Open-Vocabulary Object Detection on OV-LVIS (val)
40.1APrExDet (DeCLIP)
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| ExDet (DeCLIP)Backbone=ViT-L/14, Training Data=LVIS-base2026.06 | 40.1 | 38.2 | 37 | 38 | |
| ExDet (CLIPSelf)Backbone=ViT-L/14, Training Data=LVIS-base2026.06 | 39.7 | 38.1 | 37 | 37.9 | |
| OV-DQUOBackbone=ViT-L/14, Training Data=LVIS-base2025.08 | 39.5 | 28.5 | 36 | 33.7 | |
| OV-DQUOBackbone=ViT-L/14, Training Data=LVIS-base2026.06 | 39.5 | 28.5 | 36 | 33.7 | |
| DVtor (DeCLIP)Backbone=ViT-L/14, Training Data=LVIS-base2025.08 | 39 | 37.6 | 35.9 | 37.1 | |
| DVtor (DeCLIP)Backbone=ViT-L/14, Training Data=LVIS-base2026.06 | 39 | 37.6 | 35.9 | 37.1 | |
| F-ViT (DeCLIP)Backbone=ViT-L/14, Training Data=LVIS-base2025.08 | 37.2 | 36.5 | 35.2 | 36 | |
| F-ViT (DeCLIP)Backbone=ViT-L/14, Training Data=LVIS-base2026.06 | 37.2 | 36.5 | 35.2 | 36 | |
| DVtor (CLIPSelf)Backbone=ViT-L/14, Training Data=LVIS-base2025.08 | 36.4 | 36.9 | 35.8 | 36.3 | |
| DVtor (CLIPSelf)Backbone=ViT-L/14, Training Data=LVIS-base2026.06 | 36.4 | 36.9 | 35.8 | 36.3 | |
| ExDet (CLIP)Backbone=RN50x16, Training Data=LVIS-base2026.06 | 35 | 33.5 | 35.6 | 34.7 | |
| F-ViT (CLIPSelf)Backbone=ViT-L/14, Training Data=LVIS-base2025.08 | 34.9 | 35.6 | 34.6 | 35.1 | |
| F-ViT (CLIPSelf)Backbone=ViT-L/14, Training Data=LVIS-base2026.06 | 34.9 | 35.6 | 34.6 | 35.1 | |
| DVtor (CLIP)Backbone=RN50x16, Training Data=LVIS-base2025.08 | 33.1 | 33 | 34.1 | 33.5 | |
| DVtor (CLIP)Backbone=RN50x16, Training Data=LVIS-base2026.06 | 33.1 | 33 | 34.1 | 33.5 | |
| OWL-ViTBackbone=ViT-L/14, Training Data=O365 + VG2025.08 | 31.2 | — | — | 34.6 | |
| OWL-ViTBackbone=ViT-L/14, Training Data=O365 + VG2026.06 | 31.2 | — | — | 34.6 | |
| ExDet (DeCLIP)Backbone=ViT-B/16, Training Data=LVIS-base2026.06 | 30.5 | 31.6 | 27.1 | 29.4 | |
| F-VLM (CLIP)Backbone=RN50x16, Training Data=LVIS-base2025.08 | 30.4 | — | — | 32.1 | |
| F-VLM (CLIP)Backbone=RN50x16, Training Data=LVIS-base2026.06 | 30.4 | — | — | 32.1 | |
| ExDet (CLIPSelf)Backbone=ViT-B/16, Training Data=LVIS-base2026.06 | 30.2 | 31.1 | 26.3 | 28.8 | |
| OV-DQUOBackbone=ViT-B/16, Training Data=LVIS-base2025.08 | 29.4 | 23.8 | 27.7 | 26.5 | |
| OV-DQUOBackbone=ViT-B/16, Training Data=LVIS-base2026.06 | 29.4 | 23.8 | 27.7 | 26.5 | |
| YOLOEBackbone=YOLOv11-L*, Training Data=O365 + GoldG2025.08 | 29.1 | 36.5 | 35 | 35.2 | |
| YOLOEBackbone=YOLOv11-L*, Training Data=O365 + GoldG2026.06 | 29.1 | 36.5 | 35 | 35.2 | |
| DVtor (DeCLIP)Backbone=ViT-B/16, Training Data=LVIS-base2025.08 | 28.1 | 31 | 23.7 | 27.3 | |
| DVtor (DeCLIP)Backbone=ViT-B/16, Training Data=LVIS-base2026.06 | 28.1 | 31 | 23.7 | 27.3 | |
| MM-OVODBackbone=RN50* Agg, Training Data=LVIS-base + IN-L2025.08 | 27.3 | — | — | 33.1 | |
