Research map / Detection and segmentation / Open-set and few-shot detection
A Dense Material Segmentation Dataset for Indoor and Outdoor Scene Parsing
Paul Upchurch, Ransen Niu
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 |
|---|---|---|---|
| train on DMS data instead of OpenSurfaces (same architecture) | ResNet-50 dilated + PPM · DMS-Val (25 materials) | Acc: 0.589 -> 0.753 | Improved |
| train on DMS data instead of OpenSurfaces (same architecture) | ResNet-50 dilated + PPM · Avg-Val (25 materials) | mIoU: 0.416 -> 0.500 | Improved |
| train on DMS data instead of OpenSurfaces (same architecture) | ResNet-50 dilated + PPM · Cross-dataset gap Delta (25 materials) | Acc delta between datasets: 0.231 -> 0.047 | Improved |
| train on DMS data instead of OpenSurfaces (same architecture) | ResNet-50 dilated + PPM · OS-Val (25 materials) | mAcc: 0.702 -> 0.692 | Hurt |
| train on DMS data instead of OpenSurfaces | ResNet-50 dilated + PPM · DMS-Val photographic type: distant unreachable outdoor things | per-pixel accuracy: 0.225 -> 0.736 | Improved |
| train on DMS data instead of OpenSurfaces | ResNet-50 dilated + PPM · DMS-Val photographic type: a real surface without context | per-pixel accuracy: 0.348 -> 0.691 | Improved |
| balance skin types in training data (DMS instead of OpenSurfaces) | skin recognition, ResNet-50 dilated + PPM · DMS-Val skin types | accuracy gap across skin types: 13.3% -> 3% | Improved |
| select images to increase the underrepresented skin group | DMS image selection · DMS skin type distribution | ratio of largest to smallest skin group: 9.4 -> 2.2 | Improved |
| select images by photographic type to grow outdoor coverage | DMS image selection · DMS outdoor material counts | occurrences of outdoor materials: growing the dataset by 17% more than doubled 9 kinds of outdoor materials | Improved |
| drop six rare classes (52 -> 46 materials) | DMS-52 · DMS-Val | mIoU: 0.392 -> 0.435 | Improved |
| add OpenSurfaces as extra training data to DMS-52 | DMS-52 · DMS-Val (52 materials) | worst classes | No clear effect |
Closest papers by meaning
- Infusing Synthetic Data with Real-World Patterns for Zero-Shot Material State Segmentation (NeurIPS 2024 · similarity 0.87)
- Multimodal Material Segmentation (CVPR 2022 · similarity 0.87)
- Intrinsic Image Diffusion for Indoor Single-view Material Estimation (CVPR 2024 · similarity 0.85)
- MaSS13K: A Matting-level Semantic Segmentation Benchmark (CVPR 2025 · similarity 0.85)
- Hierarchy-Agnostic Unsupervised Segmentation: Parsing Semantic Image Structure (NeurIPS 2024 · similarity 0.85)
- SUM Parts: Benchmarking Part-Level Semantic Segmentation of Urban Meshes (CVPR 2025 · similarity 0.84)
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.
