Install the depencies by running:
bash Requirements.txt
The important parts of the code are organized as follows.
main.py # main script (maybe parse arguments here too)
config.py # file with all the parameters used across all modules
Requirements.py # file with the dependencies
data_utils
├── data_loader.py # loader for the something-something v2 dataset
├── preprocess.py #
├── metadata_loader.py #
├── somethingsomethingv2 # default parser
├── video_loader.py # loader for the videos
└── default
├── category.txt # file with all categories as rows
└── someth..thingv2.py # default parser of the dataset
training
└── train.py # called to start training
data
├── 20bn-something-something-v2-video # the full dataset
├── 20bn-something-something-v2-jason # jason labels comming with the dataset
├── 20bn-something-something-v2-frame # extraced frames
└── 20bn-something-something-v2-label # labels for train, test and valid set
models
└── Xpection.py # Xcpection based model
In main.py we keep the progresss towards the endgoal. All implementations take place outside, either in data_utils, models or training.
- Windows/Liunux/Mac: Follow the steps shown here to get the gcloud CLI.
- Ubuntu (apt-get): Follow the steps shown here to get the gcloud CLI.
It will prompt you to login once the installation is finished. The log in is done through your webb browser, make sure to use log in using the same account as the one with the 50$ on GCP. The it will ask you to type in the number corresponding to our project. If you are correctly added to them team you should see DL-DD2424 as a GCP project.
If using Powershell on Windows, make sure to restart powershell in order to use the gcloud commands.
Once the installation is confirmed, run
gcloud components list # to see all you installed components
gcloud components install app-engine-python # to install the app engine support for python
pip install keras-tuner