About ViTencoder setting

[Followmeczx]

hello. I don't really understand your approach to vit encoder. I didn't see specific building code for the ViT structure.
Also, I want to add the normal image stitpled with the input image into the ViT encoder, so I need to change the number of channels from 3 to 9, I wonder where I can change the code?
And I want to change the input image resolution from (256,192) to (256,256). This must involve some modification, can you give me some suggestions please!

[linjing7]

Hi, the code for the ViT encoder is here:

OSX/main/transformer_utils/mmpose/models/backbones/vit.py

Line 176 in 118cf97

class ViT(BaseBackbone):

[wangjunkaiyangzilong]

Hello, may I ask how the pre trained model of the encoder was obtained during training?