set w/o restoration module

[lijain]

DIffbir How to set up w/o restoration module training

[0x3f3f3f3fun]

In our latest code, to train w/o restoration module, you just need to remove a few lines of code (as shown below):

DiffBIR/train_stage2.py

Lines 80 to 89 in e7974cb

	swinir: SwinIR = instantiate_from_config(cfg.model.swinir)
	sd = {
	(k[len("module."):] if k.startswith("module.") else k): v
	for k, v in torch.load(cfg.train.swinir_path, map_location="cpu").items()
	}
	swinir.load_state_dict(sd, strict=True)
	for p in swinir.parameters():
	p.requires_grad = False
	if accelerator.is_local_main_process:
	print(f"load SwinIR from {cfg.train.swinir_path}")

DiffBIR/train_stage2.py

Line 108 in e7974cb

swinir.eval().to(device)

DiffBIR/train_stage2.py

Line 131 in e7974cb

clean = swinir(lq)

and directly use lq as condition:

cond = pure_cldm.prepare_condition(clean, prompt) => cond = pure_cldm.prepare_condition(lq, prompt)