Research map / Detection and segmentation / Open-set and few-shot detection
Inferring and Leveraging Parts From Object Shape for Improving Semantic Image Synthesis
Yuxiang Wei, Zhilong Ji, Xiaohe Wu, Jinfeng Bai, Lei Zhang, Wangmeng Zuo
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 global adversarial loss to the OASIS baseline (ablation) | semantic image synthesis · Cityscapes | FID: 47.7 -> 43.6 | Improved |
| add object-level CLIP style loss after global adversarial loss (ablation) | semantic image synthesis · Cityscapes | FID: 43.6 -> 42.8 | Improved |
| add inferred parts and sequential PSM to the loss-augmented model (ablation) | semantic image synthesis · Cityscapes | obj FID: 37.5 -> 30.4 | Improved |
| replace sequential PSM with SPADE part injection (ablation) | iPOSE · Cityscapes | FID: 41.3 -> 42.9 | Hurt |
| replace sequential PSM with concatenation part injection (ablation) | iPOSE · Cityscapes | FID: 41.3 -> 42.7 | Hurt |
| replace object-level CLIP style loss with VGG loss (ablation) | iPOSE · Cityscapes | FID: 41.3 -> 41.6 | Hurt |
| increase support maps from 1 to 3 (ablation) | PartNet · novel part categories | AC: 81.41 -> 85.00 | Improved |
| increase support maps from 1 to 3 (ablation) | PartNet-powered synthesis · COCO | FID: 15.76 -> 15.72 | Improved |
| add a multi-scale encoder (ablation) | PartNet · novel part categories | AC: 83.21 -> 85.00 | Improved |
| increase cross-attention blocks from 1 to 3 (ablation) | PartNet · novel part categories | AC: 82.28 -> 85.00 | Improved |
| increase cross-attention blocks from 3 to 4 (ablation) | PartNet · novel part categories | AC: 85.00 -> 83.41 | Hurt |
| replace predicted part maps with ground-truth part maps (ablation) | iPOSE · Cityscapes | FID: 41.3 -> 40.8 | Improved |
Closest papers by meaning
- Learning Partonomic 3D Reconstruction from Image Collections (CVPR 2025 · similarity 0.88)
- Beyond Holistic Object Recognition: Enriching Image Understanding With Part States (CVPR 2018 · similarity 0.88)
- PartNet: A Large-Scale Benchmark for Fine-Grained and Hierarchical Part-Level 3D Object Understanding (CVPR 2019 · similarity 0.88)
- PARTONOMY: Large Multimodal Models with Part-Level Visual Understanding (NeurIPS 2025 · similarity 0.88)
- PartGen: Part-level 3D Generation and Reconstruction with Multi-view Diffusion Models (CVPR 2025 · similarity 0.87)
- SUM Parts: Benchmarking Part-Level Semantic Segmentation of Urban Meshes (CVPR 2025 · 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.