| MM-OVODBackbone=RN50* (Agg), Training Data=LVIS-base + IN-L2026.06 | 27.3 | — | — | 33.1 | |
| F-ViT (DeCLIP)Backbone=ViT-B/16, Training Data=LVIS-base2025.08 | 26.8 | 29.8 | 22.4 | 26 | |
| F-ViT (DeCLIP)Backbone=ViT-B/16, Training Data=LVIS-base2026.06 | 26.8 | 29.8 | 22.4 | 26 | |
| MM-OVODBackbone=RN50* Avg, Training Data=LVIS-base + IN-L2025.08 | 26.5 | — | — | 32.8 | |
| DVtor (CLIPSelf)Backbone=ViT-B/16, Training Data=LVIS-base2025.08 | 26.3 | 30.4 | 23.2 | 26.6 | |
| DVtor (CLIPSelf)Backbone=ViT-B/16, Training Data=LVIS-base2026.06 | 26.3 | 30.4 | 23.2 | 26.6 | |
| F-ViT (CLIPSelf)Backbone=ViT-B/16, Training Data=LVIS-base2025.08 | 25.3 | 29.1 | 21.8 | 25.2 | |
| F-ViT (CLIPSelf)Backbone=ViT-B/16, Training Data=LVIS-base2026.06 | 25.3 | 29.1 | 21.8 | 25.2 | |
| YOLO-WorldBackbone=YOLOv8-L*, Training Data=O365 + GoldG2025.08 | 22.9 | 35.4 | 24.9 | 28.7 | |
| YOLO-WorldBackbone=YOLOv8-L*, Training Data=O365 + GoldG2026.06 | 22.9 | 35.4 | 24.9 | 28.7 | |
| DK-DETRBackbone=RN50, Training Data=LVIS-all2025.08 | 22.2 | 40.2 | 32 | 33.5 | |
| DK-DETRBackbone=RN50, Training Data=LVIS-all2026.06 | 22.2 | 40.2 | 32 | 33.5 | |
| RegionCLIPBackbone=RN50x4*, Training Data=CC3M2025.08 | 22 | 36.9 | 32.1 | 32.3 | |
| RegionCLIPBackbone=RN50x4*, Training Data=CC3M2026.06 | 22 | 36.9 | 32.1 | 32.3 | |
| RKDWTFBackbone=RN50* WTF8x, Training Data=LVIS-base + IN-L2025.08 | 21.1 | 29.1 | 25 | 25.9 | |
| RKDWTFBackbone=RN50* WTF8x, Training Data=LVIS-base + IN-L2026.06 | 21.1 | 29.1 | 25 | 25.9 | |
| MM-OVODBackbone=RN50* Avg, Training Data=LVIS-base2025.08 | 20.7 | — | — | 30.5 | |
| OWL-ViTBackbone=ViT-B/16, Training Data=O365 + VG2025.08 | 20.6 | — | — | 27.2 | |
| OWL-ViTBackbone=ViT-B/16, Training Data=O365 + VG2026.06 | 20.6 | — | — | 27.2 | |
| MM-OVODBackbone=RN50* Agg, Training Data=LVIS-base2025.08 | 19.3 | — | — | 30.6 | |
| MM-OVODBackbone=RN50* (Agg), Training Data=LVIS-base2026.06 | 19.3 | — | — | 30.6 | |
| RKDWTFBackbone=RN50* RKDPIS, Training Data=LVIS-base + IN-L2025.08 | 17.3 | 25.5 | 20.9 | 22.1 | |
| RKDWTFBackbone=RN50* RKDPIS, Training Data=LVIS-base + IN-L2026.06 | 17.3 | 25.5 | 20.9 | 22.1 | |
| RegionCLIPBackbone=RN50*, Training Data=CC3M2025.08 | 17.1 | 34 | 27.4 | 28.2 | |
| RKDWTFBackbone=RN50* WTF, Training Data=LVIS-base + IN-L2025.08 | 17.1 | 26.7 | 21.4 | 22.8 | |
| RegionCLIPBackbone=RN50*, Training Data=CC3M2026.06 | 17.1 | 34 | 27.4 | 28.2 | |
| RKDWTFBackbone=RN50* WTF, Training Data=LVIS-base + IN-L2026.06 | 17.1 | 26.7 | 21.4 | 22.8 | |
| RKDWTFBackbone=RN50* Base, Training Data=LVIS-base + IN-L2025.08 | 12.2 | 26.4 | 19.4 | 20.9 | |
| RKDWTFBackbone=RN50* Base, Training Data=LVIS-base + IN-L2026.06 | 12.2 | 26.4 | 19.4 | 20.9 |