CVComp - Compares Computer Vision Models

This project is a proof of concept for comparing different computer vision models and their performance.

It supports running various models locally using multiprocess and remotely using redis pubsub.

Before setting up the environment

Note that you may need to install a rust compiler in order to build packages for particular platforms. Visit https://rustup.rs for more information.

Manually setting up the environment using Anaconda

Create a new Python environment based on Python 3.10 using Anaconda and activate it.

Note: Environment name is cvcomp

conda install -n cvcomp "numpy<1.24"
python -m pip install "opencv-python<4.8" 
python -m pip install redis --upgrade-strategy only-if-needed

Tensorflow

Please install an adequate distribution of tensorflow for your platform or follow the instructions below if available for your platform.

MacOS ARM (M1) - Apple Silicon

Guides:

• TensorFlow
• PyTorch

python -m pip install tensorflow-macos tensorflow-metal --upgrade-strategy only-if-needed

Finalizing environment setup

python -m pip install tensorflow-hub --upgrade-strategy only-if-needed

Automatically setting up environment using Anaconda

On Apple Silicon (M1)

conda env create -f environment-m1.yml

Downloading YoloV4 data

• yolov4.cfg
• yolov4.weights
• coco.names

Add the files to the data folder with the following structure

project_root
├── data
│   ├── coco.names
│   ├── cfg
│   │   └── yolov4.cfg
│   └── weights
│       └── yolov4.weights    

Running the application

Before running the demos, please visit the config.py file if you want to change the default settings

The application contains 2 runnable demo scripts

run_locally.py

This will run the models locally on your machine using multiprocessing

PYTHONPATH=$PYTHONPATH:./src python ./src/run_locally.py

run_redis.py

This will run the models on a remote machine using redis pubsub

Before running the script, please make sure you have a redis server running on the default port 6379

Using redis-cli run the following command to increase the pubsub limits as follows hard 256mb, soft 128mb

redis-cli CONFIG SET client-output-buffer-limit "normal 0 0 0 slave 268435456 67108864 60 pubsub 268435456 134217728 60"

You may now run the distributed demo

PYTHONPATH=$PYTHONPATH:./src python ./src/run_redis.py

TODO

• refactor and clean up code
• handle startup and shutdown for the distributed components
• change from sending binary data to sending just bounding boxes and confidence scores when using redis
• add more models
• add more demos
• create docker images
• add GitHub workflows for building images