Video-to-command generation on Modified Something-Something Standard Object Sets V2 (test)

0.626BLEU-1Intern3.5VL-4B

Evaluation results

Method
Intern3.5VL-4B2026.06	0.626	0.427	0.339	0.251	0.614	0.521	1.422
Decoupled Object-Centric Video Understanding framework (Ours Full)2026.06	0.607	0.506	0.397	0.337	0.598	0.544	1.812
NVILA-8B2026.06	0.557	0.389	0.329	0.286	0.561	0.475	1.773
Qwen2.5-VL-3B2026.06	0.531	0.322	0.254	0.207	0.547	0.428	1.307
Intern2.5VL-4B2026.06	0.525	0.319	0.244	0.172	0.511	0.429	1.078
ChatGPT-4o-mini2026.06	0.522	0.41	0.323	0.269	0.574	0.478	1.636
Watch-and-Act2026.06	0.394	0.271	0.248	0.187	0.405	0.311	1.434
DeepseekVL7B2026.06	0.368	0.275	0.227	0.182	0.384	0.334	0.801
V2CNet2026.06	0.357	0.231	0.201	0.153	0.369	0.264	0.895
Video2Command2026.06	0.312	0.196	0.173	0.15	0.33	0.234	1.103
LLaVA-NeXT7B2026.06	0.28	0.171	0.105	0.075	0.347	0.234	0.493