Open-vocabulary Object Detection on LVIS v1
29.3AP (rare, box)UniDetector
Evaluation results
| Method | Links | ||
|---|---|---|---|
| UniDetectorTraining protocol=LVIS based training, Data=LVIS base + IN-L2023.03 | 29.3 | 36.8 | |
| DeticTraining protocol=LVIS based training, Data=LVIS base + IN-L2023.03 | 26.7 | 36.2 | |
| PromptDetTraining protocol=LVIS based training, Data=LVIS base2023.03 | 21.8 | 27.3 | |
| DetProTraining protocol=LVIS based training, Data=LVIS base2023.03 | 20.8 | 28.4 | |
| UniDetectorTraining protocol=Unrestricted open-vocabulary training, Backbone=ResNet50, Data=O365 (10%), VG2023.03 | 20.2 | 23.4 | |
| OWL-ViTTraining protocol=Unrestricted open-vocabulary training, Backbone=ViT-B/32, Data=O365 (all), VG2023.03 | 18.9 | 22.1 | |
| RegionCLIPTraining protocol=LVIS based training, Data=LVIS base2023.03 | 17.1 | 28.2 | |
| ViLDTraining protocol=LVIS based training, Data=LVIS base2023.03 | 16.7 | 27.8 | |
| GLIPTraining protocol=Unrestricted open-vocabulary training, Backbone=Swin-T, Data=O365 (all), GlodG, ...2023.03 | 10.1 | 17.2 |