How to prediction on single image

[masudrehman]

sir, How to do prediction on single image?

[moein-shariatnia]

sir, How to do prediction on single image?

Hey, If you are following along the Colab notebook, you can use the following code snippet to do prediction on a single image and visualize the final result. Note that you need to define the model (as noted in the notebook) and load the pretrained weights that I've provided there:

img = PIL.Image.open("black_and_white.jpg")
img = img.resize((256, 256))
# to make it between -1 and 1
img = transforms.ToTensor()(img)[:1] * 2. - 1.
model.eval()
with torch.no_grad():
    preds = model.net_G(img.unsqueeze(0).to(device))
plt.imshow(lab_to_rgb(img.unsqueeze(0), preds.cpu())[0])

I hope this helps you out. Please let me know if there is any problem. Good Luck!

[masudrehman]

thanks it worked

[masudrehman]

Thank you sir

On Sun, Dec 27, 2020, 10:05 PM moein-shariatnia notifications@github.com wrote:

sir, How to do prediction on single image?

Hey, If you are following along the Colab notebook, you can use the following code snippet to do prediction on a single image and visualize the final result. Note that you need to define the model (as noted in the notebook) and load the pretrained weights that I've provided there:

img = PIL.Image.open("black_and_white.jpg") img = img.resize((256, 256))

to make it between -1 and 1

img = transforms.ToTensor()(img)[:1] * 2. - 1. model.eval() with torch.no_grad(): preds = model.net_G(img.unsqueeze(0).to(device)) plt.imshow(lab_to_rgb(img.unsqueeze(0), preds.cpu())[0])

I hope this helps you out. Please let me know if there is any problem. Good Luck!

— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub https://github.com/moein-shariatnia/Deep-Learning/issues/1#issuecomment-751491923, or unsubscribe https://github.com/notifications/unsubscribe-auth/AOUBCEOSFMJZATYA42OXGQDSW5SOXANCNFSM4VK2HBJQ .

[moein-shariatnia]

Hey, you should be actually saving "model.state_dict()" instead of "model" itself. I think this would resolve your error.

On Sun, Dec 27, 2020 at 9:01 PM masudbarki notifications@github.com wrote:

also sir if you help me with this error AttributeError: Can't get attribute 'MainModel' on <module 'main'> this error happens when i saved the model with torch.save(model," fullmodel.pt') when i was using the same colab in which i have done the training path, the torch.load("fullmodel.pt") worked perfectly.. but in another colab where i uploaded the fullmodel.pt it gives the error AttributeError: Can't get attribute 'MainModel' on <module 'main'> when loading the model with torch.load("fullmodel.pth")..

— You are receiving this because you commented. Reply to this email directly, view it on GitHub https://github.com/moein-shariatnia/Deep-Learning/issues/1#issuecomment-751494569, or unsubscribe https://github.com/notifications/unsubscribe-auth/ARXTC26FYROP5TKF2PRY2UTSW5VPJANCNFSM4VK2HBJQ .

[masudrehman]

Thanks..it works now perfectly...

On Tue, Dec 29, 2020, 3:54 PM moein-shariatnia notifications@github.com wrote:

Hey, you should be actually saving "model.state_dict()" instead of "model" itself. I think this would resolve your error.

On Sun, Dec 27, 2020 at 9:01 PM masudbarki notifications@github.com wrote:

also sir if you help me with this error AttributeError: Can't get attribute 'MainModel' on <module 'main'> this error happens when i saved the model with torch.save(model," fullmodel.pt') when i was using the same colab in which i have done the training path, the torch.load("fullmodel.pt") worked perfectly.. but in another colab where i uploaded the fullmodel.pt it gives the error AttributeError: Can't get attribute 'MainModel' on <module 'main'> when loading the model with torch.load("fullmodel.pth")..

— You are receiving this because you commented. Reply to this email directly, view it on GitHub < https://github.com/moein-shariatnia/Deep-Learning/issues/1#issuecomment-751494569 , or unsubscribe < https://github.com/notifications/unsubscribe-auth/ARXTC26FYROP5TKF2PRY2UTSW5VPJANCNFSM4VK2HBJQ

.

— You are receiving this because you modified the open/close state. Reply to this email directly, view it on GitHub https://github.com/moein-shariatnia/Deep-Learning/issues/1#issuecomment-752034171, or unsubscribe https://github.com/notifications/unsubscribe-auth/AOUBCEK4ZFMHXP5GTTOCXN3SXGYP7ANCNFSM4VK2HBJQ .

[radudiaconu0]

sir, How to do prediction on single image?

Hey,

If you are following along the Colab notebook, you can use the following code snippet to do prediction on a single image and visualize the final result. Note that you need to define the model (as noted in the notebook) and load the pretrained weights that I've provided there:

img = PIL.Image.open("black_and_white.jpg")

img = img.resize((256, 256))

# to make it between -1 and 1

img = transforms.ToTensor()(img)[:1] * 2. - 1.

model.eval()

with torch.no_grad():

    preds = model.net_G(img.unsqueeze(0).to(device))

plt.imshow(lab_to_rgb(img.unsqueeze(0), preds.cpu())[0])

I hope this helps you out. Please let me know if there is any problem. Good Luck!

Why we transforms it between -1 and 1 and not 0 1

[moein-shariatnia]

Why we transforms it between -1 and 1 and not 0 1 @radudiaconu0

The model has been trained in this way so the evaluation must be done in the same way. The generator architecture contains a Tanh function in the last layer, so, the generated image is between -1 and 1. So, the real images coming to discriminator should be in the same range. That's the reason why we scale between -1 and 1 in the dataset.