Application

This repository is for testing and comparing different approaches to social distancing.

Requirements

Clone repository with submodules (git clone --recurse-submodules ...).

• Cuda 10.1
• gcc min 7

Install the submodules in editable mode

conda create -n socdist-env python=3.7
conda activate socdist-env

conda install pytorch==1.4.0 torchvision==0.5.0 cudatoolkit=10.1 -c pytorch

pip install -e monoculardepth/monodepth2
pip install -e monoculardepth/mannequinchallenge

pip install -e object-detection-segmentation/yolact 
pip install -e git+https://github.com/CharlesShang/DCNv2@master#egg=dcnv2

pip install -e tracking_wo_bnw

pip install -e human_depth_dataset

pip install -r requirements.txt

Retrieve checkpoint for mannequinchallenge

cd monoculardepth/mannequinchallenge && ./fetch_checkpoints.sh && cd ../..

Download https://vision.in.tum.de/webshare/u/meinhard/tracking_wo_bnw-output_v2.zip and unzip into tracking_wo_bnw/output.

Usage

python run.py --video_source samples/mot16.webm --depth_merger median

API Usage

uvicorn rest:app --reload
curl --location --request POST 'localhost:8000/predict' --form 'file=@samples/people_002.png'