search

Solacex / CCM

[ECCV2020] Content-Consistent Matching for Domain Adaptive Semantic Segmentation
MIT License
94 stars 10 forks source link

Questions about the two models #15

Closed ArcticOc closed 3 years ago

ArcticOc commented 3 years ago

Hi   I am a little confused about these two models, the ImageNet-pretrained Model and the Source-only Model. Why does the Source-only Model appear in this ccm_config.yml file？CCM and Source only, aren't they two different methods ? https://github.com/Solacex/CCM/blob/4e80b98007219e64d577fb105a35140cb2d13238/config/ccm_config.yml#L70   In addition, from the section 4.2 of paper, "We start from DeepLabV2-Res101with the backbone pretrained on the ImageNet. Then we firstly finetune the whole network on the source data and use such a source-trained network to initialize the target (adaptation) model.", The ImageNet-pretrained Model should be the adaption model. I am a little confused. THANKS!   PS: There is a bug with 'News: GTA-4K list is avalible!'.

Solacex commented 3 years ago

Hello, Thanks for your interest in our work!

The training consists of two stages, the source-only stage and the adaptation stage. The two stages of training are conducted consecutively, i.e., the second stage takes the trained model from the first stage as the initialization.

Thus, the initial weight of source only stage and adaptation stage are imagenet-pretrained model and the source-only model, accordingly.

ArcticOc commented 3 years ago

THANKS! I get it now.