search

yu4u / noise2noise

An unofficial and partial Keras implementation of "Noise2Noise: Learning Image Restoration without Clean Data"
MIT License
1.07k stars 237 forks source link

When using a generator for validation data, you must specify a value for `validation_steps` #15

Open zjz5250 opened 5 years ago

zjz5250 commented 5 years ago

When using a generator for validation data, you must specify a value for validation_steps

I trained the model successfully before，but i when i train it again， i met this problem，how should i do，thx！

yu4u commented 5 years ago

You might using Keras < 2.1.2. Please use Keras >= 2.1.2.

zjz5250 commented 5 years ago

yes，after i update to 2.2.0，it can work now。 thank you very much！ by the way，"train model using (noise, noise) pairs", does it mean use two noised image of one image？one image occured twice with different noise，rght？

zjz5250 commented 5 years ago

in fact，l want to know，what does it mean ： (noise, noise) or (noise, clean)

zjz5250 commented 5 years ago

I don't understand why each image of your train dataset is clean？if all of my dataset are noise image，some of them are the same image with different noise，can i train a general model？

zjz5250 commented 5 years ago

@yu4u hope your answer，thanks a lot

yu4u commented 5 years ago

by the way，"train model using (noise, noise) pairs", does it mean use two noised image of one image？one image occured twice with different noise，rght？

Yes. clean image -> noisy image1 -> noisy image2

(noise, noise) or (noise, clean) means (input, target) pair. (noise, clean) is a standard training setting.

XuDongHecs commented 5 years ago

---python clean_patch = image[i:i + image_size, j:j + image_size] x[sample_id] = self.source_noise_model(clean_patch) y[sample_id] = self.target_noise_model(clean_patch)

this means you train with input clean data + noise and output is org clean data + noise ?