notesum.ai

Published at April 28

DiffusionFake: Enhancing Generalization in Deepfake Detection via Guided Stable Diffusion

NeurIPS

Released Date: April 28, 2024

Authors: Ke Sun¹, Shen Chen², Taiping Yao², Hong Liu³, Xiaoshuai Sun¹, Shouhong Ding², Rongrong Ji¹

Aff.: ¹Key Laboratory of Multimedia Trusted Perception and Efficient Computing, Ministry of Education of China, Xiamen University, P.R. China; ²Youtu Lab, Tencent, P.R. China; ³Osaka University, Japan

Arxiv: https://openreview.net/pdf/36c9663bea5245faf8ad6ee0029decb78a239cce.pdf

Method	Celeb-DF		Wild Deepfake		DFDC-P		DFD		DiffSwap		Average
Method	AUC	EER	AUC	EER	AUC	EER	AUC	EER	AUC	EER	AUC	EER
Xception [7]	65.27	38.77	66.17	40.14	69.80	35.41	87.86	21.04	74.25	32.04	72.67	33.48
Face X-ray [42]	74.20	-	-	-	70.00	-	85.60	-	-	-	-	-
F3-Net* [29]	71.21	34.03	67.71	40.17	72.88	33.38	86.10	26.17	76.89	30.83	74.96	32.92
MAT* [50]	70.65	35.83	70.15	36.53	67.34	38.31	87.58	21.73	79.93	27.77	75.13	32.03
GFF* [28]	75.31	32.48	66.51	41.52	71.58	34.77	85.51	25.64	78.38	28.15	75.46	32.51
LTW [40]	77.14	29.34	67.12	39.22	74.58	33.81	88.56	20.57	77.95	29.01	77.07	30.39
LRL [4]	78.26	29.67	68.76	37.50	76.53	32.41	89.24	20.32	-	-	-	-
DCL [41]	82.30	26.53	71.14	36.17	76.71	31.97	91.66	16.63	80.21	27.37	80.40	27.73
PCL+I2G [51]	81.80	-	-	-	-	-	-	-	-	-	-	-
SBI* [35]	80.76	26.97	68.22	38.11	76.53	30.22	88.13	17.25	75.20	31.49	77.77	28.81
UIA-ViT [53]	82.41	-	-	-	75.80	-	94.68	-	-	-	-	-
RECCE* [2]	70.50	35.34	67.93	39.82	75.88	32.41	89.91	19.95	77.59	29.38	76.36	31.38
UCF [48]	75.27	-	-	-	75.94	-	80.74	-	-	-	-	-
CADDM* [9]	77.56	30.63	72.56	33.63	72.45	33.56	82.90	25.20	75.58	31.01	76.21	30.81
EN-b4* [42]	73.51	34.17	70.04	37.03	70.51	33.98	87.57	21.31	77.38	29.44	75.80	31.19
VIT-B* [42]	74.64	33.07	75.46	31.53	74.24	34.29	84.38	24.15	78.50	28.14	77.44	30.24
En-b4+Ours	83.17	24.59	75.17	33.25	77.35	30.17	91.71	16.27	82.02	25.55	81.88	25.97
VIT-B+Ours	80.46	27.51	80.14	29.62	80.95	27.66	90.36	19.73	86.98	21.32	83.78	25.17