search

xhu248 / semi_cotrast_seg

Apache License 2.0
65 stars 7 forks source link

Unable to match results of the paper #3

Open sarahmhooper opened 2 years ago

sarahmhooper commented 2 years ago

Thank you for your paper and code. When using the provided code for data preprocessing and model training, I am unable to match the results of the paper. My results are worse than the baseline numbers (random init) reported in the paper. Could the authors comment on how to replicate results? Perhaps hyperparameter settings need to be refined, or maybe there is a bug in the codebase?

mengmeng0068 commented 2 years ago

It seems that the pre-training model is not loaded inside the fine-tuned model：I don't see the parameter config.load_model in the SegExperiment python file

wangyihub commented 1 year ago

I have a question tha:i am not sure that the percent of labeled data to use in ablation experiments in table 1 is pionted in step 2 (pixel contrast in supervised learing) ,or step 3(fine tune).