psychemme/PsycheMME-logs

PsycheMME Evaluation Records

Question-level scoring records and tables for the PsycheMME paper: 19 model entries × 1,000 questions = 19,000 records. The entries cover 17 frozen models and two adapted checkpoints.

Benchmark · Code · Project

Download

pip install huggingface_hub datasets
hf download psychemme/PsycheMME-logs --repo-type dataset \
  --revision v1.0.2 --local-dir ./PsycheMME-logs

from datasets import load_dataset
scores = load_dataset("psychemme/PsycheMME-logs", "scores", revision="v1.0.2", split="test")

Files

File	Contents
results/scoring_inputs.jsonl	19,000 records of answer credit, rationale coverage, support, logic and status
results/per_question_scores.jsonl	19,000 question-level Raw and Reasoning-Verified scores
results/paper_table_2.json	190 main-table values for 19 entries
results/leaderboard.csv	Overall model results
results/per_capability_scores.csv	Results by capability
dataset/metadata_only.jsonl	1,000 question IDs, categories and episode IDs used by replay
metadata/result_ids.jsonl	Mapping from compact result IDs to the original benchmark question and episode IDs

Use original_question_id and original_episode_id in the mapping to join these records with the benchmark. The benchmark's question and episode IDs are preserved.

Recompute Paper Scores

Use Python 3.11 with the evaluation package installed from GitHub:

psychemme replay --root ./PsycheMME-logs --output ./runs/paper_replay.json

The command recomputes all 19,000 question-level scores from the saved answer credit and reasoning factors, compares them with the saved scores, then checks all 190 displayed table values. It runs on CPU without API keys or videos. Expected result: PASS.

To evaluate your own model, follow the evaluation quick start.

License and Citation

Original numerical results and index metadata are released under CC BY 4.0. Please cite PsycheMME when using them:

@misc{hu2026psychemme,
  title={PsycheMME: A Psychology-Informed Benchmark for Social Understanding and Interaction in Videos},
  author={Heran Hu and Yuhang Dong and Zhucun Xue and Kaiwei Chen and Yong Liu and Jiangning Zhang},
  year={2026},
  eprint={2610.28625},
  archivePrefix={arXiv},
  primaryClass={cs.CV},
  url={https://arxiv.org/abs/2610.28625}
}