Research map / Vision-language and multimodal / Human and embodied modeling
Towards In-context Scene Understanding
Ivana Balazevic, David Steiner, Nikhil Parthasarathy, Relja Arandjelovic, Olivier J Henaff
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 memory contextualization (cont.) to MoCLR (ablation) | MoCLR · PASCAL VOC (NN retrieval) | mIoU: 38.6 -> 55.6 | Improved |
| add memory contextualization (cont.) to MoCLR (ablation) | MoCLR · ADE20K (NN retrieval) | mIoU: 4.9 -> 15.3 | Improved |
| replace mean pooling with [CLS] pooling (ablation) | MoCLR · PASCAL VOC (NN retrieval) | mIoU: 38.6 -> 64.5 | Improved |
| add contextualization on top of [CLS] pooling (ablation) | MoCLR + [CLS] · PASCAL VOC (NN retrieval) | mIoU: 64.5 -> 65.6 | Improved |
| use QK attention pooling plus contextualization instead of [CLS] (ablation) | MoCLR + cont. · PASCAL VOC (NN retrieval) | mIoU: 65.6 -> 68.7 | Improved |
| add a value head (QKV att.) to QK attention pooling and contextualization (ablation) | Hummingbird · PASCAL VOC (NN retrieval) | mIoU: 68.7 -> 70.5 | Improved |
| remove contextualization from QKV attention pooling (ablation) | Hummingbird · PASCAL VOC (NN retrieval) | mIoU: 70.5 -> 68.0 | Hurt |
| add retrieval-based supervision (Hummingbird++) | Hummingbird ViT-B IN1K · PASCAL VOC (NN retrieval) | mIoU: 70.5 -> 72.1 | Improved |
| add retrieval-based supervision (Hummingbird++) | Hummingbird ViT-B IN1K · NYUv2 depth (NN retrieval) | RMSE: .718 -> .738 | Hurt |
| scale pretraining data from IN1K to IN22K | Hummingbird ViT-B · PASCAL VOC (NN retrieval) | mIoU: 70.5 -> 73.5 | Improved |
| scale pretraining data from IN1K to IN22K | MAE ViT-B · PASCAL VOC (NN retrieval) | mIoU: 6.6 -> 9.8 | No clear effect |
| scale encoder from ViT-B to ViT-L | Hummingbird IN22K · PASCAL VOC (NN retrieval) | mIoU: 73.5 -> 76.9 | Improved |
Closest papers by meaning
- LSceneLLM: Enhancing Large 3D Scene Understanding Using Adaptive Visual Preferences (CVPR 2025 · similarity 0.86)
- Aggregate-and-Adapt Natural Language Prompts for Downstream Generalization of CLIP (NeurIPS 2024 · similarity 0.86)
- Hummingbird: High Fidelity Image Generation via Multimodal Context Alignment (ICLR 2025 · similarity 0.86)
- What's in the Image? A Deep-Dive into the Vision of Vision Language Models (CVPR 2025 · similarity 0.86)
- Crab: A Unified Audio-Visual Scene Understanding Model with Explicit Cooperation (CVPR 2025 · similarity 0.86)
- CoPL: Contextual Prompt Learning for Vision-Language Understanding (AAAI 2024 · 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.
