Gregor Lenz

thank you @fabhertz95. You make a very good point. Uploading the data ourselves could get us in trouble. Rethinking our options, I think I'm going to write to Prophesee to...

it used to be in there but I removed it (see this commit https://github.com/neuromorphs/tonic/commit/a7d78a747477b24eaf6fa4a1fbd56c58eca5b0ff) because in order to read the files, I was dependent on the loris package https://github.com/neuromorphic-paris/loris but...

We can reintroduce it (see the old commit), however I just want to have loris as optional dependency. So I suggest to do the following: copy the dataset file from...

Good point! Please try this. Thank you 🙏 Le 20 avril 2022 15:09:56 GMT+01:00, Fabrizio Ottati ***@***.***> a écrit : >I noticed something: on [this line](https://github.com/neuromorphs/tonic/blob/bc9fd1317a709c291bc0f16881d2ee7de2a2fcb6/tonic/io.py#L25), loris is used directly...

After getting in touch with Prophesee, they informed me that they plan on releasing more hdf5 datasets / files with their next release in September. Let's hope for the best!

I'm not really sure yet to be honest. I still don't understand how torchvisions new API is going to look like for datasets and transforms combined. I think it'll be...

good point, there is lots of stuff that they did already and I have yet to understand. like this one too https://github.com/pytorch/vision/blob/main/torchvision/prototype/datasets/_api.py

keeping track of supported datasets here: https://github.com/neuromorphs/tonic/tree/develop/tonic/prototype#currently-supported-datasets

that is a very good next step. If we are able to test the prototype datasets, we can start including them in the package. In general it would be good...

Do you mean to use this as a drop-in replacement to the current tonic.download_utils.download_url? I think this would work, although some datasets call it with the md5 argument as well...