search

PangzeCheung / Dual-task-Pose-Transformer-Network

[CVPR 2022] Exploring Dual-task Correlation for Pose Guided Person Image Generation
Other
95 stars 11 forks source link

Fashion result looks terrible #15

Closed soon-yau closed 1 year ago

soon-yau commented 1 year ago

fashionWOMENBlouses_Shirtsid0000523301_1front_2_fashionWOMENBlouses_Shirtsid0000523301_4full_vis

I downloaded the model and run inference. By the way, the image pre-processing file did not work straight away, had to make some changes to get the path name correct. Anyway, the image results I got look aweful, not just a few, but all of them. Have I done anything wrong?

I unzip epoch190.zip into checkpoints/DPTN_fashion and run python test.py --name=DPTN_fashion --model=DPTN --dataset_mode=fashion --dataroot=./dataset/fashion --which_epoch latest --results_dir ./results/DPTN_fashion --batchSize 1 --gpu_id=0

PangzeCheung commented 1 year ago

@soon-yau Thanks for your attention. Our preprocessing method is identical to the PATN method, which employs 256 * 256 images rather than high-resolution image datasets. I hope this information proves helpful to you~