Helpful OpenCV Tutorials
September 14, 2026 · SentientSight Team
Why cv2.VideoCapture(url) stops scaling past a handful of streams, how to pin decode to NVDEC/VA-API with a GStreamer pipeline, a reader-thread pattern that reconnects and never grows latency, and the per-camera frame budget to size the box before you quote it.
Read moreSeptember 11, 2026 · SentientSight Team
How to ship a working detector before the production line exists: CAD-to-render pipelines, OpenCV 5 compositing and augmentation, domain randomization ranges that actually transfer, and the real-image validation set that tells you when to stop rendering.
Read moreSeptember 10, 2026 · SentientSight Team
Thermal cores are cheap now, but a thermal frame alone cannot tell you which component is hot. How to read radiometric 16-bit data properly, calibrate a thermal camera against an RGB camera, choose a registration model your geometry justifies, and report per-component deltas an inspector can act on.
Read moreSeptember 9, 2026 · SentientSight Team
Blurring faces after the fact is where video projects stall. YuNet and an ONNX plate detector in the OpenCV 5 DNN engine, tracking-based temporal smoothing so nothing flickers back into view, irreversible redaction instead of invertible blur, and the audit manifest a DPO will accept.
Read moreSeptember 9, 2026 · SentientSight Team
Zero defect labels required: train on good parts only, register frames with OpenCV 5, run a PatchCore/anomalib model through the DNN engine, and choose reject thresholds from escape and false-reject costs instead of best-F1.
Read more