This job offer is not available in your country.
Computer Vision Consultant (ReID / Action Recognition) — Part-Time
SafeSpace Globaldombivli, maharashtra, in6 days ago
Job description5+ years in CV for detection / tracking / ReID; shipped models in production. Strong with PyTorch; experience with TorchReID / FastReID and TensorRT / DeepStream . Tracker tuning (NMS / iou / appearance-matching) and multi-cam association. Performance profiling (Nsight Systems / Compute), real-time constraints. CV-CUDA, Triton Inference Server, ONNX export; pose frameworks (RTM-Pose / OpenPifPaf). Experience with privacy-preserving analytics (masking, on-device embeddings). Familiarity with Apache-friendly / open licenses for models & datasets. ReID linking tuned with zone-specific thresholds + metrics board. Action cues (throw / shove / encircle-assist) producing event confidences ≤30 ms / frame on target GPU. Written calibration & retraining SOP + benchmark report (latency, ID-switch, precision / recall). Live / Whiteboard / IDE (90-120 min) around Multi-cam ReID thresholding & linking, Tracker stability & ID-switch control, Lieghtweight “throw” action head and Deployment & profiling Take-Home (
Location : Hyderabad, India
About the engagement
We’re building higher-order safety analytics on an NVIDIA-accelerated edge box (SafeCore) with ViPE on top. We need a senior CV consultant to harden multi-camera ReID linking , stabilize tracking, and add lightweight action cues (e.g., throw, shove, encircle) that feed our CEP / graph detectors.
Time : 4–10 hrs / week (flex). Location : Remote with occasional India / US time zone overlap.
What you’ll do
- Design and tune cross-camera ReID pipelines : embedding export, normalization, gallery management, thresholding per zone / time band.
- Stabilize tracking (ByteTrack / OC-SORT / StrongSORT) and hand-off signals for CEP.
- Add light action heads (pose+trajectory) for “throw”, “shove / push”, “handoff”, “encircle geometry assist”.
- Optimize for edge : INT8 / FP16 deployment (TensorRT / Triton), pruning / quantization, batch & stream concurrency.
- Define evaluation harness : ROC / PR per zone, re-link rate, ID-switch metrics, latency budget adherence.
- Advise on data strategy : mining hard negatives, synthetic augmentation, and privacy-safe curation.
Must-have
Nice-to-have
Deliverables (first 4–6 weeks)
Selection Criteria :
Next steps : If interested please email your cover letter and resume to
Ashwin at ashwin.jaiswal@safespaceglobal.ai
Create a job alert for this search
Consultant • dombivli, maharashtra, in