Deepfake and Real Images Dataset
Dataset composed of manipulated (deepfake) and real facial images, preprocessed in 256x256 pixel resolution, intended for the detection of falsifications.
Approximately 190,000 256x256 pixel JPG images, real faces and deepfake
Creative Commons Attribution 4.0 International (CC BY 4.0)
Description
The dataset Deepfake and Real Images includes approximately 190,000 images of faces in 256x256 JPG format, some of which are manipulated by various deepfake techniques, and the other part corresponds to real faces. This corpus makes it possible to work on the detection of visually falsified content.
What is this dataset for?
- Develop and train deepfake detection models
- Improving facial recognition security and fraud prevention
- Study image techniques generated and manipulated in computer vision
Can it be enriched or improved?
This dataset can be enriched by finer annotations (type of manipulation, source), by metadata on the acquisition conditions or by adding new types of deepfakes.
🔎 In summary
🧠 Recommended for
- Researchers in detection AI
- Security developers
- Computer vision students
🔧 Compatible tools
- PyTorch
- TensorFlow
- OpenCV
- Detectron2
- DeepFaceLab
💡 Tip
Combine this dataset with other deepfake corpora to increase the robustness of the models.
Frequently Asked Questions
What is the standard resolution of the images in this dataset?
All images are resized to 256x256 pixels, standard for CNN models.
Does this dataset contain annotations on deepfake types?
No, annotations are limited to distinguishing between real and manipulated images.
Can this dataset be used for commercial use?
Yes, the CC BY 4.0 license allows commercial use under attribution.
