2014-eccv-part.md

Part-based R-CNNs for Fine-grained Category Detection

2014 ECCV, Ning Zhang, Jeff Donahue, Ross Girshick, Trevor Darrell.

Flowchart

Algorithm

Semantic part localization can facilitate fine-grained categorization

1. part-based RCNN:

   • use RCNN to treat objects and parts as independent object categories
   • train a one-versus-all linear SVM on feature descriptors extracted over region proposals

2. Geometric constraints:

   • consider the locations of object of parts to choose best ones

3. Fine-grained categrization:

   • Use fine-tuned CNN model to extract feature for object and parts independently
   • one-versus-all linear SVM to classify