Predict black images

[albertoursino]

Hello, why the model predicts a black image?

If I pass as input a jpg image:

Same behaviour with other images.

I downloaded the pretrained model using these lines of code you provided in the README:

# Imports
import torch
import torch.hub

# Create the model
model = torch.hub.load(
    repo_or_dir='guglielmocamporese/hands-segmentation-pytorch',
    model='hand_segmentor',
    pretrained=True
)

Then I created the dir "checkpoint" on your project directory and put the model there.
Finally, I executed the main passing the predict parameters:

python main.py \
	--mode predict \
	--data_base_path './test_images' \
	--model_checkpoint "./checkpoint/checkpoint.ckpt" \
	--model_pretrained

Note that I'm using Windows, that's why I didn't run the sh(s) you provided.

Thanks for your time!

[guglielmocamporese]

Hi albertoursino,

thank you for pointing out this issue. I solved it with the latest commit - this is what I obtained with the fixed code (I used the same commands you reported above in your comment to obtain this output):

Let me know,

Guglielmo

[guglielmocamporese]

Hi Tanveer, Thanks for reaching out - probably the issue is that you are not normalizing the input correctly. Here you can find the correct function used as "transform": transform = transforms.Compose([

transforms.Resize((256, 256)), transforms.ToTensor(), lambda x: x if x.shape[0] == 3 else x.repeat(3, 1, 1), transforms.Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225]),

]) Hope this help. Best, Guglielmo

…

On Wed, Jun 14, 2023 at 10:50 AM Tanveer Hussain ***@***.***> wrote: def preprocess_image(img): transform = T.Compose([T.Resize((256, 256)), T.ToTensor()]) x = transform(img) x = torch.unsqueeze(x, 0) x = x.cuda() return x def postprocess_image(preds, H, W): # preds = F.softmax(output, 1).argmax(1)[0] * 255 # [h, w] preds = Image.fromarray(preds.numpy().astype(np.uint8), 'L') preds = preds.resize((W, H)) return preds — Reply to this email directly, view it on GitHub <#5 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AHUB6O3Z5PEYIQRHEEUOV73XLF3OLANCNFSM52TPGT6A> . You are receiving this because you modified the open/close state.Message ID: <guglielmocamporese/hands-segmentation-pytorch/issues/5/1590759638@ github.com>