Veritas

KO | EN

gitlite — search

#react #animation #cloud-native #css #rag #devops #awesome-list #automation #proxy #graphql

(★ 122)

[ICLR 2026 Oral] Veritas: Generalizable Deepfake Detection via Pattern-Aware Reasoning.

파일 탐색기

ZIP 다운로드

- banner.png
- discord_qr.jpg
- wechat.png
- - .gitignore
  - auc_fromaug.py
  - auc_table1_fromrecord.py
  - auc_table2_fromrecord.py
  - curve_draw.py
  - explore_pretrained.py
  - frequency.py
  - heatmap_tab2.py
  - model_archi.py
  - number_frames.py
  - plot_curve.py
  - tsne.py
- - readme.md
- - archi.png
- - config.yaml
  - dataset2lmdb_test.py
  - preprocess.py
  - rearrange.py
- - test.py
  - train.py
- .DS_Store
- .gitignore
- install.sh
- README.md
- - dpo_data.png
  - grpo.png
  - grpo_clevr_count.png
  - grpo_code.png
  - grpo_countdown.png
  - grpo_countdown_1.png
  - grpo_geoqa.png
  - grpo_openr1_multimodal.png
  - image-20240201122323540.png
  - image-20240201122441874.png
  - image-20240201122625473.png
  - image-20240201122725477.png
  - image-20240201131811038.png
  - image-20240201132124061.png
  - image-20240201132139698.png
  - image-20240201132308260.png
  - image-20240201132421298.png
  - image-20240201132454465.png
  - kto_data.png
  - web-ui-en.jpg
  - web-ui.jpg
- - .readthedocs.yaml
  - conf.py
  - index.rst
- - .readthedocs.yaml
  - conf.py
  - index.rst
- .DS_Store
- make.bat
- Makefile
- README.md
- - llm.sh
  - mllm.sh
- - dataset.py
  - infer.sh
  - model.py
  - sft.sh
- - merge_lora.sh
  - ollama.sh
  - push_to_hub.sh
- - cli_demo.sh
  - demo.py
  - demo_agent.py
  - demo_bert.py
  - demo_grounding.py
  - demo_hf.py
  - demo_lora.py
  - demo_mllm.py
  - demo_reward_model.py
- - infer.sh
  - lora_sft.sh
- .DS_Store
- README.md
- dependency_links.txt
- entry_points.txt
- not-zip-safe
- PKG-INFO
- requires.txt
- SOURCES.txt
- top_level.txt
- docs.txt
- eval.txt
- framework.txt
- install_all.sh
- seq_parallel.txt
- swanlab.txt
- tests.txt
- - exp.py
  - exp_utils.py
  - generate_report.py
- - plot_loss.py
  - run_dataset_info.py
  - run_model_info.py
  - run_template.py
- .DS_Store
- - deploy_model.sh
  - deploy_reward_model.sh
- - cal_metrics.py
  - infer_hydrafake.sh
  - infer_hydrafake_vllm.sh
  - infer_pt.py
  - infer_vllm.py
  - infer_vllm_single.py
- - train_grpo.sh
  - train_mipo.sh
  - train_pgrpo.sh
  - train_sft.sh
- .DS_Store
- .DS_Store
- data.png
- data_overview.png
- method.png
- reason_compare.png
- - __init__.py
  - app.py
  - deploy.py
  - eval.py
  - export.py
  - infer.py
  - main.py
  - merge_lora.py
  - pt.py
  - rlhf.py
  - sample.py
  - sft.py
  - web_ui.py
- - __init__.py
  - constant.py
  - hub.py
- - __init__.py
  - base.py
  - data_loader.py
  - utils.py
- - __init__.py
  - init.py
- - __init__.py
  - callback.py
  - loss.py
  - loss_scale.py
  - metric.py
  - multi_turn.py
  - optimizer.py
  - orm.py
  - prm.py
  - tools.py
  - tuner.py
- - __init__.py
  - arguments.py
  - callback.py
  - mixin.py
  - rlhf_arguments.py
  - torchacc_mixin.py
  - trainer_factory.py
  - trainers.py
  - utils.py
  - xtuner.py
- - __init__.py
  - adapter.py
  - base.py
  - llamapro.py
  - lora.py
  - lora_layers.py
  - mapping.py
  - neftune.py
  - part.py
  - peft.py
  - prompt.py
  - reft.py
  - restuning.py
  - restuning_components.py
  - side.py
  - utils.py
- - __init__.py
  - app.py
  - base.py
- - __init__.py
  - constants.py
  - env.py
  - import_utils.py
  - io_utils.py
  - logger.py
  - np_utils.py
  - tb_utils.py
  - torch_utils.py
  - torchacc_utils.py
  - utils.py
- .DS_Store
- __init__.py
- version.py
- - test_app.py
- - test_dataset.py
  - test_logprobs.py
- - test_eval.py
- - test_quant.py
- - test_arch.py
  - test_dataset.py
  - test_model.py
  - test_stream.py
  - test_template.py
- - __init__.py
  - test_check_model.py
- - test_agent.py
  - test_infer.py
  - test_logprobs.py
  - test_main.py
  - test_max_memory.py
  - test_mllm.py
- - __init__.py
  - load_model.py
  - load_template.py
  - test_custom.py
  - test_dataset.py
  - test_ollama_export.py
  - test_run.py
  - test_run3.py
  - test_template.py
  - test_utils.py
- - test_export.py
  - test_model.py
  - test_save.py
  - test_train.py
- - test_flash_attn.py
  - test_llm.py
  - test_mllm.py
- - test_client.py
- - test_cls.py
  - test_lmdeploy_vlm.py
  - test_padding_side.py
  - test_rlhf_loss.py
  - test_vllm_vlm.py
- - test_cls.py
  - test_freeze.py
  - test_grounding.py
  - test_grpo.py
  - test_kto.py
  - test_liger.py
  - test_multilabel.py
  - test_packing.py
  - test_ppo.py
  - test_pt.py
  - test_rlhf.py
  - test_sample.py
  - test_sft.py
  - test_train_eval.py
- - __init__.py
  - test_extra_state_dict.py
  - test_merged_linear.py
  - test_neft.py
  - test_peft.py
  - test_scetuning.py
  - test_swift_base.py
  - test_swift_device_map.py
  - test_swift_restuning.py
- - __init__.py
  - test_io_utils.py
  - test_torch_utils.py
- .DS_Store
- __init__.py
- model_tag.py
- run.py
- run_config.yaml
- test_utils.py
LICENSE
Makefile
MANIFEST.in
README.md
requirements.txt
setup.cfg
setup.py

