In the segmentation.py file on line 59:
hs, memory = self.detr.transformer(src_proj, mask, self.detr.query_embed.weight, pos[-1])
you refer to self.detr.query_embed.weight, but in transformer.py on line 123, this embedding layer is commented out. I'd like to understand the reason behind this and whether this is still valid for segmentation cases.
Hi, team!
In the segmentation.py file on line 59:
hs, memory = self.detr.transformer(src_proj, mask, self.detr.query_embed.weight, pos[-1])
you refer to self.detr.query_embed.weight, but in transformer.py on line 123, this embedding layer is commented out. I'd like to understand the reason behind this and whether this is still valid for segmentation cases.Thanks!