Research map / Detection and segmentation / Open-set and few-shot detection
Visual Concept Reasoning Networks
Taesup Kim, Sungwoong Kim, Yoshua Bengio
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 VCR to ResNeXt-50 | ResNeXt-50 · ImageNet validation | Top-1 error (%): 21.10 -> 19.97 | Improved |
| Add VCR to ResNeXt-101 | ResNeXt-101 · ImageNet validation | Top-1 error (%): 19.82 -> 18.84 | Improved |
| Add VCR to ResNeXt-50 backbone | Mask R-CNN with FPN · COCO 2017 validation | bbox AP: 40.70 -> 41.81 | Improved |
| Add VCR to ResNeXt-50 backbone | Mask R-CNN with FPN · COCO 2017 validation | mask AP: 36.75 -> 37.71 | Improved |
| Use pixel-level feature modulation instead of channel-level modulation (ablation) | VCR on COCO 2017 · COCO 2017 validation | bbox AP: 41.81 -> 42.02 | Improved |
| Use pooling sampler instead of dynamic attention (ablation) | VCR · ImageNet validation | Top-1 error (%): 19.97 -> 20.21 | Hurt |
| Use static-query attention instead of dynamic attention (ablation) | VCR · ImageNet validation | Top-1 error (%): 19.97 -> 20.18 | Hurt |
| Add concept interaction edges (ablation) | VCR without concept edges · ImageNet validation | Top-1 error (%): 20.23 -> 19.97 | Improved |
| Use dynamic rather than static concept edges (ablation) | VCR with static edges · ImageNet validation | Top-1 error (%): 20.02 -> 19.97 | Improved |
| Add shifting to scale-only feature modulation (ablation) | VCR · ImageNet validation | Top-1 error (%): 20.13 -> 19.97 | Improved |
| Add scaling to shift-only feature modulation (ablation) | VCR · ImageNet validation | Top-1 error (%): 20.05 -> 19.97 | Improved |
| Use BN in channel-level VCR instead of removing BN (ablation) | VCR · ImageNet validation | Top-1 error (%): 20.15 -> 19.97 | Improved |
Closest papers by meaning
- VGent: Visual Grounding via Modular Design for Disentangling Reasoning and Prediction (CVPR 2026 · similarity 0.86)
- Probabilistic Conceptual Explainers: Trustworthy Conceptual Explanations for Vision Foundation Models (ICML 2024 · similarity 0.86)
- VGR: Visual Grounded Reasoning (ICLR 2026 · similarity 0.85)
- Visual Graph Arena: Evaluating Visual Conceptualization of Vision and Multimodal Large Language Models (ICML 2025 · similarity 0.85)
- Beyond the Doors of Perception: Vision Transformers Represent Relations Between Objects (NeurIPS 2024 · similarity 0.85)
- VLM-R³: Region Recognition, Reasoning, and Refinement for Enhanced Multimodal Chain-of-Thought (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.
