 <user>@<device-ip>:$HOME/models/
scp mobilenet.json                     <user>@<device-ip>:$HOME/labels/
scp ai_demo_sample.mp4   <user>@<device-ip>:$HOME/media/

Connect to device

# Run from your host machine — replace <user> and <device-ip>
ssh <user>@<device-ip>

Set Environment Variables

Qualcomm Linux

export MODEL_NAME=levit_float.tflite
export LABELS_NAME=mobilenet.json
export SRC_VIDEO_NAME=ai_demo_sample.mp4

Run the Pipeline

gst-launch-1.0 -e --gst-debug=2 \
filesrc location=$HOME/media/$SRC_VIDEO_NAME ! qtdemux ! h264parse ! \
v4l2h264dec capture-io-mode=4 output-io-mode=4 ! video/x-raw,format=NV12 ! queue ! \
tee name=t ! qtimetamux name=class_mux ! qtivoverlay ! waylandsink fullscreen=true sync=true \
t. ! queue ! qtimlvconverter ! queue ! \
qtimltflite model=$HOME/models/$MODEL_NAME delegate=external external-delegate-path=libQnnTFLiteDelegate.so external-delegate-options="QNNExternalDelegate,backend_type=htp,log_level=(string)1;" ! queue ! \
qtimlpostprocess module=mobilenet labels=$HOME/labels/$LABELS_NAME settings="{\"confidence\": 51.0}" ! text/x-raw ! queue ! class_mux.

31 Mobile-VIT Lightweight hybrid CNN-Transformer for mobile devices.

Steps to Run on QIMSDK

🟡 Float32 - Model Precision

Download Required Files

File	Save as
Model	mobile_vit_float.tflite
mobilenet.json	mobilenet.json
Input video	ai_demo_sample.mp4

If any downloaded file is a .zip archive, extract it on your host machine before copying: unzip filename.zip

Copy Files to Device

# Replace $HOME to the appropriate device path before running the commands.
# For QLI:    /root
# For Ubuntu: /home/ubuntu
# Modify this based on your platform and ensure files are copied to the correct location on the device.

ssh <user>@<device-ip> "mkdir -p $HOME/{models,labels,media,media/output}"
scp mobile_vit_float.tflite             <user>@<device-ip>:$HOME/models/
scp mobilenet.json                      <user>@<device-ip>:$HOME/labels/
scp ai_demo_sample.mp4    <user>@<device-ip>:$HOME/media/

Connect to device

# Run from your host machine — replace <user> and <device-ip>
ssh <user>@<device-ip>

Set Environment Variables

Qualcomm Linux

export MODEL_NAME=mobile_vit_float.tflite
export LABELS_NAME=mobilenet.json
export SRC_VIDEO_NAME=ai_demo_sample.mp4

Run the Pipeline

gst-launch-1.0 -e --gst-debug=2 \
filesrc location=$HOME/media/$SRC_VIDEO_NAME ! qtdemux ! h264parse ! \
v4l2h264dec capture-io-mode=4 output-io-mode=4 ! video/x-raw,format=NV12 ! queue ! \
tee name=t ! qtimetamux name=class_mux ! qtivoverlay ! waylandsink fullscreen=true sync=true \
t. ! queue ! qtimlvconverter ! queue ! \
qtimltflite model=$HOME/models/$MODEL_NAME delegate=gpu ! queue ! \
qtimlpostprocess module=mobilenet labels=$HOME/labels/$LABELS_NAME settings="{\"confidence\": 70.0}" ! text/x-raw ! queue ! class_mux.

32 NASNet Neural architecture search network for mobile devices.

Steps to Run on QIMSDK

🟡 Float32 - Model Precision

Download Required Files

File	Save as
Model	nasnet_float.tflite
mobilenet.json	mobilenet.json
Input video	ai_demo_sample.mp4

If any downloaded file is a .zip archive, extract it on your host machine before copying: unzip filename.zip

Copy Files to Device

# Replace $HOME to the appropriate device path before running the commands.
# For QLI:    /root
# For Ubuntu: /home/ubuntu
# Modify this based on your platform and ensure files are copied to the correct location on the device.

ssh <user>@<device-ip> "mkdir -p $HOME/{models,labels,media,media/output}"
scp nasnet_float.tflite                   <user>@<device-ip>:$HOME/models/
scp mobilenet.json                        <user>@<device-ip>:$HOME/labels/
scp ai_demo_sample.mp4      <user>@<device-ip>:$HOME/media/

Connect to device

# Run from your host machine — replace <user> and <device-ip>
ssh <user>@<device-ip>

Set Environment Variables

Qualcomm Linux

export MODEL_NAME=nasnet_float.tflite
export LABELS_NAME=mobilenet.json
export SRC_VIDEO_NAME=ai_demo_sample.mp4

Run the Pipeline

gst-launch-1.0 -e --gst-debug=2 \
filesrc location=$HOME/media/$SRC_VIDEO_NAME ! qtdemux ! h264parse ! \
v4l2h264dec capture-io-mode=4 output-io-mode=4 ! video/x-raw,format=NV12 ! queue ! \
tee name=t ! qtimetamux name=class_mux ! qtivoverlay ! waylandsink fullscreen=true sync=false \
t. ! queue ! qtimlvconverter ! queue ! \
qtimltflite model=$HOME/models/$MODEL_NAME delegate=external external-delegate-path=libQnnTFLiteDelegate.so external-delegate-options="QNNExternalDelegate,backend_type=htp,log_level=(string)1;" ! queue ! \
qtimlpostprocess module=mobilenet-softmax labels=$HOME/labels/$LABELS_NAME settings="{\"confidence\": 51.0}" ! text/x-raw ! queue ! class_mux.

Blog

Overview

Getting Started

Building AI Pipelines

Examples

Reference

Advanced Topics