Pytorch code for Revisiting Sequence-to-Sequence Video Object Segmentation with Multi-Task Loss and Skip-Memory
Dependencie:
numpy==1.18.1,
sacred==0.8.1,
torch==1.5,
torchvision==0.6,
tqdm
For training, download the YoutubeVOS from here. Modify the configurations and the paths in train.py and run the following command.
python train.py
For inference with the pretrained model, download the weights from here and put them under Model directory.
Modify the data paths in submit_ytvos.py and run the following command.
python submit_ytvos.py with model_name='weights.pth'
In case of questions, please contact fatemeh.azimi@dfki.de.