Open loftusa opened 2 years ago
I have a sane problem, now fixed with this: if lr_img.shape[2] == 4:
lr_img = lr_img[:, :, :3]
full code:
import numpy as np from PIL import Image from ISR.models import RRDN, rdn # Load the image img = Image.open('dd.png') # Resize the image to match the expected input shape (e.g., 3 color channels) # Convert the image to a NumPy array lr_img = np.array(img) if lr_img.shape[2] == 4: # If it has four channels, remove the alpha channel (assumed to be the fourth channel) lr_img = lr_img[:, :, :3] # Load the pre-trained RDN model # rdn = RDN(weights='noise-cancel') rrdn = RRDN(weights='gans') sr_img = rrdn.predict(lr_img) # Convert the NumPy array back to an Image object sr_img = Image.fromarray(sr_img) # Save the super-resolved image sr_img.save('output.png')
Hi, I'm getting this error when I try to predict after training:
I trained on a set of 64x64 images, with 512x512 upscaled versions. I split the original full set of 64x64 images into a training set and a validation set, and tried to predict with the validation set. That's when I got this error. I'm not sure why the generator is expecting a 40x40 image as input, given that these weights were trained on 64x64 images.
Here is the full code for training / running: