Welcome to the final project of CSC249!

In this final project, you are going to build deep learning models for two tasks on A2D dataset. Please read final_project.pdf for more details of the project requirement.

Update 1 [April 3 1:30pm] : The dataset has been updated. If you have already download the older one, please remove it by rm -rf A2D. Please download the latest dataset (the same link given below) and extract the frames from videos again. You don't have to train your model again if you train it on the older dataset.

Before start working on a specific task, please do the following preparation on your Google Cloud.

• Clone the repository

  Please use the following command to clone this repository (please do not download the zip file):

  git clone --recursive https://github.com/rochesterxugroup/csc249_final_proj.git

  If there is any updates of the repository, please use the following commands to update:

  git submodule update --remote --merge
  git pull --recurse-submodules

  cd to the cloned repo:

  cd csc249_final_proj

• Environment Configuration

  1. Download and install miniconda from https://repo.anaconda.com/miniconda/Miniconda3-latest-Linux-x86_64.sh

  2. Install the virtual enviroment and its dependencies by:

     conda env create -f env.yml

  3. Activate the virtual environment by (please remember to activate virtual environment everytime you login on google cloud):

     conda activate pytorch_0_4_1

  4. Then, install Pytorch 0.4.1 and torchvision

     conda install pytorch=0.4.1 cuda92 -c pytorch
     conda install torchvision

  5. Install the ffmpeg via

     sudo apt install ffmpeg

• Download A2D dataset

  Please make sure you are at the csc249_final_proj directory.

  1. Download the A2D dataset

     curl http://www.cs.rochester.edu/~cxu22/t/249S19/A2D.tar.gz --output A2D.tar.gz

  2. Decompress the tar ball and remove tar ball.

     tar xvzf A2D.tar.gz
     rm A2D.tar.gz

  3. Extract frames from videos

     (Tip: Since it takes a long time to extract frames from video, you can execute the command in screen or tmux, in case the disconnection happens.)

     python extract_frames.py

Please read submission/README.md for more details of submission format.