Research map / Vision-language and multimodal / Human and embodied modeling
Unified Perceptual Parsing for Scene Understanding
Tete Xiao, Yingcheng Liu, Bolei Zhou, Yuning Jiang, Jian Sun
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 Pyramid Pooling Module to FPN (ablation) | FPN (/4) with ResNet-50 · ADE20K val (object parsing) | mean IoU / pixel accuracy: 4.87/3.09 | Improved |
| fuse features from all FPN levels (ablation) | FPN+PPM (/4) · ADE20K val (object parsing) | mean IoU: 40.13 -> 41.22 | Improved |
| increase FPN output resolution from /16 to /4 (ablation) | FPN · ADE20K val (object parsing) | mean IoU: 34.46 -> 35.26 | Improved |
| use FPN instead of dilated network (ablation) | FPN (/16) vs DilatedNet reproduced · ADE20K val (object parsing) | training time (hr): 18.1 vs 53.9 hr; mIoU 34.46 vs 34.28 | Improved |
| use UPerNet instead of PSPNet (ablation) | UPerNet ResNet-50 · ADE20K val (object parsing) | training time (hr): 61.1 -> 38.7 | Improved |
| add part annotations to object training (ablation) | UPerNet · Broden+ val (object) | mean IoU: 24.72 -> 23.92 | No clear effect |
| add scene classification branch (ablation) | UPerNet object+part · Broden+ val (object) | mean IoU: 23.92 -> 23.83 | No clear effect |
| add material parsing to joint training (ablation) | UPerNet object+part+scene · Broden+ val (object) | mean IoU: 23.83 -> 23.36 | Hurt |
| add material parsing to joint training (ablation) | UPerNet object+part+scene · Broden+ val (part) | mean IoU with background: 30.10 -> 28.75 | Hurt |
| train material jointly with object, part and scene (ablation) | UPerNet · Broden+ val (material) | mean IoU: 52.78 -> 54.19 | Improved |
| add part annotations to object training | UPerNet · Broden+ val (part) | mean IoU with background: 30.21 | Mixed |
| fine-tune texture branch separately after other tasks (not mixed in) | UPerNet · Broden+ val (texture) | top-1 accuracy: 35.10 | Mixed |
Closest papers by meaning
- Scene Parsing With Global Context Embedding (ICCV 2017 · similarity 0.87)
- UniPercept: Towards Unified Perceptual-Level Image Understanding across Aesthetics, Quality, Structure, and Texture (ICML 2026 · similarity 0.86)
- Bridging the gap to real-world language-grounded visual concept learning (NeurIPS 2025 · similarity 0.86)
- Aligning Forest and Trees in Images & Long Captions for Visually Grounded Understanding (NeurIPS 2026 · similarity 0.85)
- PerceptionCLIP: Visual Classification by Inferring and Conditioning on Contexts (ICLR 2024 · similarity 0.85)
- Perceive Anything: Recognize, Explain, Caption, and Segment Anything in Images and Videos (NeurIPS 2025 · similarity 0.85)
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.
