The project involves separating the object recognition task from activity recognition, so that, the activity recognition is agnostic to the objects that are involves, and thus improves learning when the dataset is heavily skewed towards certain objects performing certain activities, which wont lead to better generalization if different objects are performing the same activity.
Intra Activity Skewness
Inter Activity Skewness
All the models are in the models directory
Follow the following steps:
- Add your makefile.config in caffe-sl directory
- Run
make all -j10 - All done
Thats it. Note that this caffe-sl version is stripped off of unwanted files in the original caffe-sl version for the project
- Install Torch
- Install PyTorch
- Requires torch packages:
luarocks install nnlr cudnn rnnto install. - Requires python packages:
pip install h5py numpy scikit-learnto install
All the data is to be kept in the project-data directory and queried from there.
All the experiment scripts should be in Experiments directory and should call the models directory from there.
There are some utility scripts that we can use. It is not properly strucutred. As the project is in progress