# CDN으로 사용하기

jsDelivr

jsDelivr는 공개 GitHub 리포지토리를 별도 설정 없이 CDN으로 즉시 서빙합니다. 버전과 파일을 고르면 웹페이지에 바로 붙일 수 있는 링크와 예시 코드가 만들어집니다.

버전

파일

링크

예시

// repository documentation

Was this content helpful?

★ 0 (0 ratings)

[ 📄[**Paper**](https://arxiv.org/pdf/2508.21048) | 🔗[**Data**](https://www.modelscope.cn/datasets/EricTanh/HydraFake) | 🚀[**Model**](https://www.modelscope.cn/models/EricTanh/Veritas) ]

## News

- 🔥 `2026.7.30` We release our new work 🚀🚀🚀[**Veritas++**](https://github.com/EricTan7/VeritasPP) for **AI-generated image detection**🔥🔥🔥! This is an extended work of Veritas, which targets at general image detection. Code and Model are released [ [**Paper**](https://arxiv.org/abs/2607.27113) | [**Model**](https://www.modelscope.cn/models/EricTanh/VeritasPP) ]
- 🔥 `2026.2.10` We release our new work 🚀[**VideoVeritas**](https://github.com/EricTan7/VideoVeritas) for **AI-generated video detection**🔥🔥🔥! Dataset and Model are released [ [**Paper**](https://arxiv.org/pdf/2602.08828) | [**Data**](https://www.modelscope.cn/datasets/EricTanh/MintVid) | [**Model**](https://www.modelscope.cn/models/EricTanh/VideoVeritas) ]
- 🔥 `2026.2.6` The [**training data**](https://www.modelscope.cn/datasets/EricTanh/HydraFake/tree/master/jsons/train) is released. [**Veritas**](https://www.modelscope.cn/models/EricTanh/Veritas) and [**Veritas-Cold-Start**](https://www.modelscope.cn/models/EricTanh/Veritas-Cold-Start) are both released. We recommend using Veritas-Cold-Start to customize your own detector (see below for more details).
- 🔥 `2026.2.6` Veritas is selected as ICLR 2026 **Oral**.
- 🔥 `2026.1.26` Veritas has been accepted to **ICLR 2026**.
- 🔥 `2025.9.17` We release the inference code for MLLMs and small vision models.
- 🔥 `2025.9.17` We release the [**HydraFake**](https://www.modelscope.cn/datasets/EricTanh/HydraFake) dataset (train/val/test).

