High-Fidelity Differential-information Driven Binary Vision Transformer
This work addresses the trade-off between computational efficiency and accuracy for deploying vision transformers on resource-constrained edge devices, representing an incremental improvement over existing binary ViT methods.
The paper tackles the performance degradation in binary vision transformers for edge devices by proposing DIDB-ViT, which uses differential information and frequency decomposition to enhance information retention, achieving superior image classification and segmentation results compared to state-of-the-art methods.
The binarization of vision transformers (ViTs) offers a promising approach to addressing the trade-off between high computational/storage demands and the constraints of edge-device deployment. However, existing binary ViT methods often suffer from severe performance degradation or rely heavily on full-precision modules. To address these issues, we propose DIDB-ViT, a novel binary ViT that is highly informative while maintaining the original ViT architecture and computational efficiency. Specifically, we design an informative attention module incorporating differential information to mitigate information loss caused by binarization and enhance high-frequency retention. To preserve the fidelity of the similarity calculations between binary Q and K tensors, we apply frequency decomposition using the discrete Haar wavelet and integrate similarities across different frequencies. Additionally, we introduce an improved RPReLU activation function to restructure the activation distribution, expanding the model's representational capacity. Experimental results demonstrate that our DIDB-ViT significantly outperforms state-of-the-art network quantization methods in multiple ViT architectures, achieving superior image classification and segmentation performance.