Scripts for downloading datasets?

[ir2718]

Hi,

I think it would be much easier to use this repo if there were scripts for downloading well known benchmarks (Union14M, IIIT, IC13, IC15, ...). Ideally, users would run these scripts if they need a certain dataset, and then modify the config to suit their training/eval needs. If you're okay with this I would like to create a PR.

Additionally, I found it annoying that there is no documentation or at least more information on using the repo in the readme file. If I have time I would like to add this as well.

[Topdu]

Sure! We'd like and appreciate if you would create a PR with the Dataset Download scripts and documentation.

[ir2718]

I've only got one question regarding the possible implementation: is it ok to treat torchvision as a dependency? I see it's not present in the requirements.txt file, but it is used in the preprocess module.

[Topdu]

Yes, torchvision should be installed.

[ir2718]

Is there any way you can host the Union14 datasets on a Google Drive or server? The issue is that the authors set it up on Baidu and OneDrive. Baidu causes issues for people who are not based in asia, while downloading from OneDrive would introduce including the API for downloading in the dependencies. Alternatively, I could ask the authors to upload the dataset to Google Drive.

[ir2718]

I made a draft PR, so when you've got time, please have a look @Topdu