Research map / Vision-language and multimodal / Human and embodied modeling
Webly Supervised Concept Expansion for General Purpose Vision Models
Amita Kamath, Christopher Clark, Tanmay Gupta, Eric Kolve, Derek Hoiem, Aniruddha Kembhavi
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 WEB10K web-search data to multi-task training | GPV-2 · DCE (captioning, out-of-domain) | CIDEr-D: 65.4 -> 72.5 | Improved |
| add WEB10K web-search data to multi-task training | GPV-2 · DCE (classification) | accuracy: 36.3 -> 45.4 | Improved |
| add WEB10K web-search data to multi-task training | GPV-2 · DCE (localization) | mAP: 74.8 -> 75.9 | Improved |
| add WEB10K web-search data to multi-task training | VL-T5 · DCE (captioning, out-of-domain) | CIDEr-D: 31.6 -> 45.0 | Improved |
| add WEB10K web-search data to multi-task training | GPV-2 · COCO (VQA test-dev) | VQA accuracy: 71.1 -> 71.4 | No clear effect |
| add WEB10K web data (closed world, COCO-SCE DETR features) | GPV-2 · DCE (captioning, out-of-domain) | CIDEr-D: 21.1 -> 33.4 | Improved |
| add classifier re-calibration (ablation) | GPV-2 · DCE (classification) | accuracy: 34.0 -> 45.8 | Improved |
| add Conceptual Captions pre-training (ablation) | GPV-2 · DCE (captioning, out-of-domain) | CIDEr-D: -13.9 | Improved |
| use VinVL features instead of DETR features (ablation) | GPV-2 · DCE (captioning, out-of-domain) | CIDEr-D: -41.7 | Improved |
| use WEB10K in multi-task training rather than as a second pre-training round (ablation) | GPV-2 · DCE (captioning, out-of-domain) | CIDEr-D: -8.2 | Improved |
| add language-based localization (LBL) (ablation) | GPV-2 · DCE (captioning, out-of-domain) | CIDEr-D: -2.4 | Improved |
| add WEB10K web data to zero-shot attribute recognition | GPV-2 · VAW | mAP: 53.2 -> 52.4 | No clear effect |
Closest papers by meaning
- Data-efficient Large Vision Models through Sequential Autoregression (ICML 2024 · similarity 0.86)
- Concept-Aware Batch Sampling Improves Language-Image Pretraining (CVPR 2026 · similarity 0.86)
- Explanatory Instructions: Towards Unified Vision Tasks Understanding and Zero-shot Generalization (ICML 2025 · similarity 0.85)
- Bridging the gap to real-world language-grounded visual concept learning (NeurIPS 2025 · similarity 0.85)
- Concept-Guided Prompt Learning for Generalization in Vision-Language Models (AAAI 2024 · similarity 0.85)
- Unleashing Large-Scale Video Generative Pre-training for Visual Robot Manipulation (ICLR 2024 · 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.
