Common Sense Reasoning for Deep Fake Detection,
Yue Zhang, Ben Colman, Xiao Guo, Ali Shahriyari, and Gaurav Bharaj
ECCV 2024
Please see our DD-VQA dataset in folder DQ_FF++
. The images are cropped frames from the FF++.
- The dataset is split based on FF++.
- For each type of manipulation, we provide the corresponding question-answer paris using a JSON file with the following format.
{ "manitupateid_videoid" {
"question_id":{
"question": "Does the image/ the person's eyes/nose/mouth/eyebrows/ look fake?",
"answer":["answer1", "answer2", "..."]
}
}
}
"%The manipulate_id is the manipulation types: 0: Deepfakes, 1:Face2face, 2: FaceShift, 3: FaceSwap,5: Original, 6: NeuralTexture."