30 Apr 2025
This research from Joshigakuin Senior High School and AIST/Oxford proposes a straightforward method for identifying visual artifacts in videos generated by OpenAI's Sora model. By training 2D convolutional neural networks on specific artifact categories, the study achieved a 94.14% mean accuracy with ResNet-50, indicating that current AI-generated videos exhibit consistent and detectable flaws.
There are no more papers matching your filters at the moment.
