Jiajun Fan

TPAMI 2026

PRANCE: Joint Token-Optimization and Structural Channel-Pruning for Adaptive ViT Inference

In brief

What the paper does.

PRANCE is a Vision Transformer compression framework that jointly optimizes token usage and structural channel pruning, cutting FLOPs by about 50% and keeping only about 10% of tokens with lossless Top-1 accuracy.

Published in
IEEE Transactions on Pattern Analysis and Machine Intelligence (TPAMI 2026)
Paper
arxiv.org/abs/2407.05010

Cite

Cite the paper.

@article{li2026prance,
  title={{PRANCE}: Joint Token-Optimization and Structural Channel-Pruning for Adaptive {ViT} Inference},
  author={Li, Ye and Tang, Chen and Meng, Yuan and Fan, Jiajun and Chai, Zenghao and Ma, Xinzhu and Wang, Zhi and Zhu, Wenwu},
  journal={IEEE Transactions on Pattern Analysis and Machine Intelligence},
  year={2026},
  url={https://arxiv.org/abs/2407.05010}
}