question about pre-aligning the visual encoder

[chloroplast567]

Great job! Thank you so much for open-sourcing your code and models! I have a few more questions. When pre-aligning the visual encoder, do you perform any post-processing on each task, or is it just used as a feature extractor?