facebookresearch/AVT

Data Augmentation and End-to-end Training

Closed this issue · 3 comments

okay-okay commented

Hi,

I'm wondering if you have any results regarding performance without data augmentation and end-to-end training? If not, do you know how much augmentations helped in terms of overfitting and performance? Currently, I'm trying to run longer time-horizon experiments without augmentations, but it doesn't seem to perform much better than 2 second horizon experiments. Thank you!

rohitgirdhar commented

Hi, unfortunately I don't; I just used the standard video augmentations. You should be able to remove those and train using this code though.

okay-okay commented

Thank you for your response!

Where can I find the standard video parameters you've used for data augmentations? I can't seem to find any experiment configs, besides some txt files in expts/ folder.

rohitgirdhar commented

All the experiment configs are the in the expts/ folder. They override the defaults here.