search

huangzehao / caffe-vdsr

A Caffe-based implementation of very deep convolution network for image super-resolution
MIT License
273 stars 134 forks source link

Data Augmentation #23

Closed saeed-anwar closed 7 years ago

saeed-anwar commented 7 years ago

Hi,

I would like to know how you increased the images from 91 to 1638.

Any info will be highly appreciated.

Regards,

huangzehao commented 7 years ago

Hi, check this issue. #19 Thanks.

saeed-anwar commented 7 years ago

Thanks,

I seems that you have applied around 17 to 18 different transformation, can you share the training file, where these transformation occurs.

Cheers,

huangzehao commented 7 years ago

Sorry, I can't fine the code of data augmentation right now since I wrote this code long ago.

You can write it by yourself. It's very simple in matlab. Like this:

image_rotate = imrotate(image,rotate_angle);
image_flip = flip(image,2);
image_translate = immove(image,translate_x,translate_y);
image_zoom = imresize(image,zoom_scale,'bicubic');
saeed-anwar commented 7 years ago

I am doing it in matlab, but I am not getting the same results as yours. Thats why I asked for the specific data augmentation types.

cheers.

huangzehao commented 7 years ago

How many training samples you used during training? And what's your solver setting?

saeed-anwar commented 7 years ago

The solver setting is same as yours, and the number of samples are around 600k from (582 data augmented, flipped only from 291 images)

huangzehao commented 7 years ago

Hi, you should modify the solve setting since your number of training samples is different from mine. Check the readme I just updated, and do training again. I hope this helps. Thanks!