In this work, we introduce:

> 📍**HydraFake Dataset**: A deepfake detection dataset with rigorous training and evaluation protocol.
>
> 📍**Veritas Model**: A reasoning model achieving remarkable generalization on OOD forgeries, capable of providing transparent and human-aligned decision process.

## Installation
```bash
conda create -n veritas python=3.10
conda activate veritas

# Install the dependencies
pip install -e .
```

## 🔥 Training
Download training data [here](https://www.modelscope.cn/datasets/EricTanh/HydraFake/tree/master/jsons/train), including `sft_36k.json`, `mipo_3k.json`, `pgrpo_8k.json`.

### 1. Cold-start (SFT)
```bash
sh self_scripts/train/train_sft.sh sft sft_36k
```

### 2. Cold-start (MiPO)
```bash
sh self_scripts/train/train_mipo.sh mipo mipo_3k
```

### 3. P-GRPO

1.Deploy reward model. Download reward model [here](https://huggingface.co/CodeGoat24/UnifiedReward-qwen-3b), and replace the path in `swift/plugin/prm.py` and `self_scripts/deploy/deploy_reward_model.sh`.

> Note: the choice of reward model is **flexible**. More powerful models may lead to better performance, e.g., UnifiedReward-qwen-7B, Qwen3-VL-8B or UnifiedReward-2.0-qwen3vl-8B.
```bash
sh self_scripts/deploy/deploy_reward_model.sh
```

2.P-GRPO training.
```bash
sh self_scripts/train/train_pgrpo.sh pgrpo pgrpo_8k
```

## 🚀 Customize your own detector

We recommend using Veritas-Cold-Start + P-GRPO for further customization:
- Veritas is fine-tuned on in-domain datasets, i.e., the latest generative model is SD-XL. As mentioned in our paper, for practical usage, you can further fine-tune [Veritas-Cold-Start](https://www.modelscope.cn/models/EricTanh/Veritas-Cold-Start) on your own collected data. 
- If you adopt our P-GRPO, the only thing you should prepare is the binary labels of your data. Arrange them similar to our `pgrpo_8k.json`.
- We also encourage the development of (1) novel GRPO-style algorithm, e.g., involving grounded reward design to deliver more precise cross-modal signals, and (2) collaborative framework with small vision models, which can be exciting furture works.

## 🔎 Inference on single image
We recommend using `vLLM` for model deployment:
```bash
sh self_scripts/deploy/deploy_model.sh /path/to/your/model
```
Inference on a single image:
```bash
python self_scripts/infer/infer_vllm_single.py \
--image_path /path/to/your/image
```

## ⌛ Test your model on HydraFake Dataset
### 1. Data Preparation
Download the [HydraFake](https://docs.google.com/forms/d/e/1FAIpQLSf0uMg4thR4YcsNwpRqdWtc4K4z-txy24ileQPaUQzRIuMDYg/viewform?usp=header) dataset and the json files. Put the json files under `./datasets`. The data structure should be like:
```
hydrafake
├── test                # testing images
|   ├── AdobeFirefly
|   |   ├── 0_real
|   |   │   └── *.png
|   |   ├── 1_fake
|   |   │   └── *.png
|   |── ...
├── val                 # validation images
|   ├── real
|   |   └── *.png
|   ├── fake
|   |   └── *.png
├── train               # training images
|   ├── fake
|   |   ├── FS
|   |   |   ├── blendface
|   |   |   │   └── *.png
|   |   |   ├── ...
|   |   ├── FR
|   |   |   ├── Aniportrait
|   |   |   │   └── *.png
|   |   |   ├── ...
|   |   ├── EFG
|   |   |   ├── Dall-E1
|   |   |   │   └── *.png
|   |   |   ├── ...
├── jsons
|   ├── test
|   |   ├── id
|   |   │   └── *.json 
|   |   ├── cm
|   |   │   └── *.json 
|   |   ├── cf
|   |   │   └── *.json 
|   |   ├── cd
|   |   │   └── *.json 
|   ├── val
|   |   └── *.json 
|   ├── train
|   |   ├── fake
|   |   |   ├── FS
|   |   |   │   └── *.json 
|   |   |   ├── FR
|   |   |   │   └── *.json 
|   |   |   ├── EFG
|   |   |   │   └── *.json 
|   |   ├── real
|   |   │   └── *.json 
```
You can also put the dataset in other places, then you should change the json file path in `./swift/llm/dataset/dataset/data_utils.py` and the image path in the json files.

