Open ljc19940403 opened 3 years ago
Hi, I am not sure about what kind of loss can help resolve it. I would suggest using more real-world data to train/finetune the model (e.g., you can use our released real-world portrait dataset to finetune the model), or adding some prior to post-processing (e.g., keep the largest connected component).
Hi, thank you very much for your great work! At present, I encounter the following problems. In the following scenarios, filamentous objects are detected as foreground. Can I solve such problems by adding another Loss, such as BCE. Or do you have any suggestions? Looking forward to your reply!