README.md

---
pretty_name: ForgeryVCR Data
tags:
  - image-forensics
  - image-manipulation-detection
  - multimodal
---

# ForgeryVCR Data

This repository stores the preprocessed 832-size assets used by the public
ForgeryVCR training and evaluation workflow.

```text
datasets/
├── train/
│   ├── sft/   # CASIA v2 archives for expert selection and Agent SFT
│   └── rl/    # IMD2020 and FantasticReality archives for GRPO
└── test/      # public evaluation benchmark archives
```

Tampered-set JSON files use paths relative to their own dataset directory, for
example `./forged/example.png` and `./gt/example_gt.png`.

## Download

Download all archives:

```bash
hf download ForgeryVCR-Team/ForgeryVCR \
  --repo-type dataset \
  --include "datasets/train/sft/*.zip" "datasets/train/rl/*.zip" "datasets/test/*.zip" \
  --local-dir downloads/forgeryvcr
```

Download only the SFT datasets:

```bash
hf download ForgeryVCR-Team/ForgeryVCR \
  --repo-type dataset \
  --include "datasets/train/sft/*.zip" \
  --local-dir downloads/forgeryvcr
```

Download only the RL datasets:

```bash
hf download ForgeryVCR-Team/ForgeryVCR \
  --repo-type dataset \
  --include "datasets/train/rl/*.zip" \
  --local-dir downloads/forgeryvcr
```

Download only the evaluation datasets:

```bash
hf download ForgeryVCR-Team/ForgeryVCR \
  --repo-type dataset \
  --include "datasets/test/*.zip" \
  --local-dir downloads/forgeryvcr
```

Extraction commands and the complete directory contract are documented in the
[ForgeryVCR repository](https://github.com/youqiwong/ForgeryVCR).