4 ms·

Convolutional Neural Networks for Visual Recognition

• ks2048 2y ago
  One page says "CS231n: Deep Learning for Computer Vision" and another says "CS231n: Convolutional Neural Networks for Visual Recognition". Did they change it recently to recognize other methods (ViT), or?
  • danjl 2y ago
    Certainly still worth learning CNNs. Still unclear if ViT is better. And there's certainly enough for a full course on CNNs and a separate course on vision transformers.
    • fzliu 2y ago
      Agreed. ViTs are better if you're looking to go multimodal or use attention-specific mechanisms such as cross-attention. If not, there's evidence out there that ViTs are not better than convnets for small networks and at scale (https://frankzliu.com/blog/vision-transformers-are-overrated https://frankzliu.com/blog/vision-transformers-are-overrated).
      • abrichr 2y ago
        ViTs also have proven to be more effective for zero-shot generalization tasks due to their ability to capture global context and relationships in the input data, which CNNs struggle with. https://arxiv.org/abs/2304.02643 https://arxiv.org/abs/2304.02643