### 2. Test your MLLMs
#### 2.1 Inference with **ms-swift**
Run inference on HydraFake:
```bash
sh self_scripts/infer/infer_hydrafake.sh /path/to/your/model
```
Inference on a specific subset:
```bash
swift infer \
    --val_dataset cd_gpt4o \
    --model /path/to/your/model \
    --infer_backend pt \
    --max_model_len 8192 \
    --max_new_tokens 2048 \
    --dataset_num_proc 16 \
    --max_batch_size 8 \
    --metric self_acc_tags
```

#### 2.2 Inference with **vLLM**
Step1: Deploy your model:
```bash
sh self_scripts/deploy/deploy_model.sh /path/to/your/model  # models/Qwen2.5-VL-7B-Instruct
```
Step2: Run inference (put your model path in `self_scripts/infer/infer_vllm.py`):
```bash
sh self_scripts/infer/infer_hydrafake_vllm.sh 
```

### 3. Test your vision models
We provide a script based on [DeepfakeBench](https://github.com/SCLBD/DeepfakeBench).
```bash
# Effort for example
python DeepfakeBench/training/test.py \
--detector_cfg DeepfakeBench/training/config/detector/effort.yaml \   
--dataset_cfg DeepfakeBench/training/config/dataset/hydrafake.yaml \
--weights_path /path/to/your/model
```

## 🛡️ HydraFake Dataset

📍 **Overview:**

(a) We carefully collect and reimplement advanced deepfake techniques to construct our HydraFake dataset. Real images are collected from 8 datasets. Fake images are from classic datasets, high-quality public datasets and our self-constructed deepfake data. (b) We introduce a rigorous and hierarchical evaluation protocol. Training data contains abundant samples but limited forgery types. Evaluations are split into four distinct levels. (c) Illustration of the subsets in different evaluation splits. (d) The performance of prevailing detectors on our HydraFake dataset. **Most detectors shows strong generalization on Cross-Model setting but poor ability on Cross-Forgery and Cross-Domain scenarios.**

📍 **Statistics:**

HydraFake contains 52K images in total for evaluation, with 14K in-domain testing, 11K cross-model testing, 12K cross-forgery testing and 15K cross-domain testing.

## 🛰️ Method

📍 We introduce a pattern-aware reasoning framework, including three basic thinking patterns (*fast judgement*, *reasoning*, *conclusion*) and two advanced patterns (*planning* and *self-reflection*).

📍 Two-stage training pipeline:

(1) **Pattern-guided Cold-Start** (SFT + MiPO): Internalize thinking patterns and align reasoning process

(2) **Pattern-aware Exploration** (P-GRPO): Scale up effective patterns, improve reflection quality.

## Citation
If you find our work useful, please cite our paper:
```
@inproceedings{tan2025veritas,
  title={Veritas: Generalizable Deepfake Detection via Pattern-Aware Reasoning},
  author={Tan, Hao and Lan, Jun and Tan, Zichang and Liu, Ajian and Song, Chuanbiao and Shi, Senyuan and Zhu, Huijia and Wang, Weiqiang and Wan, Jun and Lei, Zhen},
  booktitle={International Conference on Learning Representations},
  year={2026}
}
```

## License
This repo is released under the [Apache 2.0 License](https://github.com/EricTan7/Veritas/blob/main/LICENSE).

## Acknowledgements

This repo benefits from [ms-swift](https://github.com/modelscope/ms-swift) and [DeepfakeBench](https://github.com/SCLBD/DeepfakeBench). Thanks for their great works!

Veritas

오너의 다른 레포

파일 탐색기

# CDN으로 사용하기

Was this content helpful?

Recent Feedback

Veritas

오너의 다른 레포

파일 탐색기

# CDN으로 사용하기

Was this content helpful?

Recent Feedback

Download README