Towards Human-Machine Cooperation: Self-supervised Sample Mining for Object Detection

Keze Wang, Xiaopeng Yan, Dongyu Zhang, Lei Zhang, Liang Lin

Sun Yat-Sen University, Presented at CVPR2018

For Academic Research Use Only!

If you find SSM useful in your research, please consider citing:

@InProceedings{Wang_2018_CVPR,
          author = {Wang, Keze and Yan, Xiaopeng and Zhang, Dongyu and Zhang, Lei and Lin, Liang},
          title = {Towards Human-Machine Cooperation: Self-Supervised Sample Mining for Object Detection},
          booktitle = {The IEEE Conference on Computer Vision and Pattern Recognition (CVPR)},
          month = {June},
          year = {2018}
        }

The code is built on top of R-FCN. Please carefully read through py-R-FCN and make sure py-R-FCN can run within your enviornment.

1. In our paper, we used Pascal VOC2007/VOC2012 and COCO as our datasets, and ResNet-101 model as our pre-trained model.

2. Please download ImageNet-pre-trained ResNet-101 model manually, and put them into $SSM_ROOT/data/imagenet_models

1. training

   Before training, please prepare your dataset and pre-trained model and store them in the right path as R-FCN. You can go to ./tools/ and modify train_net.py to reset some parameters.Then, simply run sh ./train.sh.

2. testing

   Before testing, you can modify test.sh to choose the trained model path, then simply run sh ./test.sh to get the evaluation result.

Tested on Ubuntu 14.04 with a Titan X GPU (12G) and Intel(R) Xeon(R) CPU E5-2623 v3 @ 3.00GHz.

Thanks for the contribution of Xiaoxi Wang.