CLIP-Mamba: CLIP Pretrained Mamba Models with OOD and Hessian Evaluation
This work addresses the need for parameter-efficient and robust vision models for researchers and practitioners, though it is incremental as it applies existing methods to a new model type.
The authors tackled the problem of training transferable Mamba models using CLIP pretraining, finding that a 67M-parameter Mamba model matches a 307M-parameter ViT in zero-shot classification and excels in OOD generalization under specific conditions like image contrast or high-pass filtering.
State space models and Mamba-based models have been increasingly applied across various domains, achieving state-of-the-art performance. This technical report introduces the first attempt to train a transferable Mamba model utilizing contrastive language-image pretraining (CLIP). We have trained Mamba models of varying sizes and undertaken comprehensive evaluations of these models on 26 zero-shot classification datasets and 16 out-of-distribution (OOD) datasets. Our findings reveal that a Mamba model with 67 million parameters is on par with a 307 million-parameter Vision Transformer (ViT) model in zero-shot classification tasks, highlighting the parameter efficiency of Mamba models. In tests of OOD generalization, Mamba-based models exhibit exceptional performance in conditions of OOD image contrast or when subjected to high-pass filtering. However, a Hessian analysis indicates that Mamba models feature a sharper and more non-convex landscape compared to ViT-based models, making them more challenging to train. The source code is available at https://github.com/raytrun/mamba-clip.