Questions about the role of mixture of vision encoders.

[yifang23]

Hello, thank you very much for your excellent work. I have two questions to ask.

1. I noticed that in your paper it is mentioned that the Mixture of vision encoders has brought performance improvements on 12 - 14 benchmarks. However, I haven't seen the ablation experiment regarding the Mixture of vision encoders. May I ask if there are specific data that can be shared? How much performance improvement has it actually brought?
2. I noticed that the tasks where the Mixture of vision encoders brings the greatest performance improvement are OCR and Chart/Document VQA tasks, which seem to be small - target recognition and understanding tasks. May I ask if this is because tailing is used as input, which retains more picture details?