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

Can not generate the weight file when use my train images #49

Open alvin2ye opened 4 years ago

alvin2ye commented 4 years ago

Thank you for your share.

My train command:

python train.py --image_dir dataset/train --test_dir dataset/test --image_size 128 --batch_size 8 --lr 0.001 --source_noise_model text,0,50 --target_noise_model text,0,50 --val_noise_model text,25,25 --loss mae --output_path text_noise When i use your train images, It's working but the result is not very good. So I use my train images: (All files is "JPG" and the size is : 100K~200K), but It's stop here

image

Thank you

alvin2ye commented 4 years ago

stop here about 2 hours.

alvin2ye commented 4 years ago

when the train 4500 files, test 700 files is stop.

but 1000 files, test 150 files is ok.

I am afraid the effect

alvin2ye commented 4 years ago

Sorry , It's also stop. at Epoch 3

yu4u commented 4 years ago

Hmm, it's storage and hard to reproduce... The number of val images might matter because current code reads all val images and stores them on main memory. How about reducing the number of val images again?