Research map / Vision-language and multimodal / Human and embodied modeling
Visual Recognition by Request
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 ViRReq instead of conventional joint multi-task training | SegFormer-B5 · ADE20K validation | HPQ (All, %): 31.8 -> 33.9 | Helped |
| Use ViRReq instead of conventional joint multi-task training | SegFormer-B5 · ADE20K validation, classes with parts | HPQ (P, %): 32.0 -> 36.9 | Helped |
| Add CLIP text embeddings to semantic segmentation | SegFormer-B0 · CPP validation, Level-1 | mIoU (%): 76.54 -> 77.35 | Helped |
| Add CLIP text embeddings to semantic segmentation | SegFormer-B5 · CPP validation, Level-1 | mIoU (%): 82.25 -> 82.40 | Helped |
| Add part-level classes to CLIP-based semantic segmentation (ablation) | SegFormer-B0 · CPP validation, Level-1 | mIoU (%): 77.35 -> 77.39 | Helped |
| Add part-level classes to CLIP-based semantic segmentation (ablation) | SegFormer-B5 · CPP validation, Level-1 | mIoU (%): 82.40 -> 82.33 | Hurt |
| Add mask sampling during training (ablation) | CondInst-R50 · CPP validation, non-probing instance segmentation | AP (%): 36.8 -> 37.8 | Helped |
| Add mask sampling during training (ablation) | CondInst-R50 · CPP validation, probing instance segmentation (γ=1.0) | AP (%): 24.5 -> 33.5 | Helped |
| Use denser non-probing sample locations (stride 16 rather than stride 32) (ablation) | CondInst-R50 · CPP validation, non-probing instance segmentation without mask sampling | AP (%): 33.6 -> 36.8 | Helped |
| Reduce the retained part annotations from 100% to 15% (ablation) | ViRReq · CPP validation, semantic and instance labels retained | HPQ (%): 60.2 -> 57.3 | Hurt |
| Reduce retained semantic and part annotations from 100% to 30% (ablation) | ViRReq · CPP validation, incomplete semantic and part annotations | HPQ (%): 60.2 -> 51.7 | Hurt |
| Move probing locations farther from the instance mass center (γ=0.0 to γ=1.0) (ablation) | CondInst with CLIP and mask sampling · CPP validation, probing instance segmentation | AP (%): 39.4 -> 33.5 | Hurt |
Closest papers by meaning
- VISion On Request: Enhanced VLLM efficiency with sparse, dynamically selected, vision-language interactions (CVPR 2026 · similarity 0.88)
- Reading Images Like Texts: Sequential Image Understanding in Vision-Language Models (ICLR 2026 · similarity 0.86)
- LaViP: Language-Grounded Visual Prompting (AAAI 2024 · similarity 0.86)
- ORIC: Benchmarking Object Recognition under Contextual Incongruity in Large Vision-Language Models (CVPR 2026 · similarity 0.86)
- Modeling Collaborator: Enabling Subjective Vision Classification With Minimal Human Effort via LLM Tool-Use (CVPR 2024 · similarity 0.86)
- Self-Evolving Visual Concept Library using Vision-Language Critics (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.