How to solve issues during inference code

[intMinsu]

When you run inference like repo says:
python ./src/main.py validate --config ./configs/val.yaml --trainer.logger null
you will face many errors.
This issue will guide how to solve these.

1. Error in "python src/main.py fit --config ./configs/train_stage1.yaml" #6 error: Parser key "model"
   Reason: possibly related to torchmetrics version mismatch.
   Fix: change the lines importing FID and LPIPS in UniRestore/src/core/base/eval_image_restoration.py and UniRestore/src/core/base/eval_multi_task.py

• Original code :

from torchmetrics.image import FrechetInceptionDistance as FID
from torchmetrics.image import LearnedPerceptualImagePatchSimilarity as LPIPS

• Fixed code :

from torchmetrics.image.fid import FrechetInceptionDistance as FID
from torchmetrics.image.lpip import LearnedPerceptualImagePatchSimilarity as LPIPS

2. Errors Encountered During Inference #4 ImportError: cannot import name 'TaskEditorV1c' from 'modules.diffuie.taskeditor'
   Reason: wrong class name.
   Fix: change class name in UniRestore/src/modules/diffuie/autoencoder.py

• Original Code :

if self.tedit_type == "TFA":
    from .taskeditor import TaskEditorV1c as TaskEditor

• Fixed Code :

if self.tedit_type == "TFA":
    from .taskeditor import TaskFeatureAdapter as TaskEditor

3. Errors Encountered During Inference #4 TypeError: SkipConnectedAutoEncoder.forward() missing 1 required positional argument: 'task'
   Reason: calflops.calculate_flops cannot handle stable diffusion VAE forward naively
   Fix: just comment out calculate_flops part in UniRestore/src/modules/diffuie/unifie.py

• Fixed Code :

# flops, macs, params = calculate_flops(model = self.ae,
#                                       input_shape=input_shape,
#                                       output_as_string=True,
#                                       output_precision=4)
# print("UniRestore.AEs FLOPs:%s   MACs:%s   Params:%s \n" %(flops, macs, params))
# raise TypeError

# for name, param in self.ae.vae.decoder.named_parameters():
#     param.register_hook(lambda grad, name=name: print(f"Gradient computed for: {name}"))

Hope it helps