video-maskrcnn

Video Mask R-CNN

View project on GitHub

Video Object Self-Supervised Learning

Alt text

Visit our Project Page for accessing the paper, and the pre-computed results.

We tested the code on python 3.7, PyTorch 1.2.0 and CUDA 10.1

Installation

If you need to run Self-Annotation, please install Annotation Interface and interfere the training process in run.sh.

Prepare Data

mkdir -p -- CityScapes
cd CityScapes
mkdir -p -- val
cd val
mkdir -p -- Raw
cd Raw

Download leftImg8bit_sequence_trainvaltest.zip (324GB) and extract all sequences of val-set to CityScapes/val/Raw. For example, ./CityScapes/val/Raw/frankfurt/frankfurt_000000_000275_leftImg8bit.jpg

Download and extract our pre-trained model to CityScapes/val/Initial_model

Citations

Please consider citing this project in your publications if it helps your research:

@Inproceedings{ltnghia-WACV2020,
  Title          = {Toward Interactive Self-Annotation For Video Object Bounding Box: Recurrent Self-Learning And Hierarchical Annotation Based Framework},
  Author         = {Trung-Nghia Le and Akihiro Sugimoto and Shintaro Ono and Hiroshi Kawasaki},
  BookTitle      = {IEEE Winter Conference on Applications of Computer Vision},
  Year           = {2020},
}

License

The code is released under the Creative Commons Attribution-NonCommercial-ShareAlike 3.0 License, and used for academic purpose only.

Contact

Trung-Nghia Le.