Understanding Vision Transformers (ViTs): Hidden properties, insights, and robustness of their representations
We study the learned visual representations of CNNs and ViTs, such as texture bias, how to learn good representations, the robustness of pretrained models, and finally properties that emerge from trained ViTs.
We study the learned visual representations of CNNs and ViTs, such as texture bias, how to learn good representations, the robustness of pretrained models, and finally properties that emerge from trained ViTs.
What's Your Reaction?
