Data-Uncertainty Guided Multi-Phase Learning for Semi-supervised Object Detection

This is the mmdetection implementation of our CVPR2021 paper:

Zhenyu Wang, Yali Li, Ye Guo, Lu Fang, Shengjin Wang. Data-Uncertainty Guided Multi-Phase Learning for Semi-supervised Object Detection. ArXiv.

Installation

This code is based on mmdetection v2.18. Please install the code according to the mmdetection step first. Run:

pip install ensemble_boxes

to prepare for ensembling the results.

data preparation

multiphase
├──data
|  ├──VOCdevkit
|  |  ├──VOC2007
|  |  ├──VOC2012
|  ├──coco
|  |  ├──annotations
|  |  |  ├──instances_train2014.json
|  |  |  ├──instances_valminusminival2014.json
|  |  |  ├──instances_minival2014.json
|  |  ├──images
|  |  |  ├──train2014
|  |  |  ├──val2014

Running scripts

pascal voc

Run:

python tools/dataset_converters/pascal_voc.py data/VOCdevkit -o labels

to prepare the dataset. Then, to train the supervised model, run (the default gpu number for VOC is 4):

bash tools/dist_train.sh configs/multiphase/pascal_voc/faster_rcnn_r50_fpn_1x_voc07_sup.py 4

With the supervised model, generating pseudo labels for the first phase:

bash scripts/pascal_voc/extract_pl_phase1.sh 4 labels/rvoc.pkl labels/voc12_trainval_pl_phase1.pkl 

Then, perform semi-supervised learning for the first phase:

bash tools/dist_train.sh configs/multiphase/pascal_voc/faster_rcnn_r50_fpn_1x_voc0712_semi_phase1.py 4

Generating pseudo labels for the second phase:

bash scripts/pascal_voc/extract_pl_phase2.sh 4 labels/rvoc.pkl labels/rvoc2.pkl labels/voc12_trainval_pl_phase2.pkl

Semi-supervised learning for the second phase:

bash tools/dist_train.sh configs/multiphase/pascal_voc/faster_rcnn_r50_fpn_1x_voc0712_semi_phase2.py 4

Finally, model ensemble for the detection results:

bash scripts/pascal_voc/ensemble_test.sh 4

coco

For the COCO dataset, the basic pipieline is the same, the default gpu number is 8:

bash tools/dist_train.sh configs/multiphase/coco/faster_rcnn_r50_fpn_1x_coco_sup.py 8
bash scripts/coco/extract_pl_phase1.sh 8 labels/rvcoco.pkl labels/coco115k_trainval_pl_phase1.json 
bash tools/dist_train.sh configs/multiphase/coco/faster_rcnn_r50_fpn_1x_coco_semi_phase1.py 8
bash scripts/coco/extract_pl_phase2.sh 8 labels/rvcoco.pkl labels/rvcoco2.pkl labels/coco115k_trainval_pl_phase2.json
bash tools/dist_train.sh configs/multiphase/coco/faster_rcnn_r50_fpn_1x_coco_semi_phase2.py 8
bash scripts/coco/ensemble_test.sh 8

Future features

• Experiments on COCO partial (1%, 2%, 5%, 10% ratio for labeled images)

Citation

If you find this repo useful for your research, please consider citing the paper as follows:

@inproceedings{wang2021data,
  title={Data-uncertainty guided multi-phase learning for semi-supervised object detection},
  author={Wang, Zhenyu and Li, Yali and Guo, Ye and Fang, Lu and Wang, Shengjin},
  booktitle={Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition},
  year={2021}
}

Contact us for any questions.