Research map / Vision-language and multimodal / CLIP and zero-shot
T-REN: Learning Text-Aligned Region Tokens Improves Dense Vision-Language Alignment and Scalability
Savya Khosla, Sethuraman T V, Aryan Chadha, Alex Schwing, Derek Hoiem
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 T-REN region tokens on top of dino.txt | DINOv3 dino.txt · ADE20K OVSS | mIoU: 24.7 -> 30.6 | Improved |
| add T-REN region tokens | DINOv3 dino.txt · Cityscapes OVSS | mIoU: 36.9 -> 52.7 | Improved |
| add T-REN region tokens | DINOv3 dino.txt · Visual Haystacks D=1K | accuracy (%): 52.1 -> 65.2 | Improved |
| add T-REN track tokens | DINOv3 dino.txt · Ego4D query localization | Recall@1: 36.8 -> 52.4 | Improved |
| add T-REN track tokens | DINOv3 dino.txt · VSPW scene parsing | mIoU: 20.7 -> 38.3 | Improved |
| add region pooling jointly with text alignment (ablation) | T-REN · ADE20K OVSS | mIoU: 25.4 -> 30.6 | Improved |
| add joint text alignment training to region pooling (ablation) | T-REN · ADE20K OVSS | mIoU: 19.5 -> 30.6 | Improved |
| add region pooling to patch-level text alignment (ablation) | T-REN · Visual Haystacks retrieval D=10 | accuracy (%): 76.1 -> 87.2 | Improved |
| add token merging within and across frames (ablation) | T-REN · VSPW scene parsing | mIoU | No clear effect |
| add temporal token merging (tau_track 0.5) (ablation) | T-REN · Ego4D query localization subset | recall tIoU=0.25 (%): 39.4 -> 57.7 | Improved |
| raise tau_track to 0.9 (ablation) | T-REN · Ego4D query localization subset | recall tIoU=0.25 (%): 57.7 -> 45.2 | Hurt |
| predict k=3 region tokens per prompt instead of one (ablation) | T-REN · COCO retrieval | recall | Improved |
Closest papers by meaning
- DINOv2 Meets Text: A Unified Framework for Image- and Pixel-Level Vision-Language Alignment (CVPR 2025 · similarity 0.90)
- Kernel-based Unsupervised Embedding Alignment for Enhanced Visual Representation in Vision-language Models (ICML 2025 · similarity 0.89)
- REN: Fast and Efficient Region Encodings from Patch-Based Image Encoders (NeurIPS 2025 · similarity 0.89)
- Locality Alignment Improves Vision-Language Models (ICLR 2025 · similarity 0.88)
- Fine-Grained Verifiers: Preference Modeling as Next-token Prediction in Vision-Language Alignment (ICLR 2025 · similarity 0.88)
- Aligning Forest and Trees in Images & Long Captions for Visually Grounded Understanding (NeurIPS 2026 · similarity 0.88)
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.
