|
|
--- |
|
|
license: apache-2.0 |
|
|
task_categories: |
|
|
- image-text-to-text |
|
|
tags: |
|
|
- multimodal |
|
|
- visual-question-answering |
|
|
- spatial-reasoning |
|
|
- reinforcement-learning |
|
|
- transit-maps |
|
|
language: |
|
|
- en |
|
|
--- |
|
|
|
|
|
# ReasonMap-Plus Dataset |
|
|
|
|
|
This repository hosts the `ReasonMap-Plus` dataset, an extended dataset introduced in the paper [RewardMap: Tackling Sparse Rewards in Fine-grained Visual Reasoning via Multi-Stage Reinforcement Learning](https://huggingface.co/papers/2510.02240). |
|
|
|
|
|
## Paper Abstract |
|
|
Fine-grained visual reasoning remains a core challenge for multimodal large language models (MLLMs). The recently introduced ReasonMap highlights this gap by showing that even advanced MLLMs struggle with spatial reasoning in structured and information-rich settings such as transit maps, a task of clear practical and scientific importance. However, standard reinforcement learning (RL) on such tasks is impeded by sparse rewards and unstable optimization. To address this, we first construct ReasonMap-Plus, an extended dataset that introduces dense reward signals through Visual Question Answering (VQA) tasks, enabling effective cold-start training of fine-grained visual understanding skills. Next, we propose RewardMap, a multi-stage RL framework designed to improve both visual understanding and reasoning capabilities of MLLMs. RewardMap incorporates two key designs. First, we introduce a difficulty-aware reward design that incorporates detail rewards, directly tackling the sparse rewards while providing richer supervision. Second, we propose a multi-stage RL scheme that bootstraps training from simple perception to complex reasoning tasks, offering a more effective cold-start strategy than conventional Supervised Fine-Tuning (SFT). Experiments on ReasonMap and ReasonMap-Plus demonstrate that each component of RewardMap contributes to consistent performance gains, while their combination yields the best results. Moreover, models trained with RewardMap achieve an average improvement of 3.47% across 6 benchmarks spanning spatial reasoning, fine-grained visual reasoning, and general tasks beyond transit maps, underscoring enhanced visual understanding and reasoning capabilities. |
|
|
|
|
|
## Dataset Overview |
|
|
`ReasonMap-Plus` addresses the core challenge of fine-grained visual reasoning for multimodal large language models (MLLMs). It extends the original `ReasonMap` dataset by introducing dense reward signals through Visual Question Answering (VQA) tasks, enabling effective cold-start training of fine-grained visual understanding skills. This dataset is crucial for the `RewardMap` framework, which aims to improve both visual understanding and reasoning capabilities of MLLMs in structured and information-rich settings like transit maps. |
|
|
|
|
|
The dataset includes `ReasonMap-Plus` for evaluation and `ReasonMap-Train` for `RewardMap` training. |
|
|
|
|
|
## Links |
|
|
- **Project Page:** [https://fscdc.github.io/RewardMap](https://fscdc.github.io/RewardMap) |
|
|
- **Code Repository:** [https://github.com/fscdc/RewardMap](https://github.com/fscdc/RewardMap) |
|
|
|
|
|
<p align="center"> |
|
|
<img src="https://github.com/fscdc/RewardMap/raw/main/assets/rewardmap.svg" width = "95%" alt="RewardMap Framework Overview" align=center /> |
|
|
</p> |
|
|
|
|
|
## Sample Usage |
|
|
|
|
|
To get started with the RewardMap project and utilize the ReasonMap-Plus dataset, follow the steps below. |
|
|
|
|
|
### 1. Install dependencies |
|
|
|
|
|
If you face any issues with the installation, please feel free to open an issue. We will try our best to help you. |
|
|
|
|
|
```bash |
|
|
pip install -r requirements.txt |
|
|
``` |
|
|
|
|
|
### 2. Download the dataset |
|
|
|
|
|
<p align="center"> |
|
|
<img src="https://github.com/fscdc/RewardMap/raw/main/assets/overview_dataset.svg" width = "95%" alt="Dataset Overview" align=center /> |
|
|
</p> |
|
|
|
|
|
You can download `ReasonMap-Plus` for evaluation and `ReasonMap-Train` for RewardMap Training from HuggingFace or by running the following command: |
|
|
|
|
|
```bash |
|
|
python utils/download_dataset.py |
|
|
``` |
|
|
|
|
|
Then, put the data under the folder `data`. |
|
|
|
|
|
### 3. Data Format Example |
|
|
|
|
|
The data will be transferred into a format like: |
|
|
|
|
|
```json |
|
|
{ |
|
|
"conversations": [ |
|
|
{ |
|
|
"from": "human", |
|
|
"value": "<image> Please solve the multiple choice problem and put your answer (one of ABCD) in one \"\\boxed{}\". According to the subway map, how many intermediate stops are there between Danube Station and lbn Battuta Station (except for this two stops)? \ |
|
|
A) 8 \ |
|
|
B) 1 \ |
|
|
C) 25 \ |
|
|
D) 12 \ |
|
|
" |
|
|
}, |
|
|
{ |
|
|
"from": "gpt", |
|
|
"value": "B" |
|
|
} |
|
|
], |
|
|
"images": [ |
|
|
"./maps/united_arab_emirates/dubai.png" |
|
|
] |
|
|
}, |
|
|
``` |
|
|
|
|
|
## Citation |
|
|
|
|
|
If you find this paper useful in your research, please consider citing our paper: |
|
|
|
|
|
```bibtex |
|
|
@article{feng2025rewardmap, |
|
|
title={RewardMap: Tackling Sparse Rewards in Fine-grained Visual Reasoning via Multi-Stage Reinforcement Learning}, |
|
|
author={Feng, Sicheng and Tuo, Kaiwen and Wang, Song and Kong, Lingdong and Zhu, Jianke and Wang, Huan}, |
|
|
journal={arXiv preprint arXiv:2510.02240}, |
|
|
year={2025} |
|
|
} |
|
|
``` |