高级计算机视觉工程师（现场，纽约市，纽约）

Sourced from hirelala.com

Pakket builds remote control systems for warehouses, tracking forklifts, pallet jacks, people, and trucks via IP cameras and projecting them onto a shared bird's-eye-view floor plan. As a Senior Computer Vision Engineer, you will own the perception core of the real-time multi-camera tracking pipeline deployed on GPU edge servers. This is a hands-on founding-engineer role with meaningful equity and broad ownership across the stack.

• Build and improve the real-time pipeline: multi-camera RTSP capture, GPU decoding, detection, fusion, and tracking
• Improve cross-camera tracking quality: association, track lifecycle, ID switches, occlusion, and track recovery
• Own end-to-end detection model workflow from dataset preparation to TensorRT/ONNX deployment
• Optimize compute budgets: batching, quantization, latency/throughput on edge GPUs
• Build evaluation tooling with offline replay and MOTA/IDF1/HOTA-level metrics and controlled A/B ablations
• Own adjacent systems: backend/API, deployment, observability, and internal tooling
• Diagnose production failures from logs, replays, and captured data

• 4+ years of hands-on experience building computer vision systems
• Real-time multi-camera video: RTSP capture, FFmpeg/PyAV decoding, GPU (NVDEC) decoding, frame pacing, backpressure
• End-to-end object detection with YOLO/Ultralytics: dataset prep, fine-tuning, deployment; inference optimization (batching, ONNX/TensorRT, CUDA)
• Camera geometry and calibration: intrinsics + lens distortion, homography estimation/application, projection to a shared ground plane; strong NumPy/OpenCV/SciPy
• Linux + remote GPU operations: SSH to edge machines and cloud VMs, running systemd/Docker services
• Experience building production systems on real video and discussing common failure modes

• Warehouse/logistics or industrial computer vision experience
• Multi-camera object tracking
• Small-model classification (MobileNet-like) and sequence smoothing (e.g., HMM)
• Video encoding/distribution: NVENC, RTMP/HLS, MediaRTX
• Familiarity with startup environments