Skip to main content

​
Welcome

The QIM SDK App Builder guides walk you through constructing your own GStreamer-based application on Qualcomm platforms — without needing deep GStreamer expertise. IoT engineering teams span a wide spectrum of needs, so the App Builder SDKs meet developers where they are:
  • Python — high-level interfaces for rapid prototyping and orchestration, ideal for partners and hobbyists.
  • C++ — high-performance environments for commercial product builders.

​
Why These SDKs

  • Allow developers to build Python-based applications with hardware acceleration across Camera, AI/NPU, Video (Encode/Decode), Audio, GPU, and Display IPs.
  • Easy APIs with an extremely low learning curve.
  • Easy AI model pipeline enablement, with custom pre- and post-processing.
  • Hosting compiler-independent applications at AI Hub.
  • Eliminate the learning curve involved with native SDKs.
  • Quick prototyping to help ODMs/customers quantify SoC capabilities, including:
    • Single-model performance.
    • Single / daisy-chain / multi-stream pipeline performance — ISP camera (single/multi), video decode/encode capabilities, latencies, GPU%, CPU%, thermal readouts.
  • Amazon Fleet 3.0: 5 cameras, 13 AI streams, 2 network streams, 10 encode, overlays, WebRTC/RTSP.
  • Geotab: 2 cameras, two four-stage daisy-chain AI pipelines with 4 AI models running cascaded.
  • Help ODMs/partners in the SoC-selection process, reducing dependency on CE and Dev Engineering teams — BD/Marketing teams can share the public documentation and customers evaluate on their own.
  • Help non-Edge-AI/app developers such as camera sensor tuners, AI model trainers, and data collectors.
  • Meets ODM requests for Nvidia-equivalent SDKs to keep migration to Qualcomm easy, offering the same Pythonic SDK style they expect.
Introduction

​
What These SDKs Cover

The App Builder SDKs provide comprehensive building blocks to build a complete AI application, covering all possible facets:
  • Data capture — ISP/USB camera, network camera (WebRTC/HLS/RTSP), file source, audio (mic/USB/RTSP).
  • Inference — tensor in/out via TFLite/LiteRT, ONNXRT, QNN, SNPE.
  • Hardware-accelerated ML pre-processing — GPU (GLES), CPU (Neo Kernels), EVA Kernels — covering a wide range of AI Hub models.
  • ML post-processing — covering a wide range of AI Hub models.
  • An easy mechanism for developers to write their own pre- and post-processing for AI model onboarding.
  • Hardware-accelerated overlays — bounding boxes, masks, labels, privacy masks, arbitrary shapes/sizes — with full developer access to build their own.
  • Zero-copy throughout the pipeline — multi-threaded, parallel workloads with optimal hardware acceleration utilization and no overhead.
  • Video decode/encode — all formats supported by the SoC, including JPEG encode/decode and MJPEG.
  • Wide range of muxers/parsers and file containers (200+) covering audio, video, and metadata streams.
  • Wide range of software codecs (100+).
  • Wide range of streaming stacks — WebRTC, RTSP, HLS, RTMP.
  • Multi-stream AI / daisy-chain — nth-stage AI metadata cascading, JSON formatting, AI metadata streaming.
  • IoT message broker protocols — MQTT, Kafka, and Redis.
  • Concurrent AI — vision AI, audio AI, and sensor AI (time-series), with time synchronization.
  • Support for running these SDKs in a Docker container, giving the same experience at both the platform and container level.
  • Exception-based unified error handling.
  • A rich set of sample applications and documentation (API documentation and blog-style content).

​
Two Language Bindings

These guides cover the essential application-level concepts required to initialize GStreamer, construct and configure a pipeline programmatically, manage pipeline execution state, and handle runtime events — using QIM SDK plugins alongside standard GStreamer elements. Once you’re comfortable with the core concepts, explore the Sample Applications section for ready-to-run C++ and Python applications demonstrating AI and multimedia use cases built on these same concepts.