Code for CVPR2021 paper

Zero-shot Instance Segmentation

Code requirements

• python: python3.7
• nvidia GPU
• pytorch1.1.0
• GCC >=5.4
• NCCL 2
• the other python libs in requirement.txt

Install

conda create -n zsi python=3.7 -y
conda activate zsi

conda install pytorch=1.1.0 torchvision=0.3.0 cudatoolkit=10.0 -c pytorch

pip install cython && pip --no-cache-dir install -r requirements.txt
   
python setup.py develop

Dataset prepare

• Download the train and test annotations files for zsi from annotations, put all json label file to

  data/coco/annotations/
  

• Download MSCOCO-2014 dataset and unzip the images it to path：

  data/coco/train2014/
  data/coco/val2014/
  

• Training:

  • 48/17 split:

       chmod +x tools/dist_train.sh
       ./tools/dist_train.sh configs/zsi/train/zero-shot-mask-rcnn-BARPN-bbox_mask_sync_bg_decoder.py 4
    

  • 65/15 split:

    chmod +x tools/dist_train.sh
    ./tools/dist_train.sh configs/zsi/train/zero-shot-mask-rcnn-BARPN-bbox_mask_sync_bg_65_15_decoder_notanh.py 4
    

• Inference & Evaluate:

  • ZSI task:

    • 48/17 split ZSI task:

      • download 48/17 ZSI model, put it in checkpoints/ZSI_48_17.pth

      • inference:

        chmod +x tools/dist_test.sh
        ./tools/dist_test.sh configs/zsi/48_17/test/zsi/zero-shot-mask-rcnn-BARPN-bbox_mask_sync_bg_decoder.py checkpoints/ZSI_48_17.pth 4 --json_out results/zsi_48_17.json
        

      • our results zsi_48_17.bbox.json and zsi_48_17.segm.json can also downloaded from zsi_48_17_reults.

      • evaluate:

        • for zsd performance

          python tools/zsi_coco_eval.py results/zsi_48_17.bbox.json --ann data/coco/annotations/instances_val2014_unseen_48_17.json
          

        • for zsi performance

          python tools/zsi_coco_eval.py results/zsi_48_17.segm.json --ann data/coco/annotations/instances_val2014_unseen_48_17.json --types segm
          

    • 65/15 split ZSI task:

      • download 65/15 ZSI model, put it in checkpoints/ZSI_65_15.pth

      • inference:

        chmod +x tools/dist_test.sh
        ./toools/dist_test.sh configs/zsi/65_15/test/zsi/zero-shot-mask-rcnn-BARPN-bbox_mask_sync_bg_65_15_decoder_notanh.py checkpoints/ZSI_65_15.pth 4 --json_out results/zsi_65_15.json
        

      • our results zsi_65_15.bbox.json and zsi_65_15.segm.json can also downloaded from zsi_65_15_reults.

      • evaluate:

        • for zsd performance

          python tools/zsi_coco_eval.py results/zsi_65_15.bbox.json --ann data/coco/annotations/instances_val2014_unseen_65_15.json
          

        • for zsi performance

          python tools/zsi_coco_eval.py results/zsi_65_15.segm.json --ann data/coco/annotations/instances_val2014_unseen_65_15.json --types segm
          

  • GZSI task:

    • 48/17 split GZSI task:
      • use the same model file ZSI_48_17.pth in ZSI task
      • inference:

        chmod +x tools/dist_test.sh
        ./tools/dist_test.sh configs/zsi/48_17/test/gzsi/zero-shot-mask-rcnn-BARPN-bbox_mask_sync_bg_decoder_gzsi.py checkpoints/ZSI_48_17.pth 4 --json_out results/gzsi_48_17.json
        

      • our results gzsi_48_17.bbox.json and gzsi_48_17.segm.json can also downloaded from gzsi_48_17_results.
      • evaluate:
        • for gzsd

          python tools/gzsi_coco_eval.py results/gzsi_48_17.bbox.json --ann data/coco/annotations/instances_val2014_gzsi_48_17.json --gzsi --num-seen-classes 48
          

        • for gzsi

          python tools/gzsi_coco_eval.py results/gzsi_48_17.segm.json --ann data/coco/annotations/instances_val2014_gzsi_48_17.json --gzsi --num-seen-classes 48 --types segm
          

    • 65/15 split GZSI task:
      • use the same model file ZSI_48_17.pth in ZSI task
      • inference:

        chmod +x tools/dist_test.sh
        ./tools/dist_test.sh configs/zsi/65_15/test/gzsi/zero-shot-mask-rcnn-BARPN-bbox_mask_sync_bg_65_15_decoder_notanh_gzsi.py checkpoints/ZSI_65_15.pth 4 --json_out results/gzsi_65_15.json
        

      • our results gzsi_65_15.bbox.json and gzsi_65_15.segm.json can also downloaded from gzsi_65_15_results.
      • evaluate:
        • for gzsd

          python tools/gzsi_coco_eval.py results/gzsi_65_15.bbox.json --ann data/coco/annotations/instances_val2014_gzsi_65_15.json --gzsd --num-seen-classes 65
          

        • for gzsi

          python tools/gzsi_coco_eval.py results/gzsi_65_15.segm.json --ann data/coco/annotations/instances_val2014_gzsi_65_15.json --gzsd --num-seen-classes 65 --types segm
          

License

ZSI is released under MIT License.

Citing

If you use ZSI in your research or wish to refer to the baseline results published here, please use the following BibTeX entries:

@InProceedings{zhengye2021zsi,
  author  =  {Ye, Zheng and Jiahong, Wu and Yongqiag, Qin and Faen, Zhang and Li, Cui},
  title   =  {Zero-shot Instance Segmentation},
  booktitle = {The IEEE Conference on Computer Vision and Pattern Recognition (CVPR)},
  month = {June},
  year = {2021}
}