Sungkyun Chang
Sungkyun Chang
@Mihonarium Solved by removing pathlib for argin. Also fixed same issue for --output option.
Hi @Novicei, thanks for good questions (sorry for the late reply). Actually I am interested in improving the evaluation method, too. > I would like to know if the 500...
@Novicei Thanks for the questions:) I believe this discussion will help provide a more accurate baseline for other follow-up research. > First, since the offset is only 1.2, it cannot...
TODO: - [ ] training '2s' segment model and reporting the test result
@stdio2016 Got it. Just in my implementation, it is equivalent to N uniform offsets for the query.
Yes, it's planned. But I didn't include a config file for that, because I need to test if the triplet loss works correctly after migrating to this open project. Actually...
@guillemcortes Sorry, somehow I forgot about this. I'll update it in a week.
@krishvishal No, it's still being delayed. I'll notify when it's updated.
@krishvishal @guillemcortes `Now-playing` model with `Online (semi-hard) triplet loss` is now available. ```python3 python run.py train CHECKPOINT_NAME -c now_playing ``` Results and pre-trained model will be available in this week.
@980202006 If I understood the question correctly, do you want to search without faiss using custom data (your mp3s)? : If so, [mini-search-validation](https://github.com/mimbres/neural-audio-fp/blob/b7de9e7f48de2f527725f68d9c155fad07fb25a1/model/trainer.py#L210) can be used. For example, you can...