Reading Paper

Introduction

The RCNN mainly includes three modules for training:

• Generating around 2000 category-independent region proposals.
• extracts a fixed-length features from each region with CNN.
• A set of class-specific SVMs.

The RCNN mainly includes two steps during testing:

• Generating around 2000 category-independent region proposals.
• Classifying each region with category-specific linear SVMs.

Module design:

• Region proposal: selective search
• Feature extraction: 4096-d
• A set of class-specific SVMs.

Methods