Rethinking Vision Transformer Depth via Structural Reparameterization

EfficientFSL introduces a query-only fine-tuning framework for Vision Transformers (ViTs), enhancing few-shot classification while significantly reducing computational demands. This approach leverages the pre-trained model's capabilities, achieving high accuracy with minimal parameters.

A new framework named CoEvo has been proposed for zero-shot out-of-distribution (OOD) detection in vision-language models, addressing the challenges posed by the absence of labeled negatives. CoEvo employs a bidirectional adaptation mechanism for both textual and visual proxies, dynamically refining them based on contextual information from test images. This innovation aims to enhance the reliability of OOD detection in open-world applications.

The introduction of the Diffusion-Guided Autoencoder (DGAE) marks a significant advancement in latent representation learning, enhancing the decoder's expressiveness and effectively addressing training instability associated with GANs. This model achieves state-of-the-art performance while utilizing a latent space that is twice as compact, thus improving efficiency in image and video generative tasks.