Datasetpublic

kir0504/anonymous-clear-drive

CLEAR-Drive Review Sample This repository provides an anonymized review-only sample of CLEAR-Drive, an autonomous-driving causal reasoning dataset introduced in the submitted paper. CLEAR-Drive is designed to improve the logical judgment ability of driving-oriented vision-language models. It constructs positive--negative sample pairs from Chain-of-Causality (CoC) reasoning, where the positive sample represents a causally valid reasoning trace and the negative sample is produced… See the full description on the dataset page: https://huggingface.co/datasets/kir0504/anonymous-clear-drive.

sourceHugging Facecc-by-nc-4.0updated 5mo agoView on Hugging Face

0likes39downloads

Dataset Card

CLEAR-Drive Review Sample

This repository provides an anonymized review-only sample of CLEAR-Drive, an autonomous-driving causal reasoning dataset introduced in the submitted paper.

CLEAR-Drive is designed to improve the logical judgment ability of driving-oriented vision-language models. It constructs positive--negative sample pairs from Chain-of-Causality (CoC) reasoning, where the positive sample represents a causally valid reasoning trace and the negative sample is produced by minimally perturbing the causal inference process to introduce a targeted logical error. These pairs are intended for preference-based training, such as DPO/MPO-style optimization.

This repository is provided only to help reviewers inspect the data format, sample quality, and construction protocol. It is not the full dataset release.

Repository Contents

text

.
├── README.md
├── data_image_recogdrive.zip
└── sample_mpo_recogdrive.jsonl

data_image_recogdrive.zip

This archive contains the anonymized image files corresponding to the provided ReCogDrive sample entries. The images are included only for review and format inspection.

sample_mpo_recogdrive.jsonl

This file contains representative positive--negative sample pairs derived from ReCogDrive-style autonomous-driving scenes. Each line is a JSON object describing one training example.

The core fields include:

—image: the image paths.
—question: the question given to the model.
—chosen: the preferred response, corresponding to the positive CoC reasoning sample.
—rejected: the dispreferred response, corresponding to the negative sample with a targeted causal-logic error.
—answer_gt: the ground truth answer.