Research map / Detection and segmentation / Open-set and few-shot detection
Class-Agnostic Object Detection with Multi-modal Transformer
Muhammad Maaz, Hanoona Rasheed, Salman Khan, Fahad Shahbaz Khan, Rao Muhammad Anwer, Ming-Hsuan Yang
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 language branch (captions) during training (ablation) | MAVL · COCO | AP50: 34.4 -> 39.3 | Improved |
| add language branch (captions) during training (ablation) | MDETR · Pascal VOC | AP50: 59.7 -> 63.9 | Improved |
| add caption-based language structure to training (ablation) | Deformable DETR / MAVL · Pascal VOC | AP50: 16.2 -> 61.6 | Improved |
| add multi-scale deformable attention (MSDA) (ablation) | MAVL · Pascal VOC | AP50: 59.9 -> 65.0 | Improved |
| use combined detections from several text queries instead of 'all objects' (ablation) | MAVL · COCO | AP50: 33.3 -> 42.0 | Improved |
| drop the 'all small objects' query from combined detections (ablation) | MAVL · Pascal VOC / KITTI | AP50: 63.7 -> 68.6 (VOC); 48.2 -> 45.8 (KITTI) | Mixed |
| use general text queries instead of category-wise queries (ablation) | MAVL · Pascal VOC | AP50: 61.7 -> 68.6 | Improved |
| use stronger backbone (E5 instead of ResNet-101) (ablation) | MDETR · Kitchen | AP50: 38.4 -> 53.3 | Improved |
| use task-specific query instead of general query (ablation) | MAVL · DUT-OMRON | AP50: 67.0 -> 75.5 | Improved |
| use task-specific query instead of general query (ablation) | MAVL · CHAMELEON | AP50: 30.2 -> 36.2 | Improved |
| replace RPN pseudo-labels with MAVL proposals for unknowns | ORE · ORE Task 1 | R50 unknown: 14.4 -> 50.1 | Improved |
| replace Selective Search proposals with MAVL proposals in pretraining | DETReg · Pascal VOC 10% | AP: 51.4 -> 58.8 | 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.
- Faster-RCNN — Faster R-CNN: Towards Real-Time Object Detection with Region Proposal Networks (TPAMI 2016) · better on 5 of 5 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
- Training Object Detectors From Scratch: An Empirical Study in the Era of Vision Transformer (CVPR 2022 · similarity 0.87)
- MViTv2: Improved Multiscale Vision Transformers for Classification and Detection (CVPR 2022 · similarity 0.87)
- Simple Open-Vocabulary Object Detection with Vision Transformers (ECCV 2022 · similarity 0.87)
- Multi-modal Queried Object Detection in the Wild (NeurIPS 2023 · similarity 0.87)
- Self-Supervised Learning of Intertwined Content and Positional Features for Object Detection (ICML 2025 · similarity 0.85)
- Roboflow100-VL: A Multi-Domain Object Detection Benchmark for Vision-Language Models (NeurIPS 2025 · 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.
