Total number of frames for training and testing

[ggosjw]

Thanks for open sourcing your amazing and wonderful work!
I have a question about the total number of frames for training:

I followed your Readme and downloaded the dataset only from uncompressed/scenario/training_20s.

Altogether 50 files:

I followed your paper, splitting the dataset into 80% of training and 20% of testing and validating. Then I run your data_process.py file to preprocess the data.

However, when training, i found my total number of training frame is 36111, which is far fewer than 113,622 mentioned in your paper.


So, my question is that whether i download the incomplete dataset or there is something wrong in my reproduction process.

Thank you for your kind help and patience.

[MCZhi]

Actually, there are 1000 files in the uncompressed/scenario/training 20s folder, so I suggest you go and check out the files on the website again. Note that I didn't use all the data files and only selected about 110 files in total.

[ggosjw]

OK,thank you for your kind help. That really helps me a lot.