Using DDP to pretrain Lamps:
CUDA_VISIBLE_DEVICES="0,1" python -m torch.distributed.launch --nproc_per_node=2 --master_port 28301 main.py --arch swin_base --batch_size_per_gpu 8 --data_path pretrain image path --output_dir
Pretrained weights
| Backbone | Input Resolution | Pretrain dataset | model |
|---|---|---|---|
| Swin-B | 448x448 | ChestX-ray14 | download |
| Swin-B | 448x448 | Large-scale | download |