Full Professor, Institute of automation, Chinese academy of science
2 papers at NeurIPS 2025
We propose ETT, an end-to-end tokenizer tuning approach that enables joint optimization between vision tokenization and target autoregressive tasks.