search

eti-p-doray / unet-gan-matting

Background Removal based on U-Net
MIT License
294 stars 83 forks source link

I didn't find the GAN part in the project #29

Closed bigblank closed 4 years ago

bigblank commented 4 years ago

In the beginning, I used the model I trained to eval found that the picture output was too pink. I thought maybe the dataset I trained was too little, but I used the 83256 you provided to eval the result is also pink, just like the figure 6(a) in your paper when GAN is not used.

bigblank commented 4 years ago

In the beginning, I used the model I trained to eval found that the picture output was too pink. I thought maybe the dataset I trained was too little, but I used the 83256 you provided to eval the result is also pink, just like the figure 6(a) in your paper when GAN is not used.

If the data set is large, the effect will be good. I started with too few data sets.