Research map / Detection and segmentation / Open-set and few-shot detection
ScaleDet: A Scalable Multi-Dataset Object Detector
Yanbei Chen, Manchen Wang, Abhay Mittal, Zhenlin Xu, Paolo Favaro, Joseph Tighe, Davide Modolo
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 |
|---|---|---|---|
| Increase training data from one dataset to four datasets | ScaleDet · LVIS | mAP: 33.1 -> 36.8 | Improved |
| Train on LVIS, COCO, Objects365, and OpenImages instead of training each dataset separately | ScaleDet · LVIS, COCO, Objects365, OpenImages | mAP: 33.1, 46.8, 28.8, 69.3 -> 36.8, 47.1, 30.6, 69.4 | Improved |
| Replace ResNet50 with Swin-Base | ScaleDet · Upstream datasets (mean) | mAP: 46.0 -> 56.3 | Improved |
| Use OpenCLIP instead of CLIP text embeddings | ScaleDet-T · Upstream datasets (mean) | mAP: 50.9 -> 51.4 | Improved |
| Use OpenCLIP instead of CLIP text embeddings | ScaleDet-B · Upstream datasets (mean) | mAP: 56.3 -> 56.5 | Improved |
| Fine-tune the multi-dataset model on each evaluation dataset | ScaleDet-B · Upstream datasets (mean) | mAP: 56.5 -> 58.0 | Improved |
| Use one unified semantic label classifier rather than dataset-specific classifiers | Multi-dataset detector · COCO, Objects365, OpenImages | mAP: 45.4 -> 47.7 | Improved |
| Train on four detection datasets rather than LVIS and COCO only | ScaleDet · LVIS, COCO | mAP: 38.9 -> 41.9 | Improved |
| Use ScaleDet-B instead of GLIPv2-B for fine-tune transfer | Object detector · ODinW-13 | mAP: 69.4 -> 71.8 | Improved |
| Use ScaleDet-R instead of Detic-R | Object detector · ODinW-13 direct transfer | mAP: 29.4 -> 39.4 | Improved |
| Add soft-label assignment loss to hard-label assignment (ablation) | ScaleDet trained on L+C · LVIS, COCO, Objects365, OpenImages | mAP: 34.2 -> 34.5 | Improved |
| Add soft-label assignment loss to hard-label assignment (ablation) | ScaleDet trained on L+C+O365+OID · ODinW-13 direct transfer | mAP: 37.9 -> 39.4 | Improved |
Later papers that compared with this method
1 later accepted paper reports ScaleDet as a baseline in their results tables; 1 of them report a better number on most of the results they share with it.
- Plain-Det: A Plain Multi-Dataset Object Detector (ECCV 2024 · 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
- Simple Multi-Dataset Detection (CVPR 2022 · similarity 0.90)
- Mind the Gap: Transferring Labels to Align Object Detection Datasets (CVPR 2026 · similarity 0.89)
- Large Scale Visual Recognition through Adaptation using Joint Representation and Multiple Instance Learning (JMLR 2016 · similarity 0.88)
- Object Detection Benchmarks are Incomplete: The Role of Label Errors and Annotation Uncertainty (NeurIPS 2026 · similarity 0.87)
- Open-Vocabulary Object Detection via Language Hierarchy (NeurIPS 2024 · similarity 0.85)
- InstructDET: Diversifying Referring Object Detection with Generalized Instructions (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.
