Skip to content
Codes and pretrained model for ECCV 14 paper 'Part based RCNNs for fine-grained category detection'
Branch: master
Clone or download
Fetching latest commit…
Cannot retrieve the latest commit at this time.

Part-based RCNNs for Fine-grained Category Detection

This work is created by Ning Zhang, Jeff Donahue, Ross Girshick and Trevor Darrell from UC Berkeley.

Citing this work

If you are using this code for your research, please cite the following paper:

    Author = {Zhang, Ning and Donahue, Jeff and Girshick, Ross and Darrell, Trevor},
    Title = {Part-based RCNN for Fine-grained Detection},
    Booktitle = {European Conference on Computer Vision},
    Year = {2014}


This software is under BSD 3-Clause License, please refer to LICENSE file.


  1. Caffe
  1. RCNN
  • Download source code from and follow the instructions to install.
  • Change rcnn path in init.m
  • In order to train part detectors for CUB2011 dataset, replace the following three functions in imdb/imdb_from_voc.m imdb/roidb_from_voc.m and imdb/imdb_eval_voc.m to the functions in imdb_cub folder.
  • Follow rcnn instructions to train the part detectors.
  1. Liblinear

Annotation/ has annotated part boxes on CUB200-2011 dataset.


  • run.m is the main function to reproduce the results in the paper.
  • Part detectors, finetuned models, feature representations are cached. Download the cache files by running and unzip to caches/ folder.

###Bug report If you have any issues running the codes, please contact Ning Zhang (

You can’t perform that action at this time.