CASA: Cross-Attention via Self-Attention for Efficient Vision-Language Fusion
This addresses the computational cost problem for researchers and practitioners using VLMs in high-resolution or streaming applications, representing an incremental improvement over existing cross-attention methods.
The paper tackles the efficiency-performance trade-off in vision-language models by proposing CASA, a method that reduces the performance gap with full token insertion models while maintaining scalability for long-context tasks like video captioning.
Vision-language models (VLMs) are commonly trained by inserting image tokens from a pretrained vision encoder into the textual stream of a language model. This allows text and image information to fully attend to one another within the model, but becomes extremely costly for high-resolution images, long conversations, or streaming videos, both in memory and compute. VLMs leveraging cross-attention are an efficient alternative to token insertion but exhibit a clear performance gap, in particular on tasks involving fine-grained visual details. We find that a key to improving such models is to also enable local text-to-text interaction in the dedicated cross-attention layers. Building on this, we propose CASA, Cross-Attention via Self-Attention, a simple and efficient paradigm which substantially reduces the gap with full token insertion on common image understanding benchmarks, while enjoying the same scalability as cross-attention models when applied to long-context multimodal tasks such as streaming video captioning. For samples and code, please see our project page at https://kyutai.org/casa .