The field of visual perception and domain adaptation is moving towards a more comprehensive understanding of how to effectively recognize and adapt to new environments. Recent research has focused on developing novel variational models for image decomposition, as well as improving the performance of large vision-language models in recognizing shapes, textures, and materials. Additionally, there has been a push towards enhancing source-free domain adaptation techniques, including the use of vision-and-language models and progressive curriculum labeling. Noteworthy papers in this area include:

• The work on Image Decomposition with G-norm Weighted by Total Symmetric Variation, which proposes a new model for decomposing images into cartoon and texture parts.
• The introduction of the Large Shape & Textures dataset, which is used to evaluate the performance of large vision-language models in recognizing shapes and textures.
• The development of ViLAaD, a method that enhances source-free domain adaptation by incorporating vision-and-language models.
• The proposal of ElimPCL, a progressive curriculum labeling method that eliminates noise accumulation in source-free domain adaptation.