raminrahimi6970/Pytorch_Retinaface

Retinaface get 80.99% in widerface hard val using mobilenet0.25.

Running Face Detection in Google Colab

Original Repository

This repository is a fork from Pytorch_Retinaface repository and created to show how you can run your machine learning projects during Qarantine on Google GPU machine if you have no access to local gpu machine.

Medium Article supporting content

How Google Colab helped me to speed up my video processing tasks while working in remote and had no GPU.

Requirements

Dependencies

Dependency	Version	checked
python	= 3.x	✅
pytorch	> 0.21.3	✅
opencv-python	> 3.4	✅
python-dotenv	x	✅

Operation System

• Ubuntu 16.04 (we use colab to run our program)

• GPU enabled

• cuda installed 10.1 (9.2 not sure)

How to run it

1- Get pretrained weights from here and put it in ./weights/ directory

2- upload the video or image in /upload/ directory

4- run the algorithm using this command

python  --input < file address > 

        --output-dir < output directory >
        
        --cpu <True/False>  # force cpu computation
        
        --keep_top_k <integer> # How many detection to keep
        
        --confidence_threshold <float between [0-1]> # confidence threshold
        
        --network <resnet50 or mobile0.25> # network for detection of faces
        
        --trained_model <path to trained model>