QueueTime

Finds the length of lines of people in images - Grinnell CSC 262 Computer Vision Final Project

Setup

Installing locally with pipenv (requires python 3.7) - required for easy gui use

1. Install pipenv:

pip install --user pipenv

2. Get the COCO api ready:

git submodule init
git submodule update
./gen_coco_3.sh

3. Install dependencies:

pipenv install

4. Download the coco dataset:

pipenv run ./COCO_download.sh

5. Run pull up the virtualenv:

pipenv shell

Installing with Docker (experimental)

1. Install Docker
2. Grab the coco api

git submodule init
git submodule update

3. Run ./docker.sh download to download the coco dataset

4. Run ./docker.sh shell for a bash instance in the right virtualenv

Usage:

Most executable python files have a help option available with python3 $file --help. This will list the available arguments and what they do.

Neural network portion:

Downloading images from coco:

cd src
python3 download.py [num_imgs]

Training:

cd src
python3 train.py -m $model_file -o 6000 -i 2000 -e 40 -b 10 -l 0.0005 -r True

Classifying:

IMAGE_ID should be one of the downloaded images:

cd src
python3 classify.py -m $model_file IMAGE_ID

mAP scoring:

Generate mAP ground truth data:

Note: This file does not have help

cd src
python3 mAP_formatting.py IMAGE_ID IMAGE_ID IMAGE_ID IMAGE_ID ...

Generate model data:

Should be the same set of image ids as above

cd src
python3 classify.py -m $file -f -p IMAGE_ID IMAGE_ID IMAGE_ID IMAGE_ID ...

mAP scoring test:

cd mAP
python3 main.py

Queue Classification

Note: the queue-classification folder has a different set of dependencies than the rest of the project. As such, if you are in the larger python virtualenv when you enter the folder, you should run the exit command before cding in, and then run pipenv shell in the queue-classification folder.

Generate labels with pre-trained mask-rcnn

cd queue-classification
pipenv install
pipenv shell
python3 gen_labels.py -v $video_file -o "$video-file".json

Display video

cd queue-classification
python3 queue_classification.py -v $video-file -a "$video-file".json