Intern Lumina U2: A Multi-Codebook Diffusion Large Language Model for Omni-Visual Understanding and Image Generation

Benchmark Model Rank Results
chart-question-answering-on-charxiv-dqIntern Lumina U2 (Ascend)#7Accuracy: 85.9
image-editing-on-imgedit-dataIntern Lumina U2 (Ascend)#14Overall: 3.82
text-to-image-generation-on-dpg-benchIntern Lumina U2 (NVIDIA)#17DPG-Bench Score: 87.99
text-to-image-generation-on-oneig-enIntern Lumina U2 (Ascend)#23OneIG Overall Score: 0.494
image-editing-on-gedit-bench-enIntern Lumina U2 (NVIDIA)#29Overall: 6.54
text-to-image-generation-on-genevalIntern Lumina U2 (Ascend)#35Overall: 0.89
chart-question-answering-on-chartqaIntern Lumina U2 (NVIDIA)#371:1 Accuracy: 86.2
multimodal-reasoning-on-mathvisionIntern Lumina U2 (Ascend)#44Accuracy: 36.5
video-question-answering-on-mlvuIntern Lumina U2 (NVIDIA)#49Accuracy: 64.9
multimodal-reasoning-on-mmmu-proIntern Lumina U2 (Ascend)#53Accuracy: 38.5
video-question-answering-on-mvbenchIntern Lumina U2 (Ascend)#62Avg.: 61.8
multimodal-reasoning-on-mathvistaIntern Lumina U2 (NVIDIA)#65Accuracy: 70.4
multimodal-reasoning-on-mmmuIntern Lumina U2 (NVIDIA)#71Accuracy: 55.5
visual-question-answering-vqa-on-realworldqaIntern Lumina U2 (Ascend)#79Accuracy: 62.9