Leveraging Representations from Intermediate Encoder-blocks for Synthetic Image Detection
This work addresses the integrity and safety of online information by improving synthetic image detection, though it is incremental as it builds on existing feature extraction methods from foundation models.
The paper tackled the problem of detecting synthetic images by leveraging intermediate Transformer blocks of CLIP's image-encoder to capture fine-grained details, achieving an average +10.6% absolute performance improvement across 20 test datasets with training times as low as ~8 minutes per epoch.
The recently developed and publicly available synthetic image generation methods and services make it possible to create extremely realistic imagery on demand, raising great risks for the integrity and safety of online information. State-of-the-art Synthetic Image Detection (SID) research has led to strong evidence on the advantages of feature extraction from foundation models. However, such extracted features mostly encapsulate high-level visual semantics instead of fine-grained details, which are more important for the SID task. On the contrary, shallow layers encode low-level visual information. In this work, we leverage the image representations extracted by intermediate Transformer blocks of CLIP's image-encoder via a lightweight network that maps them to a learnable forgery-aware vector space capable of generalizing exceptionally well. We also employ a trainable module to incorporate the importance of each Transformer block to the final prediction. Our method is compared against the state-of-the-art by evaluating it on 20 test datasets and exhibits an average +10.6% absolute performance improvement. Notably, the best performing models require just a single epoch for training (~8 minutes). Code available at https://github.com/mever-team/rine.