hustvl/MIMDet

Could you provide a model of 768 dimension output image features?

Pter61 opened this issue · 2 comments

Pter61 commented

There are many works that require image embeddings that have the same dimensions as text embeddings (e.g. BERT-base(768 dim), BERT-large(1024 dim). However, I noticed that you only provided the model of 1024 dimensions; therefore, could you provide more models of 768 dimensions?

Yuxin-CV commented

There are many works that require image embeddings that have the same dimensions as text embeddings (e.g. BERT-base(768 dim), BERT-large(1024 dim). However, I noticed that you only provided the model of 1024 dimensions; therefore, could you provide more models of 768 dimensions?

Hi.
The embedding dim of all the vit base models is 768.

Pter61 commented

There are many works that require image embeddings that have the same dimensions as text embeddings (e.g. BERT-base(768 dim), BERT-large(1024 dim). However, I noticed that you only provided the model of 1024 dimensions; therefore, could you provide more models of 768 dimensions?

Hi. The embedding dim of all the vit base models is 768.

OK! Thank you!