Senior Computer Vision & Video Inference Engineer (DeepStream / TensorRT / GStreamer)
Budget: $30.0 - $60.0
HOURLY / AS_NEEDED
⭐ 0.00 (0)
United Kingdom
computer-vision, python, opencv, c++, artificial-intelligence, machine-learning, software-development, gstreamer, cuda
Qualifiche preferite
- Esperienza: Esperto
- Inglese: Conversazionale
We are seeking a Senior Computer Vision & Video Inference Engineer to audit, optimize, and stabilize a real-time, multi-stream video analytics pipeline.
The core focus of this engagement is strictly the computer vision and edge/GPU inference layer—optimizing detection, multi-object tracking, hardware-accelerated video decoding, and GPU throughput across concurrent RTSP feeds. You will work alongside our core engineering team to harden the pipeline for long-running reliability and integrate metadata outputs into our existing backend services.
Key Responsibilities
- Audit and tune Ultralytics YOLO models (v8+) using TensorRT FP16/INT8 quantization for production deployment.
- Refine multi-object tracking (ByteTrack) to minimize ID switches, trajectory fragmentation, and false boundary crossings using spatial heuristics (confirmation frames, hysteresis, dead zones).
- Harden RTSP stream ingestion, decoding, and watchdog recovery using GStreamer and OpenCV to eliminate backpressure and dropped frames.
- Profile and scale inference throughput using NVIDIA DeepStream (nvstreammux, nvinfer, nvtracker) on discrete GPUs and/or Jetson targets.
- Identify and eliminate memory leaks, thread contention, and CPU-GPU transfer bottlenecks using Linux/CUDA profiling tools (e.g., Nsight Systems).
- Expose telemetry metrics (FPS, stream health, latency) and bridge metadata via ZeroMQ, REST, or lightweight IPC.
Required Qualifications
- Proven track record deploying production video pipelines with NVIDIA DeepStream SDK, TensorRT, and CUDA.
- Deep hands-on experience with GStreamer pipeline construction, RTSP protocols, and hardware decoding.
- Strong practical background with YOLO architectures, ByteTrack (or similar MOT algorithms), and boundary/line-crossing analytics.
- Advanced concurrent Python (multiprocessing, async, queues) with strong Linux systems debugging.
- Experience writing or configuring custom C++ parsers/plugins for DeepStream is a strong plus.
Apri su Upwork
AI proposal draft
Generate a short cover letter for this job. Edit before sending.
Sign in to generate an AI proposal draft.
Accedi