sangyun884 / HR-VITON

Official PyTorch implementation for the paper High-Resolution Virtual Try-On with Misalignment and Occlusion-Handled Conditions (ECCV 2022).
825 stars 171 forks source link

2. Human parse CIHP_PGN모델 전처리 결과 이미지가 다릅니다. #73

Open doithun opened 1 year ago

doithun commented 1 year ago

Preprocessing.md 에 적힌 경로의 소스코드를 다운받아서 전처리 이미지를 추출하면

[Original Result] 00006_00

[After image-parse-agnostic] 00006_00

[My Preprocessing Result] 00006_00

[After image-parse-agnostic] 00006_00

안녕하세요.

현재 HR-VITON 모델을 사용하기 위하여 전처리 데이터를 준비하고 있습니다.

  1. Human parse CIHP_PGN모델 전처리를 하였는데

위의 이미지와 같이

HR-VITON의 데이터의 이미지와, 제가 추출한 이미지의 비트 수준도 다르고

객체의 경계면도 다릅니다.

추출한 이미지를 get_im_parse_agnostic 함수를 적용하여 마스킹을 하면

원본은 잘 분리가 되는데 제가 전처리한 이미지는 헤어부분에서 마스킹을 시작합니다.

이미지의 라벨링 값이 달라서 그런것 같은데 코드의 어느 부분을 조정해야 할까요?

dataset(VITON-HD) is just for the visualization, it has 0~19 uint values.

데이터 셋 이미지 저장 후 0 ~ 19 값으로 팔레트 이미지 업스케일링을 어떤 방식으로 하셨을까요???

SamLights commented 1 year ago

did you solve this problem ?