Describe the bug
On line 15 of mmdet2trt/models/dense_heads/detr_head.py:self.positional_encoding = build_wrapper(module.positional_encoding)
Since mmdet version 3, the parameter positional_encoding is moved out from DETRHead's scope, so the module does not the attribution of positional_encoding.
environment:
torch=2.2.0
tensorrt=8.6.1
mmdetection=3.3.0
cuda=11.7
mmdet2trt=0.6.0