EmergentTwilight/MMCoT/data/ThinkMorph/Spatial_Navigation - Storage Bucket

🌟 This repo contains part of the training dataset for model ThinkMorph-7B.

Dataset Description

We create an enriched interleaved dataset centered on four representative tasks requiring varying degrees of visual engagement and cross-modal interactions, including Jigsaw Assembly, Spatial Navigation, Visual Search and Chart Refocus.

Statistics

Dataset Usage

Data Downloading

You can download the dataset by the following command (Taking downloading Jigsaw_Assembly data as an example):

from datasets import load_dataset

dataset = load_dataset("ThinkMorph/Jigsaw_Assembly", split="train")

Data Format

The dataset is provided in jsonl format and contains the following attributes:

{
    "pid": [string] Problem ID,
    "question": [string] The question text,
    "answer": [string] The correct answer for the problem,
    "problem_image_0": [image] Image in the question; Each question in our dataset is accompanied by an image,
    "resoning_thought_0": [string] The first tetextualxt segment of reasoning traces, which typically describe the operations performed on the image,
    "reasoning_image_0": [image] The intermediate images in the reasoning traces,
    "resoning_thought_1": [string] The second textual segment of the reasoning trace, which typically answers the question by leveraging the preceding interleaved content,
    "task": [string] The task of the problem, e.g., “Jigsaw_Assembly”,
    "full_text_only_thought": [string] Text-only reasoning trajectory used to solve the problem, with no images involved during the inference process
}

Usage

For usage instructions, please visit our GitHub repository.

Citation

@misc{gu2025thinkmorphemergentpropertiesmultimodal,
      title={ThinkMorph: Emergent Properties in Multimodal Interleaved Chain-of-Thought Reasoning}, 
      author={Jiawei Gu and Yunzhuo Hao and Huichen Will Wang and Linjie Li and Michael Qizhe Shieh and Yejin Choi and Ranjay Krishna and Yu Cheng},
      year={2025},
      eprint={2510.27492},
      archivePrefix={arXiv},
      primaryClass={cs.CV},
      url={https://arxiv.org/abs/2510.27492}, 
}