Research map / Detection and segmentation / Weakly supervised detection and segmentation
Exploiting Unlabeled Data with Vision and Language Models for Object Detection
Shiyu Zhao, Zhixing Zhang, Samuel Schulter, Long Zhao, Vijay Kumar B G, Anastasis Stathopoulos, Manmohan Chandraker, Dimitris N. Metaxas
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 RoI head box refinement to pseudo labels (tau=0.95) (ablation) | VL-PLM · COCO-ZS | Novel AP50: 26.0 -> 31.3 | Improved |
| add RoI head refinement at tau=0.05 (ablation) | VL-PLM · COCO-ZS | Novel AP50: 14.6 -> 19.3 | Improved |
| raise pseudo-label threshold from 0.05 to 0.95 (with RoI refinement) (ablation) | VL-PLM · COCO-ZS | Novel AP50: 19.3 -> 31.3 | Improved |
| raise threshold from 0.95 to 0.99 (ablation) | VL-PLM · COCO-ZS | Novel AP50: 31.3 -> 27.2 | Hurt |
| use RPN+RoI proposals instead of selective search (ablation) | VL-PLM · COCO-ZS | AP@PL: 5.7 -> 25.3 | Improved |
| use RPN+RoI instead of RoI head alone (ablation) | VL-PLM · COCO-ZS | AP@PL: 8.8 -> 25.3 | Improved |
| use RPN+RoI instead of RPN alone (ablation) | VL-PLM · COCO-ZS | AP@PL: 19.7 -> 25.3 | Improved |
| fuse teacher pseudo labels with VL-PLM labels (1% COCO) (ablation) | VL-PLM SSOD · COCO SSOD (1% labeled) | mAP: 13.27 -> 15.35 | Improved |
| fuse teacher pseudo labels with VL-PLM labels (10% COCO) (ablation) | VL-PLM SSOD · COCO SSOD (10% labeled) | mAP: 24.20 -> 27.23 | Improved |
| use VL-PLM labels without teacher labels instead of teacher labels only (10% COCO) (ablation) | SSOD · COCO SSOD (10% labeled) | mAP: 25.98 -> 24.20 | Hurt |
| replace STAC pseudo labels with VL-PLM (1% COCO) | STAC · COCO SSOD (1% labeled) | mAP: 13.97 -> 17.71 | Improved |
| add more unlabeled images (1:20 ratio vs none) at 5% labeled (ablation) | Faster R-CNN · COCO SSOD (5% labeled) | mAP: 17.7 -> 23.7 | Improved |
Closest papers by meaning
- MarvelOVD: Marrying Object Recognition and Vision-Language Models for Robust Open-Vocabulary Object Detection (ECCV 2024 · similarity 0.92)
- Open Vocabulary Object Detection with Pseudo Bounding-Box Labels (ECCV 2022 · similarity 0.92)
- Generative Region-Language Pretraining for Open-Ended Object Detection (CVPR 2024 · similarity 0.92)
- Taming Self-Training for Open-Vocabulary Object Detection (CVPR 2024 · similarity 0.89)
- Language-conditioned Detection Transformer (CVPR 2024 · similarity 0.89)
- Open-Vocabulary Object Detection via Language Hierarchy (NeurIPS 2024 · similarity 0.89)
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.
