ViT: The Sequence Length With a Class Token
Beginner
The problem, from first principles
A ViT prepends a learnable class token to the patch tokens. The class token's final state is used for classification, so the sequence the transformer sees has one more entry than the patch count.
From theory to code
Implement vit_seq_length(img, p), the number of tokens the transformer processes.
Constraints
- The class token is the first token.
Hints
Hint 1
Count the patches, then add one.
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.