The MaskTrack method is the baseline for state-of-the-art methods in video object segmentation like Video Object Segmentation with Re-identification and[ Lucid Data Dreaming for Multiple Object Tracking] (https://arxiv.org/abs/1703.09554). The top three methods in DAVIS 2017 challenge were based on the MaskTrack method. However, no open source code is available for the MaskTrack method. Here I provide the MaskTrack method with following specifications:
Machine configuration used for testing:
Offline training is done on DAVIS 2017 train data. The online training and testing is done on DAVIS 2017 test dataset. I recommend using conda for downloading and managing the environments.
Download the Deeplab Resnet 101 pretrained COCO model from here and place it in 'training/pretrained' folder.
If you want to skip offline training and directly perform online training and testing, download the offline trained model from here and place it in 'training/offline_save_dir/lr_0.001_wd_0.001' folder.
What things you need to install the software and how to install them
Software used:
Dependencies: Create a conda environment using the training/deeplab_resnet_env.yml file. Use: conda env create -f training/deeplab_resnet_env.yml
If you are not using conda as a package manager, refer to the yml file and install the libraries manually.
Please refer to the following instructions:
A. Offline training data generation
B. Setting paths for python files
C. Offline training
D. Online Training data generation
E. Online Training and testing
This project is licensed under the MIT License - see the LICENSE file for details
This code was produced during my internship at Nanyang Technological University under Prof. Guosheng Lin. I would like to thank him for providing access to the GPUs.
I would like to thank K.K. Maninis for providing this code: https://github.com/kmaninis/OSVOS-PyTorch for reference.