Please confirm the commands to reproduce all four stages.

[Kamlesh364]

Stage 1:
CUDA_VISIBLE_DEVICES=0 python train_base.py \ --dataset-name videomatte240k --model-backbone resnet50 \ --model-name mattingbase-resnet50-videomatte240k \ --model-pretrain-initialization "pretraining/best_deeplabv3_resnet50_voc_os16.pth" \ --epoch-end 8

Stage 2:
CUDA_VISIBLE_DEVICES=0 python train_refine.py \ --dataset-name videomatte240k \ --model-backbone resnet50 \ --model-name mattingrefine-stage2-resnet50-videomatte240k \ --model-last-checkpoint "./checkpoint/mattingbase-resnet50-videomatte240k/epoch-7.pth" \ --epoch-end 1

Stage 3:
CUDA_VISIBLE_DEVICES=0 python train_refine.py \ --dataset-name photomatte13k \ --model-backbone resnet50 \ --model-name mattingrefine-stage3-resnet50-photomatte13k \ --model-last-checkpoint "./checkpoint/mattingbase-resnet50-videomatte240k/epoch-0.pth" \ --epoch-end 25

Stage 4:
CUDA_VISIBLE_DEVICES=0 python train_refine.py \ --dataset-name distinction \ --model-backbone resnet50 \ --model-name mattingrefine-stage4-resnet50-distinction \ --model-last-checkpoint "./checkpoint/mattingrefine-stage3-resnet50-photomatte13k/epoch-24.pth" \ --epoch-end 30

Also, when I use the Stage 2 command on GPUs(0,1,2,3)-

Please verify if this is unusual!

@PeterL1n @jinzishuai @andreyryabtsev Thanks!