Skip to main content
The region-based detectors in this group evolved in three steps. R-CNN classifies region proposals that an external algorithm produces, cropping and processing each one separately. Fast R-CNN runs the backbone once over the whole image and pools each proposal’s features from the shared feature map. Faster R-CNN replaces the external proposals with a Region Proposal Network inside the detector. The sections that follow cover each step and then build Faster R-CNN in PyTorch.

​
Video references

These videos, the first seven of the ExplainingAI series Object Detection Tutorials: RCNN to YOLOv8, follow the same path and pair well with the sections.
  1. Object Detection Series: Intro, a short trailer for the series.
  2. R-CNN Explained (34 min).
  3. Mean Average Precision (mAP): Explanation and Implementation for Object Detection (28 min), the metric the detectors are compared with. See also Detection Metrics.
  4. Fast R-CNN Explained: ROI Pooling (33 min).
  5. Faster R-CNN Explanation: Region Proposal Network (41 min).
  6. Faster R-CNN PyTorch Implementation (68 min).
  7. Faster R-CNN PyTorch Code Walkthrough: Fine-Tuning and Custom Dataset Training (63 min).
The full playlist continues with YOLO, SSD and DETR.
Edit this page on GitHub or file an issue.