FF++ C23: Preprocessed Face Crops for Deepfake Detection

FF++_C23_Organised_Train_Val_Test is a preprocessed version of the FaceForensics++ dataset. It provides face crops from the C23 (high quality) compression level, organized into training, validation, and test splits. The dataset is hosted on Kaggle, but specific details on the creator, license, and exact size are not provided in the input metadata.

Use Cases

• Train deepfake detection classifiers based on pre-cropped facial images.
• Benchmark model performance on a standardized train/val/test split.
• Evaluate detection algorithms on high-quality (C23) compressed video frames.

Strengths

• Data is preprocessed into face crops, which can reduce initial data preparation effort.
• The dataset includes predefined training, validation, and test splits, aiding in standardized evaluation.

Limitations

• Description metadata is limited; actual data quality requires manual inspection after download.
• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.

Provenance

Source

FaceForensics++ (FF++) benchmark dataset.

Collection Method

Preprocessed face crops derived from the original video dataset.