Research map / Image generation / Editing and style transfer
Disentangled Person Image Generation
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| use the two-stage method instead of PG² | person image generation · DeepFashion pose-guided generation | IS: 3.090 -> 3.228 | Helped |
| use the two-stage method instead of PG² | person image generation · DeepFashion pose-guided generation | SSIM: 0.762 -> 0.614 | Hurt |
| use the two-stage method instead of PG² | person image generation · Market-1501 pose-guided generation | mask-IS: 3.435 -> 3.491 | Helped |
| use the two-stage method instead of PG² | person image generation · Market-1501 pose-guided generation | SSIM: 0.253 -> 0.099 | Hurt |
| extract foreground features from seven body ROIs rather than whole feature maps (ablation) | Ours re-ID feature encoder · Market-1501 re-ID | Rank-1: 0.307 -> 0.338 | Helped |
| apply PCA to BodyROI7 features (ablation) | BodyROI7 re-ID feature encoder · Market-1501 re-ID | Rank-1: 0.338 -> 0.355 | Helped |
| add PUL using Market-1501 data | Res50 trained on VM · Market-1501 re-ID | Rank-1: 0.338 -> 0.369 | Helped |
| add KISSME metric learning | Res50+PUL trained on VM+Market · Market-1501 re-ID | Rank-1: 0.369 -> 0.375 | Helped |
| use real pose keypoints from training data while sampling foreground and background (ablation) | BodyROI7 image sampler · Market-1501 image sampling | visual realism | Helped |
| use seven body ROIs instead of whole-body foreground features (ablation) | Ours image sampler · Market-1501 image sampling | visual realism | Helped |
| use vanilla GAN loss instead of Wasserstein GAN loss for embedding mapping (ablation) | Stage-2 embedding mapper · embedding feature mapping | training stability | Hurt |
Closest papers by meaning
- Learning Disentangled Representation for Robust Person Re-identification (NeurIPS 2019 · similarity 0.88)
- Unsupervised Person Image Synthesis in Arbitrary Poses (CVPR 2018 · similarity 0.87)
- Disentangled and Controllable Face Image Generation via 3D Imitative-Contrastive Learning (CVPR 2020 · similarity 0.87)
- Pose Guided Person Image Generation (NeurIPS 2017 · similarity 0.87)
- Person Image Synthesis via Denoising Diffusion Model (CVPR 2023 · similarity 0.87)
- Transferring disentangled representations: bridging the gap between synthetic and real images (NeurIPS 2024 · similarity 0.86)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.