search

menyifang / ADGAN

The Implementation of paper "Controllable Person Image Synthesis with Attribute-Decomposed GAN" CVPR 2020 (Oral); Pose and Appearance Attributes Transfer;
475 stars 90 forks source link

Unable to reproduce the generated image #30

Open ryhhtn opened 3 years ago

ryhhtn commented 3 years ago

I prepare dataset and scripts, but unable to reproduce the generated image.

If possible, please share the detailed READ.ME.

fashionMENJackets_Vestsid0000065304_1front jpg___fashionMENJackets_Vestsid0000065304_4full jpg_vis

ryhhtn commented 3 years ago

I'm using following library

torch==1.7.1+cu92
torchfile==0.1.0
torchinfo==0.0.9
torchmetrics==0.3.1
torchtext==0.8.1
torchvision==0.8.2+cu92
kobrien-Novetta commented 3 years ago

I can also confirm that these are the kinds of images being produced by both the 800 and 1000 .pth checkpoints. example

kobrien-Novetta commented 3 years ago

Looking at issue #6, apparently it only works on 2 gpus. If you change the gpu input to 0,0 you'll be able to run it on 1 gpu! If you can't pass 0,0 into on the command line, then you can just edit the default paramter for gpu_ids to be '0,0'

ryhhtn commented 3 years ago

thank you for your reply. I succeeded with reference to #6 and advice fashionMENTees_Tanksid0000016602_2side jpg___fashionMENTees_Tanksid0000016602_1front jpg_vis