search

Dootmaan / DSRL

Implementation of CVPR 2020 Dual Super-Resolution Learning for Semantic Segmentation
MIT License
95 stars 17 forks source link

怎么在推理阶段把超分分支删除啊 #21

Open rtrtfs opened 1 year ago

rtrtfs commented 1 year ago

可以给一个推理阶段的代码吗

Felix-fz commented 7 months ago

请问你搞明白了吗

Dootmaan commented 7 months ago

Hi @CopleManba and @rtrtfs, thank you for your questions.

This is actually pretty easy. You can split the original network into two subnets, where the first one is for segmentation that contains the encoder and the second one is the SISR decoder appended after the encoder output of the segmentation subnet. When inference, you can directly use the segmentation subnet and abandon the SISR one.

Felix-fz commented 7 months ago

Hi @Dootmaan , thank you for your answer. But I still do not konw why the segmentation subnet is optimized when the SISR brach is deleted.