Papers (not important)

Region Proposals

• Find a small set of boxes that are likely to cover all objects
• Often based on heuristics: e.g. look for “blob-like” image regions
• Relatively fast to run; e.g. Selective Search gives 2000 region proposals in a few seconds on CPU

Paper

Rich feature hierarchies for accurate object detection and semantic segmentation

Rich feature hierarchies for accurate object detection and...

R-CNN (Region-based CNN)

1. Use an off-the-shelf Region of Interest (RoI) proposal algorithm (~2k proposals per image)
2. Crop and warp (resize) each proposal image window to obtain a fixed-size network input
3. Forward propagate the fixed-size network input to get a feature representation
4. Object classification
5. Refine proposal localization with bounding-box regression (cropped image may be not optimal)

Bounding-box regression

After scoring each selective search proposal, we predict a new bounding box for the detection using a class-specific bounding-box regressor.