which backbone will be better?

[simplify23]

i see many backbones in the backbone config file and the Resnet50_FPN_Deform as default. I also find dilated and deform are embedding in backbone, will these blocks have great performance for this OCR task? What are the differents between Resnet50_FPN_Deform and ResNet50? I just see nn.ConvTranspose2d() in FPN_Deform function

[qyfff]

代码错误太多没意思