MichaelFan01 / STDC-Seg

Source Code of our CVPR2021 paper "Rethinking BiSeNet For Real-time Semantic Segmentation"
MIT License
703 stars 131 forks source link

pretrained model #7

Open guo0522 opened 3 years ago

guo0522 commented 3 years ago

@MichaelFan01 你好!首先感謝您的貢獻,這是一篇很驚艷的作品。 針對paper部分有一點疑惑希望您能幫我解惑 在abstract中有提到 BiseNet backbones borrowed from pretrained tasks, e.g., image classification, may be inefficient for image segmentation due to the deficiency of taskspecific design 您是指Xception39 or ResNet是for Image classification task 因此不是用於Semantic Segmentation task 不過在paper3.2.1節中第一句話提到 We use the pretrained STDC networks as the backbone of our encoder 想請問您關於pretrainedSTDC是用於何種task 或 Dataset 做training的 感謝您!~

MichaelFan01 commented 3 years ago

是这样,我们的意思是Xception39 or ResNet没有针对分割任务做专门的设计,不是说for Image classification task不能用于Semantic Segmentation task; pretrainedSTDC跟其他的pretrained model一样,都是在ImageNet上做的预训练

miscedence12 commented 2 years ago

@MichaelFan01 请问下,你们的预训练模型在哪可以找到?谢谢

chris2000520 commented 4 months ago

是这样,我们的意思是Xception39 or ResNet没有针对分割任务做专门的设计,不是说for Image classification task不能用于Semantic Segmentation task; pretrainedSTDC跟其他的pretrained model一样,都是在ImageNet上做的预训练

同问!