Issue link: https://resources.mouser.com/i/1442826
10 AI 1. YOLO: You Only Look Once: As opposed to traditional algorithms, YOLO applies a single CNN classifier to the entire image, which divides images into multiple regions and predicts bounding boxes and probabilities. It then uses these probabilities across overlapping regions to come up with a set of predictions with different levels of confidence, and discards based on thresholds to reach a final set. This makes YOLO much faster, but not very good at detecting smaller objects or highly overlapping ones. 2. Mobile-net SSD—Single Shot Detection: SSD operates the same way as YOLO in applying a single CNN classifier to the entire image, but uses techniques like anchor boxes and different aspect ratios to further classify each region. The prediction is done after each region has been through multiple CNN layers. This gives SSD a higher degree of accuracy, with a minor loss in performance as compared to YOLO. Our system models detection as a regression problem. It divides the image into an S x S grid and for each grid cell predicts B bounding boxes, confidence for those boxes, and C class probabilities. These predictions are encoded as an S x S x (B * 5 + C) tensor. YOLO vs SSD vs Faster-RCNN for various sizes.