search

bcmi / GracoNet-Object-Placement

Official code for ECCV2022 paper: Learning Object Placement via Dual-path Graph Completion
MIT License
73 stars 6 forks source link

Evaluation Details #5

Closed lingtianxia123 closed 1 year ago

lingtianxia123 commented 1 year ago

Thanks for sharing the code. I test GracoNet with your provided model and only obtain acc 0.740±0.005，FID 36.19 and LPIPS 0.205 on "evaluni" dataset. But I obtain acc 0.840±0.005 on "eval" dataset. Are these metrics obtained on different test data? But the paper describes "generates 10 composite images by randomly sampling 10 random vectors". Is it more reasonable to test all metrics in "evaluni" dataset?

Siyuan-Zhou commented 1 year ago

@lingtianxia123 In our paper, we use "eval" dataset for testing generation plausibility (acc and FID) and "evaluni" dataset for testing generation diversity (LPIPS).