Words Matter: Leveraging Individual Text Embeddings for Code Generation in CLIP Test-Time Adaptation
This work addresses reliability issues for users of CLIP models in real-world applications with distribution shifts, though it is incremental as it builds on existing test-time adaptation techniques.
The paper tackled the problem of performance degradation in CLIP vision-language models under distribution shifts by proposing CLIP-OT, a test-time adaptation method that uses class text embeddings and optimal transport for pseudo-labeling, achieving up to 7% performance gains over state-of-the-art methods.
Vision-language foundation models, such as CLIP, have shown unprecedented zero-shot performance across a wide range of tasks. Nevertheless, these models may be unreliable under distributional shifts, as their performance is significantly degraded. In this work, we explore how to efficiently leverage class text information to mitigate these distribution drifts encountered by large pre-trained vision-language models (VLMs) during test-time inference. In particular, we propose to generate pseudo-labels for the test-time samples by exploiting generic class text embeddings as fixed centroids of a label assignment problem, which is efficiently solved with Optimal Transport. Furthermore, the proposed adaptation method (CLIP-OT) integrates a multiple template knowledge distillation approach, which replicates multi-view contrastive learning strategies in unsupervised representation learning but without incurring additional computational complexity. Extensive experiments on multiple popular test-time adaptation benchmarks presenting diverse complexity empirically show the superiority of CLIP-OT, achieving performance gains of up to 7% over recent state-of-the-art methods, yet being computationally and memory efficient.