metadata
configs:
- config_name: default
data_files:
- split: train
path: metadata.jsonl
dataset_info:
features:
- name: file_name
dtype: string
- name: question_id
dtype: string
- name: fraud_type
dtype: string
- name: fraud_sub_types
sequence: string
- name: ground_truth
sequence: string
- name: mask_indices
sequence: int32
- name: mask_caption
dtype: string
- name: mask_path
dtype: string
- name: figure_caption
dtype: string
- name: figure_related_sentences
dtype: string
- name: inconsistent_part
dtype: string
- name: tampered_sentences
sequence: string
- name: golden_sentences
sequence: string
- name: metadata
struct:
- name: all_images_paths
sequence: string
- name: tampered
dtype: bool
- name: paper_ids
sequence: string
- name: panel_ids
sequence: string
- name: figure_ids
sequence: string
- name: scenario
dtype: string
tags:
- scientific-paper-fraud-detection
- vision-fraud-reasoning
- fraud-reasoning
- mllm-benchmark
pretty_name: THEMIS
This repository contains the data of the paper THEMIS: Towards Holistic Evaluation of MLLMs for Scientific Paper Fraud Forensics