General
AffectGPT MER-UniBench
MER-UniBench (AffectGPT, ICML 2025): per-item emotion-understanding results of the AffectGPT audio-video-text MLLM on 7 public test sets — basic emotion recognition (MER2023, MELD, IEMOCAP-four; binary emotion-wheel hit per item) and sentiment analysis (CMU-MOSI, CMU-MOSEI, CH-SIMS, CH-SIMS v2; binary polarity accuracy per item), for 7 released training checkpoints (epochs 30-60).
8,123items
1subjects
100%observed
apache-2.0license
generaldomain
textmodality
audiomodality
imagemodality
Response matrix
Fit to width. Hover for subject & item; click a cell for details.
This condition combination wasn’t evaluated — try a different attack, category, or judge.
Correct (1)Incorrect (0)Unobserved
Scale: 1 = correct · 0 = incorrect