Object Detection in Computer Vision - Lecture 22

Explore object detection techniques in this comprehensive computer vision lecture. Delve into the comparison of detection algorithms, examining R-CNN training methods including fine-tuning, feature extraction, and classifier training. Investigate Fast R-CNN's improvements, focusing on Region of Interest Pooling. Analyze the limitations of Fast R-CNN and discover the advancements made in Faster R-CNN, including the introduction of Region Proposal Networks (RPN) and anchor boxes. Gain valuable insights into the evolution of object detection methodologies within the field of computer vision.

Intro
Which algorithm is better?
Detection as Classification
R-CNN Training (Fine-tuning)
R-CNN Training (feature extraction)
R-CNN Training (train classifier)
UCF R-CNN Training (bounding box regression/prediction)
Issue #1 with R-CNN
Fast R-CNN: Another view
Fast R-CNN: Region of Interest Pooling
ROI-Pooling
Problem of Fast R-CNN?
R-CNN Summary
Region proposal network (RPN)
Anchor boxes
Faster R-CNN