Asking about the class token.

[kona419]

Hello,
https://github.com/zbwxp/SegVit/blob/85616388cd8983e2d4daafd971540b210e956666/configs/_base_/models/seg_vit-b16.py#L24C9-L24C23

I found that you set
with_cls_token=False.
Is this means that ViT encoder in your model do not use class token, and use it for the decoder(ATM)?

[AkideLiu]

We used the feature embedding from the VIT encoder without using the class token. Alternatively, we initialized a class embedding in the decoder in

SegVit/decode_heads/atm_head.py

Lines 239 to 240 in 8561638

	outputs_class = self.class_embed(qs)
	out = {"pred_logits": outputs_class[-1]}

[kona419]

We used the feature embedding from the VIT encoder without using the class token. Alternatively, we initialized a class embedding in the decoder in

SegVit/decode_heads/atm_head.py

Lines 239 to 240 in 8561638

	outputs_class = self.class_embed(qs)
	out = {"pred_logits": outputs_class[-1]}

Thank you for the reply!
By the way, I was wonder when the SegViT2 especially Continual Learning part will be released.