[Bug] Inference tab renders normalized tensors without inverse normalization.

[NANDA2726]

The Inference tab in tabs/inference.py displays the normalized image tensor directly without applying inverse normalization before rendering:

img_to_draw = transforms.ToPILImage()(sample_tensor[0])

This causes the output image to appear with distorted, psychedelic colors instead of the original image, making the inference visualization incorrect and misleading. The image tensor should be de-normalized before display so the output image appears correctly with original colours.

Proposed Fix

Add a denormalize helper function in tabs/inference.py and apply it before rendering:

def denormalize(tensor, mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225]):
    mean = torch.tensor(mean).view(3, 1, 1)
    std = torch.tensor(std).view(3, 1, 1)
    return torch.clamp(tensor.cpu() * std + mean, 0, 1)

denormalized = denormalize(sample_tensor[0])
img_to_draw = transforms.ToPILImage()(denormalized)

[dpascualhe]

Can you add a screenshot showing the reported behavior?

[NANDA2726]

Here is the screenshot:

The image appears to be normalized without inverse normalization, resulting in distorted colors during inference visualization.

[dpascualhe]

Hi Nanda, can you test it again with the lastest version? No other user has been suffering this problem. If you cannot replicate anymore, we will close the issue