askforalfred / alfred

ALFRED - A Benchmark for Interpreting Grounded Instructions for Everyday Tasks
MIT License
362 stars 78 forks source link

How much storage space do Resnet Features need? #42

Closed LWJ312 closed 4 years ago

LWJ312 commented 4 years ago

Hello~ I met some trouble and I need your help. I have a limited storage space with my PC, so I download json data you offered. I have generated some images and then I want to get Resnet features using extract_resnet.py. But after extracting features in _valseen and _valunseen, all the feature.pt files have taken 14.8G,I made an estimation that the train data features may need 200G. But you offered a 17G data including jsons and features, I don't know how much it takes after unzipping. Maybe it takes less than 200G? Thanks for your help in advance!

MohitShridhar commented 4 years ago

@LWJ312 the unzipped json_feat_2.1.0 folder is ~35GB. It's much smaller because it doesn't contain any generated images.

LWJ312 commented 4 years ago

@MohitShridhar Thank you~ Yeah, I know what you mean, but now I wonder why my feature.pt files generated following the _trajdata.json are much larger than 35GB?

MohitShridhar commented 4 years ago

So the generated feature.pt contains ResNet features for every frame in the video demonstration. If you turn off smooth_nav, it won't save extra frames used to generate smooth-looking videos.

LWJ312 commented 4 years ago

Oh! I get it. Thank you so much :)