To launch training: python -m torch.distributed.launch --master_port 25763 --nproc_per_node=2 train.py --exp_id EXPERIMENT_NAME --stage 3 --load_network saves/XMem-s012.pth
Original training data has been moved to Dodo at /data/thom/xmem_training_data
You configure the dataloaders in train.py
in the functions named renew_<datasetname>_loader
The format for each "video" is a folder of Annotations (.png images, 2D, 0,1) and a folder of images.
For in context learning you would just have 1 video.