ViT: Counting Image Patches
Beginner
The problem, from first principles
The Vision Transformer (Dosovitskiy et al., 2021) cuts an image into fixed-size patches and treats each patch as a token. The number of tokens, and hence the cost of attention, follows directly from the image size and patch size.
From theory to code
Implement num_patches(img, p), the number of patches for a square image.
Constraints
- Use integer division.
Hints
Hint 1
Divide the side length by the patch size, floor, then square.
Keywords
research-paperscomputer-visiontransformervision
Python
Run your code to see results
No results yet
Run your code to check it against the examples, or submit to run every test.