Research map / Vision-language and multimodal / Human and embodied modeling
DIP: Unsupervised Dense In-Context Post-training of Visual Representations
Sophia Sirko-Galouchenko, Spyros Gidaris, Antonin Vobecky, Andrei Bursuc, Nicolas Thome
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 |
|---|---|---|---|
| post-train DINOv2R with DIP | DINOv2R ViT-B/14 · PascalVOC (1/128 of support data) | mIoU: 60.3 -> 70.1 | Improved |
| post-train DINOv2R with DIP | DINOv2R ViT-B/14 · ADE20K (in-context, full data) | mIoU: 40.8 -> 42.6 | Improved |
| apply DIP post-training to MAE | MAE ViT-B/16 · PascalVOC (in-context) | mIoU: 13.9 -> 47.3 | Improved |
| apply DIP post-training to CLIP | CLIP ViT-B/16 · PascalVOC (in-context) | mIoU: 71.8 -> 73.8 | Improved |
| apply DIP post-training to DINOv2R ViT-L | DINOv2R ViT-L · PascalVOC (full data) | mIoU: 76.9 -> 81.1 | Improved |
| use in-context objective instead of direct dense prediction (ablation) | DIP · NYUv2 depth | RMSE: .776 -> .756 | Improved |
| use in-context objective instead of direct dense prediction (ablation) | DIP · ADE20K | mIoU: 39.9 -> 39.7 | No clear effect |
| add distractor support examples (ablation) | DIP · PascalVOC | mIoU: 78.5 -> 81.0 | Improved |
| use DINOv2R nearest-neighbor positives instead of two random crops (ablation) | DIP · PascalVOC | mIoU: 79.5 -> 81.0 | Improved |
| add DiffCut segments before K-means pseudo-labeling (ablation) | DIP · PascalVOC | mIoU: 59.1 -> 81.0 | Improved |
| use ground truth labels instead of pseudo-labels (ablation) | DIP · PascalVOC | mIoU: 81.0 -> 81.9 | No clear effect |
| vary number of pseudo-classes (300 to 2000) (ablation) | DIP · PascalVOC | mIoU: 80.8 -> 81.0 | No clear effect |
Closest papers by meaning
- Perception Encoder: The best visual embeddings are not at the output of the network (NeurIPS 2025 · similarity 0.88)
- What Matters When Repurposing Diffusion Models for General Dense Perception Tasks? (ICLR 2025 · similarity 0.87)
- Improved Baselines with Representation Autoencoders (NeurIPS 2026 · similarity 0.87)
- Win-Win: Training High-Resolution Vision Transformers from Two Windows (ICLR 2024 · similarity 0.87)
- Vision‑Language‑Vision Auto‑Encoder: Scalable Knowledge Distillation from Diffusion Models (NeurIPS 2025 · similarity 0.86)
- Beyond Encoder Accumulation: Measuring Encoder Roles in Multi-Encoder VLMs (NeurIPS 2026 · 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.
