search

vikolss / DACS

Code from the paper "DACS: Domain Adaptation via Cross-domain Mixed Sampling"
MIT License
123 stars 19 forks source link

Question about "Evaluation Conditions" in paper #10

Closed super233 closed 3 years ago

super233 commented 3 years ago

Hi, thanks for your awesome work. I have noticed that you said "DACS was evaluated as an average of three runs" in "Evaluation Conditions". Dose it mean that the reported result is the average of the best results from three runs? Or dose it mean that the reported result is the average of all results from three runs and each run contains many results?

vikolss commented 3 years ago

Each run only has one result, the one from when training is completed after 250k iterations. The reported result is the average of three runs like this. This is in contrast to several other works that we are comparing ours against, which actually use early stopping, i.e. report the best result from the course of the training, something we believe is not a fair evaluation. I hope this answers your question! :)

super233 commented 3 years ago

Thanks for your reply! :-)