MME-Emotion
MME-Emotion is a multimodal emotion benchmark designed for evaluating emotion understanding and reasoning under diverse settings.
Contents
Each subset consists of:
- A JSON annotation file
- A ZIP archive containing corresponding video clips
Due to privacy considerations, raw videos are distributed only in compressed form.
Subsets
- ER_Lab / ER_Wild
- FG_ER / FG_SA
- IR
- ML_ER
- Noise_ER
- SA
Additional audio-based explanations are provided in audio_clue/.
Usage
import json
from zipfile import ZipFile
with open("ER_Lab.json") as f:
data = json.load(f)
video_id = data[0]["video_id"] # ER_Lab/xxx.mp4