---
license: other
language:
- en
task_categories:
- image-text-to-text
tags:
- fossil
- paleontology
- visual-question-answering
- llama-factory
pretty_name: FosCap Three-Stage Training Dataset
---

# FosCap Three-Stage Training Dataset

Training data for the three-stage FosCap fossil-image understanding and captioning pipeline.

## Contents

| Stage | Annotations | Images |
| --- | --- | --- |
| Stage 1 | `foscap_stage1.json` | `foscap_stage1_images.zip` |
| Stage 2 | `foscap_stage2.json` | `foscap_stage2_images.zip` |
| Stage 3 | `foscap_stage3_new_visual_expert.json` | `foscap_stage3_images.zip` |

The annotation files use LLaMA-Factory's ShareGPT multimodal format with `messages` and `images` fields.
Image paths in each JSON file are filenames relative to the extracted image directory for that stage.

## Extraction

```bash
unzip foscap_stage1_images.zip -d stage1_images
unzip foscap_stage2_images.zip -d stage2_images
unzip foscap_stage3_images.zip -d stage3_images
```

## Related models

- [Stage 1](https://huggingface.co/LockOnN/FosCap-Qwen3.6-35B-A3B-Stage1)
- [Stage 2](https://huggingface.co/LockOnN/FosCap-Qwen3.6-35B-A3B-Stage2)
- [Stage 3](https://huggingface.co/LockOnN/FosCap-Qwen3.6-35B-A3B-Stage3)