Research map / Detection and segmentation / Open-set and few-shot detection
LocTex: Learning Data-Efficient Visual Representations From Localized Textual Supervision
Zhijian Liu, Simon Stent, Jie Li, John Gideon, Song Han
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 mouse-trace localization loss L_L (ablation) | LocTex · VOC07 linear classification | mAP: -0.9 mAP when removed | Improved |
| add mouse-trace localization loss L_L (ablation) | LocTex · COCO 10% instance segmentation | AP bbox: -0.7 when removed | Improved |
| add mouse-trace localization loss L_L (ablation) | LocTex · COCO 20% instance segmentation | AP mask: -0.5 when removed | Improved |
| add contrastive loss L_C (vs predictive captioning, VirTex-like) (ablation) | LocTex · VOC07 linear classification | mAP: -4.2 when both removed vs -0.9 without L_L | Improved |
| add contrastive loss L_C (vs predictive captioning, VirTex-like) (ablation) | LocTex · COCO 10% instance segmentation | AP bbox: -1.3 when both removed vs -0.7 without L_L | Improved |
| add 2x resolution supervision to 1x (ablation) | LocTex · VOC07 linear classification | mAP: 87.7 -> 88.4 | Improved |
| add 4x resolution supervision to 1x,2x (ablation) | LocTex · VOC07 linear classification | mAP: 88.4 -> 88.2 | Hurt |
| use oracle mouse traces instead of real ones (ablation) | LocTex · VOC07 linear classification | mAP: 88.4 -> 90.8 | Improved |
| use oracle mouse traces instead of real ones (ablation) | LocTex · COCO 10% instance segmentation | AP bbox: 25.0 -> 26.1 | Improved |
| replace VirTex (1 caption) with LocTex localized caption | VirTex · VOC07 linear classification | mAP: 84.2 -> 88.4 | Improved |
| replace VirTex with LocTex at 118K images | VirTex · COCO 10% instance segmentation | AP bbox: 23.7 -> 25.0 | Improved |
| replace VirTex with LocTex on VOC detection | VirTex · PASCAL VOC detection | AP bbox: 51.7 -> 53.9 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- VirTex (118K) — VirTex: Learning Visual Representations From Textual Annotations (CVPR 2021) · better on 4 of 4 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- LocCa: Visual Pretraining with Location-aware Captioners (NeurIPS 2024 · similarity 0.86)
- Contrastive Localized Language-Image Pre-Training (ICML 2025 · similarity 0.86)
- Aligning Forest and Trees in Images & Long Captions for Visually Grounded Understanding (NeurIPS 2026 · similarity 0.86)
- TIPS: Text-Image Pretraining with Spatial awareness (ICLR 2025 · similarity 0.85)
- Enhancing Vision-Language Pre-training with Rich Supervisions (CVPR 2024 · similarity 0.85)
- Focus, Don't Prune: Identifying Instruction-Relevant Regions for Information-Rich Image Understanding (CVPR 2026 · 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.
