Research map / Detection and segmentation / Open-set and few-shot detection
X-DETR: A Versatile Architecture for Instance-Wise Vision-Language Tasks
Zhaowei Cai, Gukyeong Kwon, Avinash Ravichandran, Erhan Bas, Zhuowen Tu, Rahul Bhotika, Stefano Soatto
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 |
|---|---|---|---|
| split merged paragraph queries into independent queries (mixed*) (ablation) | X-DETR (ResNet-50) · LVIS, no finetuning | AP: 11.5 -> 13.5 | Improved |
| split merged paragraph queries into independent queries (mixed*) (ablation) | X-DETR (ResNet-50) · RefCOCO MMIS-style ablation R@1, no finetuning | R@1: 63.9 -> 75.4 | Improved |
| add COCO boxes without category (mixed*+boxes) (ablation) | X-DETR (ResNet-50) · LVIS, no finetuning | AP: 13.5 -> 14.7 | Improved |
| add COCO boxes without category (mixed*+boxes) (ablation) | X-DETR (ResNet-50) · RefCOCO, no finetuning | R@1: 75.4 -> 78.0 | Improved |
| add CC image-caption data (ablation) | X-DETR (ResNet-50) · LVIS, no finetuning | AP: 14.7 -> 15.9 | Improved |
| add CC image-caption data (ablation) | X-DETR (ResNet-50) · RefCOCO, no finetuning | R@1: 78.0 -> 76.9 | Hurt |
| add LocNar pseudo-labeled data and OpenImages boxes (ablation) | X-DETR (ResNet-50) · Flickr30k phrase grounding, no finetuning | R@1: 80.1 -> 81.3 | Improved |
| add LocNar pseudo-labeled data and OpenImages boxes (ablation) | X-DETR (ResNet-50) · MMIS on RefCOCOg, no finetuning | R@5: 22.2 -> 21.0 | Hurt |
| add LocNar pseudo-labeled data and OpenImages boxes (ablation) | X-DETR (ResNet-50) · LVIS, finetuned on 10% | AP: 23.3 -> 25.6 | Improved |
| replace joint-modality transformer alignment (MDETR) with dot-product alignment | X-DETR · LVIS 10% finetuning | AP: 24.2 -> 29.5 | Improved |
| replace joint-modality transformer alignment (MDETR) with dot-product alignment | X-DETR · RefCOCO val MMIS, no finetuning | R@10: 2.5 -> 30.8 | Improved |
| add context region (expanded box) to R-CLIP (ablation) | R-CLIP · LVIS OVOD, 0% finetuning | AP: 12.7 -> 13.7 | Improved |
Closest papers by meaning
- X2-VLM: All-in-One Pre-Trained Model for Vision-Language Tasks (TPAMI 2023 · similarity 0.89)
- GCD: Advancing Vision-Language Models for Incremental Object Detection via Global Alignment and Correspondence Distillation (AAAI 2025 · similarity 0.87)
- Open-Det: An Efficient Learning Framework for Open-Ended Detection (ICML 2025 · similarity 0.87)
- DINOv2 Meets Text: A Unified Framework for Image- and Pixel-Level Vision-Language Alignment (CVPR 2025 · similarity 0.86)
- Unveiling Encoder-Free Vision-Language Models (NeurIPS 2024 · similarity 0.86)
- Cyclic Contrastive Knowledge Transfer for Open-Vocabulary Object Detection (ICLR 2025 · 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.
