[WIP][Bug fix] use pad_shape instead of img_shape for bevformer and recover the origin code where the first frame pre_bev is None #225
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Use
pad_shape
instead ofimg_shape
for bevformer, whereimg_shape
reserves the origin image shape andpad_shape
is the image shape after padded.Recover the origin codes where the first frame
pre_bev
is None. Present codes set the first framepre_bev
to zeros during the trainng phase in order to keep consistent with the deploy phase, but it seems to be not too right. The reasons are two below:pre_bev
should be set as None.pre_bev
is the key and value of the temporal encoder layers. Ifpre_bev
is None, the encoder layer will set the value ofpre_bev
the same as query.