search

jackroos / VL-BERT

Code for ICLR 2020 paper "VL-BERT: Pre-training of Generic Visual-Linguistic Representations".
MIT License
738 stars 110 forks source link

4-d location vector #36

Closed kememem closed 4 years ago

kememem commented 4 years ago

Hi,

I am looking at your code. Can you please point me out which part of the code is doing location vector normalization? I can see you pass 4-d raw location into the model, but I could find the place that you normalize it with image width and height.

Thanks!

jackroos commented 4 years ago

@kememem https://github.com/jackroos/VL-BERT/blob/d58c70c78bbeb9ff278ee79f93a6a0f02e66a914/common/utils/bbox.py#L55

kememem commented 4 years ago

Thank you so much for your prompt reply. Can you further elaborate in which file you call this coordinate_embeddings function?

jackroos commented 4 years ago

@kememem

https://github.com/jackroos/VL-BERT/blob/d58c70c78bbeb9ff278ee79f93a6a0f02e66a914/common/fast_rcnn.py#L170

Sorry for the late reply.