search

microsoft / VideoX

VideoX: a collection of video cross-modal models
Other
968 stars 160 forks source link

DataSet Preparation and Use of Decord #59

Closed aliman80 closed 2 years ago

aliman80 commented 2 years ago

Hi, Thank you for your wonderful work. i was trying to download all the requirements to reproduce the results. I have following trivial queries please;

1: Where and how can we use decord in this code base. 2: If possible can you elaborate the arrangment of dataset. what i understood for the option 2 is that we need to download zipped data. 3: Where in zipped files are we using train, test label files.

nbl97 commented 2 years ago

Thanks for your interest in our work~Here are my responses:

  1. We borrow the pipeline of loading video from MMAction2. Specifically, we define the pipeline in datasets/build.py#L248. All the class about data preparation and augmentation are defined in datasets/pipeline.py, where you can find DecordInit and DecordDecode.
  2. You do not need to download zipped data. The provided two options are just about how to organize the dataset, not about how to download the dataset. You can download the dataset by a means convenient to you. After the downloading, there are two options to organize them: putting all the videos (including training set and testing set) together in a FOLDER (option 1) or TAR FILE (option 2).
  3. Same with option 1, we still need train.txt and val.txt to list the videos and labels. We have updated the README for easy understanding.
aliman80 commented 2 years ago

Thank you for your kind response.