INDIFACE: Illuminating India's Deepfake Landscape with a Comprehensive Synthetic Dataset

• Published in: IEEE International Conference and Workshops on Automatic Face and Gesture Recognition : FG pp. 1 - 9
• Main Authors: Kuckreja, Kartik, Hoque, Ximi, Poddar, Nishit, Reddy, Shukesh, Dhall, Abhinav, Das, Abhijit
• Format: Conference Proceeding
• Language: English
• Published: IEEE 27.05.2024
• Subjects: Benchmark testing; Deepfakes; Face recognition; Gesture recognition; Government; Law enforcement; Synthetic data
• ISSN: 2770-8330
• DOI: 10.1109/FG59268.2024.10582046

Due to the recent progress in Deepfake generation, several datasets and manipulation techniques have been proposed in the recent literature with various effective face-swap and face-reenactment methods. Deepfake is an emerging threat to society and government as it can jeopardize law enforcement and cause personal loss. Investigations in the literature established that demographic variation had impacted the performance of Deepfake detection. To date, Deepfake detection has not been studied in the Indian context; hence, in this work, we proposed a Deepfake dataset INDIFACE entirely with Indian subjects. We have collected 101 original videos and used two different manipulation techniques for Deepfake generation. We provide detailed benchmarking with state-of-the-art methods on Deepfake datasets, showcasing that the existing model is insufficient to detect Deepfake detection for the Indian scenario. Hence, more attention is required to this area of research. The proposed dataset INDIFACE is publicly available at.