Open Images
Benchmarks
| Task Name | Dataset Name | SOTA Result | Trend | |
|---|---|---|---|---|
| Scene Graph Generation | Open Images V6 (test) | wmAPrel56.38 | 86 | |
| Label Purity | Open Images | Label Purity79.6 | 30 | |
| Object Detection | Open Images V7 | Latency (us)95.1 | 30 | |
| Multi-label classification | Open Images | mAP84.5 | 24 | |
| Multi-label Classification | Open Images (test) | mAP85 | 16 | |
| Harm Recognition | Open Images Weapon slice | F1 Score97.65 | 15 | |
| Feature Attribution Evaluation | Open Images 5000 random images (val) | AUC0.719 | 13 | |
| Object Detection | Open Images unseen classes (non-overlapping) | AR@1000 (Overall)21 | 11 | |
| Image Inpainting | Open Images V7 (test) | Coverage Radius (T1)2.13 | 10 | |
| Multi-label classification | Open Images v4 (test) | Precision (K=10)35.3 | 10 | |
| Attribution Quality Evaluation | Open Images (val) | SIC AUC0.866 | 10 | |
| Feature Reconstruction | Open Images clip_txt Original Target (test) | R^2 (variance-weighted)0.874 | 9 | |
| Multi-label classification | Open Images (val) | mAP58.11 | 9 | |
| Instance Segmentation | Open Images (test) | mAP50 (Constrained Novel)35.9 | 8 | |
| Multi-Label Recognition | Open Images GZSL | F1 Score @ K=1040.9 | 7 | |
| Multi-Label Recognition | Open Images ZSL | F1@1020.9 | 7 | |
| Lossless Image Compression | Open Images (val) | BPSP2.867 | 7 | |
| Image Expansion | Open Images (Human) | FID40.82 | 5 | |
| Image Expansion | Open Images Indoor | FID29.33 | 5 | |
| Image Expansion | Open Images Natural | FID27.43 | 5 | |
| Concept Alignment | Open Images hierarchy depth 5 | Mean Jaccard Similarity0.8018 | 5 | |
| Concept recovery probing (1D logistic probe) | Open Images 432 binary tasks (test) | CLIP Image Score0.6372 | 5 | |
| Visual Question Answering | Open Images cross-task (test) | Accuracy44.7 | 5 | |
| Visual Reasoning | Open Images (test) | Accuracy85.1 | 5 | |
| Scene Graph Detection | Open Images V6 | mR5040.7 | 5 |
Showing 25 of 37 rows