Research map / Image generation / Editing and style transfer
Appearance and Pose-Conditioned Human Image Generation Using Deformable GANs
Aliaksandr Siarohin, Stéphane Lathuilière, Enver Sangineto, Nicu Sebe
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (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 |
|---|---|---|---|
| add deformable skip connections to U-Net baseline (ablation) | Baseline · Market-1501 | Detection Score: 0.595 -> 0.629 | Improved |
| add nearest-neighbour loss instead of L1 (ablation) | DSC · Market-1501 | Detection Score: 0.629 -> 0.720 | Improved |
| nearest-neighbour loss instead of perceptual loss (ablation) | DSC · Market-1501 | Detection Score: 0.603 -> 0.720 | Improved |
| nearest-neighbour loss instead of L1 (ablation) | DSC · Market-1501 | IS: 3.442 -> 3.185 | Mixed |
| max combination instead of linear combination of affine transformations (ablation) | Full · Market-1501 | mask-IS: 3.385 -> 3.502 | Improved |
| exploit body symmetry (ablation) | Full · Market-1501 | Detection Score: 0.698 -> 0.720 | Improved |
| use part masks when aligning features (ablation) | Full · Market-1501 | Detection Score: 0.693 -> 0.720 | Improved |
| compute nearest-neighbour loss on VGG Block 1_2 instead of pixel space (ablation) | Full · Market-1501 | Detection Score: 0.580 -> 0.720 | Improved |
| increase neighborhood size from n=3 to n=5 (ablation) | Full · Market-1501 | Detection Score: 0.720 -> 0.579 | Hurt |
| increase lambda from 0.01 to 0.1 (ablation) | Full · Market-1501 | IS: 3.185 -> 2.621 | Hurt |
| augment Re-ID training with Full generated images (x2) | IDE+Euclidean · Market-1501 Re-ID | Rank-1: 73.9 -> 78.9 | Improved |
| augment Re-ID training with Baseline generated images (x2) (ablation) | IDE+Euclidean · Market-1501 Re-ID | Rank-1: 73.9 -> 68.1 | Hurt |
Closest papers by meaning
- Deformable GANs for Pose-Based Human Image Generation (CVPR 2018 · similarity 0.95)
- Pose-Normalized Image Generation for Person Re-identification (ECCV 2018 · similarity 0.91)
- Unsupervised Person Image Synthesis in Arbitrary Poses (CVPR 2018 · similarity 0.90)
- FD-GAN: Pose-guided Feature Distilling GAN for Robust Person Re-identification (NeurIPS 2018 · similarity 0.90)
- Pose Guided Person Image Generation (NeurIPS 2017 · similarity 0.89)
- Progressive Pose Attention Transfer for Person Image Generation (CVPR 2019 · similarity 0.89)
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.
