This code base is no longer maintained and exists as a historical artifact to supplement our CVPR 2016 paper. For more recent work, please see craftGBD.
by Wanli Ouyang, Xiaogang Wang, Cong Zhang, Xiaokang Yang
We cluster objects into visually similar class groups and learn deep representations for these groups separately. A hierarchical feature learning scheme is proposed. In this scheme, the knowledge from the group with large number of classes is transferred for learning features in its sub-groups. For more details, please refer to our arXiv paper.
If you find the code or the models useful, please cite this paper:
@inproceedings{wanli2016factors,
author = {Wanli Ouyang, Xiaogang Wang, Cong Zhang, Xiaokang Yang},
title = {Factors in Finetuning Deep Model for object detection},
booktitle = {CVPR},
year = {2016},
}
The caffe with multi-GPU support is recommended for fine-tuning the model. You can build and run the code by the following commands.
mkdir build && cd build
cmake .. -DUSE_MPI=ON
make && make install
mpirun -np 4 ./install/bin/caffe train --solver=<Your Solver File> [--weights=<Pretrained caffemodel>]
We are using the CRAFT Objects from Images for generating proposals.