dcgan.torch icon indicating copy to clipboard operation
dcgan.torch copied to clipboard

Image Generated are blurred

Open hellorp1990 opened this issue 8 years ago • 2 comments

Hi, I am using CPU to generate images,

At first change the code from GPU to cpu. then training by this command-- DATA_ROOT=myimages dataset=folder th main.lua

the generating image

gpu=0 batchSize=64 net=experiment1_25_net_G.t7 th generate.lua

it is generating one image and blurry.

at first i tried medical images , the image that i got is blurred. Then i thought it might be because they are grayscale , then i tried some rgb images (Horse dataset) and then did the same thing but the image generated is again blurred.

I tried three different generator (generate 5/generate 10/ generator 25) but all are blurred. I am attaching them. ![5](https://cloud.githubusercontent.com/assets/19805255/23264269/c43fdf2c-f9ae-11e6-877d-ef80e72bf1a0.png 10 5

)

Please help if I am doing wrong.

hellorp1990 avatar Feb 23 '17 15:02 hellorp1990

If you have fewer than 10000 images in your dataset you need to increase the number of epochs the model trains for. Set nist=500 or 100 and let it run.

Then use one of the higher saved models for the image generation. For example gpu=0 batchSize=64 net=experiment1_500_net_G.t7 th generate.lua

darinc avatar Mar 05 '17 16:03 darinc

I also encountered this problem using gpu and over 200k images

xhchrn avatar Mar 09 '17 03:03 xhchrn