List your company
ResearchFormerly WizWand

Multi-image Dialogue Understanding on MMDU

59.8AccuracyStochasT
8.42421.76235.148.438Oct 22, 2025Dec 3, 2025Jan 14, 2026Feb 25, 2026Apr 8, 2026May 20, 2026Jul 1, 2026
Updated 4mo ago

Evaluation results

MethodLinks
StochasTModel=Qwen-2.5-VL-3B, Evaluation setting=MultiT2026.0759.8
MultiTModel=Qwen-2.5-VL-3B, Evaluation setting=MultiT2026.0757.9
OriginalModel=Qwen-2.5-VL-3B, Evaluation setting=MultiT2026.0747
LLaVA-v1.5Backbone Parameters=7B2025.1026.37
KORErank=2562025.1023.18
Replay2025.1019.31
MoELoRA2025.1017.85
O-LoRA2025.1017.28
KORErank=2352025.1016.58
LoRA2025.1013.7
LwF2025.1013.68
EWC2025.1013.36
StochasTModel=LLaVA-1.5-7B, Evaluation setting=MultiT2026.0713.1
Full-FTLearning Strategy=Fine-Tuning2025.1013.03
MultiTModel=LLaVA-1.5-7B, Evaluation setting=MultiT2026.0712.68
OriginalModel=LLaVA-1.5-7B, Evaluation setting=MultiT2026.0710.81
SEFE2025.1010.55
CIA2025.1010.4