search

pytorch / vision

Datasets, Transforms and Models specific to Computer Vision
https://pytorch.org/vision
BSD 3-Clause "New" or "Revised" License
15.96k stars 6.92k forks source link

Places365 #5347

Open pmeier opened 2 years ago

pmeier commented 2 years ago

cc @pmeier @bjuncek

Amapocho commented 2 years ago

Can I take this up?

pmeier commented 2 years ago

Sure, go ahead.

Amapocho commented 2 years ago

The Places365-Challenge dataset is way too big for me to download and find the sha256 code, is there any other way to find it?

pmeier commented 2 years ago

I think I habe the files in disk. Send the PR without it. I'll fill them in.

Amapocho commented 2 years ago

I'm sorry but I just realized I will need to download the entire dataset to build the data pipe and I cannot do that for such a huge dataset. Please do unassign this issue, I will look into the other datasets that I will be able to download and try to go forward with them.

Regret any inconvenience caused.

Should I make a PR with resources that I have updated or should I just let it be?

pmeier commented 2 years ago

I'm sorry but I just realized I will need to download the entire dataset to build the data pipe and I cannot do that for such a huge dataset. Please do unassign this issue, I will look into the other datasets that I will be able to download and try to go forward with them.

Regret any inconvenience caused.

No worries, I could have told you before you got started and this is on me. Still, before I unassign you, would you be willing to implement this for the small variant of the dataset. This limits the download size for all files to about 30GB. IIRC, they have exactly the same structure as their larger "sisters". So the only thing I would have to do after your PR is checking if the implementation still works with the other files.

Should I make a PR with resources that I have updated or should I just let it be?

Depends. Do you have more than just the skeleton? If yes, go ahead and send the PR. Otherwise I don't think it will be of much help for someone taking over.

Amapocho commented 2 years ago

I only have 25GB on the drive which has PyTorch so I won't be able to run it on my device sadly. I have added all the links for the resource download, image type, and all the other links so I'll add a PR as it'll help start off whoever picks up the issue next.

In the meanwhile, I have looked into the RenderedSST2 and I would be able to take that up. Is there any quirk to that dataset I should be wary of before commenting on the issue?

pmeier commented 2 years ago

In the meanwhile, I have looked into the RenderedSST2 and I would be able to take that up. Is there any quirk to that dataset I should be wary of before commenting on the issue?

No, I don't think so. Go ahead, I can assign you there.

pmeier commented 2 years ago

For anyone who wants to pick this up: you can find the skeleton for the implementation in #5383.