Input image + detections
Binary mask and blobs
correct detection wrong class false positive missed ground truth

Computer Vision Simulation | Object Recognition

A hands-on object detector built from classical image processing: noise filtering, automatic Otsu thresholding, connected-component labelling and shape features, scored against exact ground truth with IoU, precision, recall and F1. All scenes are synthetic, so you can raise the noise or lower the contrast and watch precisely where and why the detector starts to fail.