Skip to content

Latest commit

 

History

History
71 lines (57 loc) · 3.58 KB

README.md

File metadata and controls

71 lines (57 loc) · 3.58 KB
Raw

BoxInst: High-Performance Instance Segmentation with Box Annotations

BoxInst: High-Performance Instance Segmentation with Box Annotations;
Zhi Tian, Chunhua Shen, Xinlong Wang and Hao Chen;
In: Proc. IEEE Conf. Computer Vision and Pattern Recognition (CVPR), 2021.
arXiv preprint arXiv:2012.02310

[Paper] [BibTeX] [Video Demo]

Installation & Quick Start

First, follow the default instruction to install the project and datasets/README.md set up the datasets (e.g., MS-COCO).

For demo, run the following command lines:

wget https://huggingface.co/tianzhi/AdelaiDet-BoxInst/resolve/main/BoxInst_MS_R_50_3x.pth?download=true -O BoxInst_MS_R_50_3x.pth
python demo/demo.py \
    --config-file configs/BoxInst/MS_R_50_3x.yaml \
    --input input1.jpg input2.jpg \
    --opts MODEL.WEIGHTS BoxInst_MS_R_50_3x.pth

For training on COCO, run:

OMP_NUM_THREADS=1 python tools/train_net.py \
    --config-file configs/BoxInst/MS_R_50_1x.yaml \
    --num-gpus 8 \
    OUTPUT_DIR training_dir/BoxInst_MS_R_50_1x

For evaluation on COCO, run:

OMP_NUM_THREADS=1 python tools/train_net.py \
    --config-file configs/BoxInst/MS_R_50_1x.yaml \
    --eval-only \
    --num-gpus 8 \
    OUTPUT_DIR training_dir/BoxInst_MS_R_50_1x \
    MODEL.WEIGHTS training_dir/BoxInst_MS_R_50_1x/model_final.pth

Models

COCO Instance Segmentation Baselines with BoxInst

Only box annotations are used during training.

Name inf. time box AP mask AP mask AP (test-dev) download
BoxInst_MS_R_50_1x 14 FPS 39.4 30.7 - model
BoxInst_MS_R_50_3x 14 FPS 41.5 31.8 32.1 model
BoxInst_MS_R_101_1x 11 FPS 41.4 32.2 32.5 model
BoxInst_MS_R_101_3x 11 FPS 43.3 33.0 33.2 model
BoxInst_MS_R_101_BiFPN_3x 10 FPS 45.4 34.1 33.9 model
BoxInst_MS_R_101_BiFPN_dcni3_3x 8 FPS 46.4 34.8 35.0 model

Disclaimer:

  • All models are trained with multi-scale data augmentation. Inference time is measured on a single NVIDIA 1080Ti with batch size 1.
  • This is a reimplementation. Thus, the numbers might be slightly different from the ones reported in our original paper.

Citing BoxInst

If you use BoxInst in your research or wish to refer to the baseline results, please use the following BibTeX entries.

@inproceedings{tian2020boxinst,
  title     =  {{BoxInst}: High-Performance Instance Segmentation with Box Annotations},
  author    =  {Tian, Zhi and Shen, Chunhua and Wang, Xinlong and Chen, Hao},
  booktitle =  {Proc. IEEE Conf. Computer Vision and Pattern Recognition (CVPR)},
  year      =  {2021}
}