2025 IEEE International Conference on Acoustics, Speech, and Signal Processing

IEEE Signal Processing Society website | 6-11 April 2025 | 2025 SP Cup Official Document
[Sponsored by MathWorks Inc. and the IEEE Signal Processing Society]

With the surge in advancements in synthetic data generation, deepfakes pose a serious threat due to their potential application in unethical and malicious purposes. This encompasses manipulating public opinion, instigating geopolitical tension, defamation, and identity threats. This year marks a significant election season globally, with more voters than ever in history expected to head to the polls across at least 64 countries. Naturally, deepfakes are of greater concern, as the generated contents in the form of image, video and audio are already widely circulated to gain political advantages. Addressing deepfakes is increasingly challenging due to the continuous development of new generation techniques, creating an ‘arms race’ between the attacker and the defender, as also mentioned in a recent BBC news report.

Of all potential deepfake content, fake images are more prevalent, as they can be easily generated by non-experts using publicly available tools that require minimal domain knowledge. In recent years, several solutions are proposed to discriminate the fake content from real. However, their effectiveness is questionable due to a lack of generalization, especially in the presence of unseen attacks not represented in the training data. To address this challenge, we present the Deepfake Face Detection In The Wild Competition (DFWild-Cup), which focuses on automatically detecting image deepfakes in real-world settings, with participants assigned to develop systems capable of distinguishing between real and computer-generated images. This DFWild-Cup competition specifically emphasizes the generalization aspects of deepfake detectors by encompassing a wide variety of data generated using diverse generation methods, attack scenarios and source datasets. While a few other machine learning challenges already addressed the deepfake detection, this represents the inaugural challenge solely focused on image deepfake detection in the wild, highlighting its emphasis on diversity, particularly in real-world scenarios.
Read the full Call for Participation for more details.

The DFWild-Cup is concerned with identifying images whether they are real or fake. The real image can be natural image collected with a camera or snapshots of real videos. Whereas the fake images may be generated with image various image synthesis or forgery techniques or can be snapshot of computer-generated fake videos. The participant will be provided with a set of training and test data. After training the detector with the dedicated training data, the task is to estimate a score of each of the test images. A higher score indicates a real image whereas a lower score indicates a fake image

Full technical details, dataset(s), evaluation metrics, and all other pertinent information about the competition is located in the 2025 SP Cup Official Document.

This competition is sponsored by the IEEE Signal Processing Society and MathWorks