Official Implementation of Token Adapter [ECCV 2024]. The implementation of the Token Adapter algorithm is located in the file tokenAdapter.py. For the experimental section of the paper, please refer to Expedit-LargeScale-Vision-Transformer and simply insert the Token Adapter to use it.

Acknowledgement

The code implementation refers to the following open source project:

• ToMe: We were inspired by this approach to compress visual tokens from a structured perspective
• Expedit-LargeScale-Vision-Transformer: all of our experimental setups are based on this repository.