Optimizing Camera Configurations for Multi-View Pedestrian Detection

Overview

Contents

• Dependencies
• Data Preparation
• Usage

Dependencies

0. Create python environment with python=3.8

   conda create -n mvconfig python=3.8
   conda activate mvconfig

1. Please install PyTorch with CUDA support with

   pip3 install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu118

2. Then, install all other dependencies via

   pip install -r requirements.txt

   and then install lap separately via

   conda install -c conda-forge lap=0.4.0

3. Verify that Carla works through docker container

   docker run --privileged --gpus 1 --net=host -e DISPLAY=$DISPLAY carlasim/carla:0.9.14 /bin/bash ./CarlaUE4.sh

Data Preparation

Running the MVconfig code requires the CARLA simulator. We recommend using the CARLA Docker image.

1. To install the Docker container, please refer to the Docker Engine installation guide and the NVIDIA Container Toolkit installation guide.

2. The CARLA Docker image is available on Docker Hub, which can be pulled using the following command

   docker pull carlasim/carla:0.9.14

Usage

In order to train or test the detection and tracking model, as well as the camera control module, please follow the instructions below.

1. Train model with default configuration

   CUDA_VISIBLE_DEVICES=0 python main.py -d carlax --reID --carla_gpu 0 --carla_cfg [cfg_name] --record_loss --carla_port 2000 --carla_tm_port 8000

2. Train model with three human expert configurations

   CUDA_VISIBLE_DEVICES=0 python main.py -d carlax --reID --carla_gpu 0 --carla_cfg [cfg_name]_[1/2/3] --record_loss --carla_port 2000 --carla_tm_port 8000

3. Train model with interactive cameras and joint training

   CUDA_VISIBLE_DEVICES=0 python main.py -d carlax --reID --interactive --carla_gpu 0 --carla_cfg [cfg_name] --epochs 50 --joint_training 1 --record_loss --carla_port 2000 --carla_tm_port 8000

4. Evaluate the trained model (with or without interactive cameras)

   CUDA_VISIBLE_DEVICES=0 python main.py -d carlax --reID [--interactive] --carla_gpu 0 --carla_cfg [cfg_name] --carla_port 2000 --carla_tm_port 8000 --eval --resume [log_path]