DifficultyAwareEmbedding

A difficulty-aware embedding of complementary deep networks for image classification It is the code released for Embedding Complementary Deep Networks for Image Classification(CVPR2019).

<p align='center'> <img src='./imgs/imagenet.png' width='400'> </p>

Citation

If you find the paper or repository is useful, please kindly cite:

@InProceedings{Chen_2019_CVPR,
author = {Chen, Qiuyu and Zhang, Wei and Yu, Jun and Fan, Jianping},
title = {Embedding Complementary Deep Networks for Image Classification},
booktitle = {The IEEE Conference on Computer Vision and Pattern Recognition (CVPR)},
month = {June},
year = {2019}
}

Setup/Prerequisite

• install the prerequisite python packages

pip install -r requirements.txt

• Run the Visdom server for monitoring the training/test process

visdom

<p align='center'> <img src='./imgs/ScreenShot-Training.png' width='400'> </p> <p align='center'> <img src='./imgs/ScreenShot-Test.png' width='400'> </p>

Training

1. Download the resnet50 pretrained model from official torchvision

wget -P checkpoints/test-resnet50 https://download.pytorch.org/models/resnet50-19c8e357.pth

2. Run the embedding algrithms for the following rounds

# for the second round
./scripts/trainval_resnet50.sh

Test

1. Download the resnet50 pretrained model and put it under checkpoints/resnet folder

2. Run the test script

./scripts/test_resnet50.sh