CC3M
Benchmarks
| Task Name | Dataset Name | SOTA Result | Trend | |
|---|---|---|---|---|
| Image Generation | CC3M SDXL v1.0 (test) | FID33.49 | 27 | |
| Aesthetic Evaluation | CC3M SDXL 1.0 (test) | HPS0.2932 | 27 | |
| Object Recognition | CC3M (test) | Recall0.738 | 21 | |
| Aesthetic Evaluation | CC3M | HPS0.2757 | 15 | |
| Image Generation | CC3M (test) | FID36.96 | 15 | |
| Multimodal Understanding | CC3M IOD | Accuracy100 | 14 | |
| Malicious Prompt Detection | CC3M (IOD) | FPR0 | 14 | |
| Text-to-Image Retrieval | CC3M | Recall45.7 | 9 | |
| Image-to-Text Retrieval | CC3M | Recall47.2 | 9 | |
| Image Classification | CC3M | Accuracy46.7 | 9 | |
| Multi-Tag Selection | CC3M (test) | Precision92.5 | 9 | |
| Text-to-image generation | CC3M | FID6.06 | 7 | |
| Image-Text Retrieval | CC3M-595K-LLaVA 1 | IR Score7.6 | 6 | |
| Conditional Inversion | CC3M first 500 images | PSNR25.24 | 6 | |
| Unconditional Inversion | CC3M first 500 images | PSNR24.41 | 6 | |
| Multi-Tag Selection | CC3M | Precision0.883 | 6 | |
| Backdoor Defense (Multimodal) | CC3M 60K subset 1.0 (test) | Clean Accuracy (CA)64.03 | 5 | |
| Vision-Language Compositional Evaluation | CC3M 50,000 random subset TripletData | Text Score92.25 | 4 | |
| Text-level Semantic Segmentation | CC3M (subset) | Caption IoU65.5 | 4 | |
| Object Recognition | CC3M | Recall86.8 | 3 |
Showing 20 of 20 rows