search

JiahuiYu / generative_inpainting

DeepFill v1/v2 with Contextual Attention and Gated Convolution, CVPR 2018, and ICCV 2019 Oral
http://jiahuiyu.com/deepfill/
Other
3.27k stars 787 forks source link

about this problem #477

Closed atztao closed 3 years ago

atztao commented 4 years ago

image image image

hey, can you meet this problem? i want to know, this problem cause by the training time not enough or others ?

Pinakee15 commented 4 years ago

image image image

hey, can you meet this problem? i want to know, this problem cause by the training time not enough or others ?

Hey did you train on custom dataset or the places2 dataset. Thanks

atztao commented 4 years ago

image image image hey, can you meet this problem? i want to know, this problem cause by the training time not enough or others ?

Hey did you train on custom dataset or the places2 dataset. Thanks

I used custom dataset with the places2. Thanks.

atztao commented 4 years ago

I found small holes is good, but some large holes has the problem.Do you meet this?

JiahuiYu commented 3 years ago

No on standard benchmarks. Maybe you can start to train with small holes then progressively change to big holes.