Question about the data process
Closed this issue · 2 comments
Hi @magehrig
After one year, I still use and learn from your work (^_^), thanks for your nice work again. Too many advanced techniques in your work I need to learn.
This time I want to ask about the data loader, I noticed two types of data loaders are used in your code, i.e., stream loader and random loader. In my impression, many works often load the dataset in a stream way, to fit the RNN training, I want to know whether the extra random loader can bring improvement for the RVT compared to the pure stream loader since the code of mixed loaders seems rather complex, and this way is originated from your or other existing works?
Hatisn
Best!
Hi @Hatins
Happy to hear that!
Yes it helps a lot. There is a small ablation in the supplementary material that shows this. Also helps with the stability of the training. I came up with that myself; have not seen it anywhere else.
OK! I got it, it's a nice innovation and deserves to be learned by us！
Thank you again for your reply despite your busy schedule.