Skip to content

zju3dv/snake

Repository files navigation

Good news! Snake algorithms exhibit state-of-the-art performances on COCO dataset: DANCE

Deep Snake for Real-Time Instance Segmentation

city

Deep Snake for Real-Time Instance Segmentation
Sida Peng, Wen Jiang, Huaijin Pi, Xiuli Li, Hujun Bao, Xiaowei Zhou
CVPR 2020 oral

Any questions or discussions are welcomed!

Installation

Please see INSTALL.md.

Testing

Testing on Cityscapes

  1. Download the pretrained model here and put it to $ROOT/data/model/rcnn_snake/long_rcnn/197.pth.
  2. Test:
    # use coco evaluator
    python run.py --type evaluate --cfg_file configs/city_rcnn_snake.yaml
    # use the cityscapes official evaluator
    python run.py --type evaluate --cfg_file configs/city_rcnn_snake.yaml test.dataset CityscapesVal
    
  3. Speed:
    python run.py --type network --cfg_file configs/city_rcnn_snake.yaml
    

Testing on Kitti

  1. Download the pretrained model here and put it to $ROOT/data/model/snake/kins/149.pth.
  2. Test:
    python run.py --type evaluate --cfg_file configs/kins_snake.yaml test.dataset KinsVal
    
  3. Speed:
    python run.py --type network --cfg_file configs/kins_snake.yaml test.dataset KinsVal
    

Testing on Sbd

  1. Download the pretrained model here and put it to $ROOT/data/model/snake/sbd/149.pth.
  2. Test:
    python run.py --type evaluate --cfg_file configs/sbd_snake.yaml test.dataset SbdVal
    
  3. Speed:
    python run.py --type network --cfg_file configs/sbd_snake.yaml test.dataset SbdVal
    

Visualization

Visualization on Cityscapes

  1. Download the pretrained model here and put it to $ROOT/data/model/rcnn_snake/long_rcnn/197.pth.
  2. Visualize:
    # Visualize Cityscapes test set
    python run.py --type visualize --cfg_file configs/city_rcnn_snake.yaml test.dataset CityscapesTest ct_score 0.3
    # Visualize Cityscapes val set
    python run.py --type visualize --cfg_file configs/city_rcnn_snake.yaml test.dataset CityscapesVal ct_score 0.3
    

If setup correctly, the output will look like

vis_city

Visualization on Kitti

  1. Download the pretrained model here and put it to $ROOT/data/model/snake/kins/149.pth.
  2. Visualize:
    python run.py --type visualize --cfg_file configs/kins_snake.yaml test.dataset KinsVal ct_score 0.3
    

Visualization on Sbd

  1. Download the pretrained model here and put it to $ROOT/data/model/snake/sbd/149.pth.
  2. Visualize:
    python run.py --type visualize --cfg_file configs/sbd_snake.yaml test.dataset SbdVal ct_score 0.3
    

Demo

We support demo for image and image folder using python run.py --type demo --cfg_file /path/to/yaml_file demo_path /path/to/image ct_score 0.3.

For example:

python run.py --type demo --cfg_file configs/sbd_snake.yaml demo_path demo_images ct_score 0.3
# or
python run.py --type demo --cfg_file configs/sbd_snake.yaml demo_path demo_images/2009_000871.jpg ct_score 0.3

If setup correctly, the output will look like

demo

Training

The training parameters can be found in project_structure.md.

Training on Cityscapes

Two-stage training:

  1. Train the detector:
    python train_net.py --cfg_file configs/city_ct_rcnn.yaml model rcnn_det
    
  2. Train the detector and snake together:
    python train_net.py --cfg_file configs/city_rcnn_snake.yaml model rcnn_snake det_model rcnn_det
    

Training on Kins

python train_net.py --cfg_file configs/kins_snake.yaml model kins_snake

Training on Sbd

python train_net.py --cfg_file configs/sbd_snake.yaml model sbd_snake

Tensorboard

We provide tensorboard for seeing the training status:

# for the rcnn_snake task
tensorboard --logdir data/record/rcnn_snake
# for the snake task
tensorboard --logdir data/record/snake

If setup correctly, the output will look like

tensorboard

Citation

If you find this code useful for your research, please use the following BibTeX entry.

@inproceedings{peng2020deep,
  title={Deep Snake for Real-Time Instance Segmentation},
  author={Peng, Sida and Jiang, Wen and Pi, Huaijin and Li, Xiuli and Bao, Hujun and Zhou, Xiaowei},
  booktitle={CVPR},
  year={2020}
}