ehofesmann / PawPal

Entertaining and Training your dog while you are away.

PawPal

Eric Hofesmann, Aman Kumar Jha, Preet Gill, Vihang Agarwal

Entertaining and Training your dog while you are away.

Using state-of-the-art computer vision algorithms, this dog localization and activity recognition system can determine what your dog is doing from a home surveillance camera.

Results

Biting C3D Recogntiion Accuracy (%)

Mean	Standard Deviation	Random Chance
68.41	6.10	50.00

Across 5 splits given in tfrecords_pawpal/split.npy in the dataset download link below

Usage

Requirements

Python 3.5

OpenCV

Tensorflow 1.0.0

Cython

Darkflow for Yolo

M-PACT Activity Recognition Platform

Detailed installation instructions below.

Installation and Setup

Follow instructions below to install darkflow and PawPal

git clone https://github.com/thtrieu/darkflow
virtualenv -p python3.5 env
source env/bin/activate
pip install tensorflow==1.0.0 
pip install Cython 
pip install opencv-python
cd darkflow
sudo apt-get install python3 python-dev python3-dev \
     build-essential libssl-dev libffi-dev \
     libxml2-dev libxslt1-dev zlib1g-dev \
     python-pip
pip install -e .
flow  (ignore any errors)
cd ..
git clone https://github.com/ehofesmann/PawPal/
cd PawPal

Download the weights for C3D Download link

Download yolo.weights Download link

mv ~/Downloads/checkpoint-532.npy ./c3d/
mv ~/Downloads/yolo.weights ../darkflow/bin/

Update the /path/to/darkflow in detect_video.py

python detect_video.py --vidpath example/example1.mp4

Testing

python detect_video.py --vidpath example/example1.mp4

Training or Finetuneing

Dataset

Dog biting vs non biting tfrecords dataset Download link

Activity Recognition Model

Install M-PACT and copy PawPal/c3d/c3d_frozen into the models directory of M-PACT