Arka Sadhu

> One thing I didn't try before is using cudnn.benchmark = True, which helps a lot. Woah, I din't know that. I am slightly unfamiliar with ec2 instances. Does g2.2xlarge...

@freedom6927 please see https://github.com/TheShadow29/zsgnet-pytorch/blob/master/DATA_README.md You need to download the annotation files in the drive link. Let me know if you run into any error.

@freedom6927 it is under vg_split / csv_dir. For training and validation, you could use `training_balanced.csv` and `val_balanced.csv` For testing, you would use `test_balanced_c2.csv` and `test_balanced_c3.csv` respectively. In table 5, we...

@freedom6927 sorry for the late reply, we only used the balanced validation set.

@freedom6927 Sorry, I don't understand your question. What do you mean by dataset division?

@freedom6927 I don't have it with me, but something along the following lines of code should be sufficient: ``` train_df = ... # read train csv test_df = ... #...

@kelikeli Thanks for your interest. If you want to create ASRL-like for your own dataset, you should follow the instructions in https://github.com/TheShadow29/vognet-pytorch/tree/master/dcode You may have to change a few things...

Hello, @1170300521 sorry for the delayed reply (for some reason this didn't turn up in my notifs). I recall using lr=1e-4 for 10 epochs, and 1e-5 for another 10 epochs....

@tejan-rgb Can you try changing `only_valid` to `only_val`?

Hi. Unfortunately, this repo uses fastai v0.7 and not fastai v1. Might be best to ask in the forums https://forums.fast.ai/