New version with yolov5: https://gitlab.com/odk/odk-frame-analyzer

Urban object detection using PyTorch and YoloV3

For more information, look at this medium post.

PyTorch implementation of an urban object detection model. This repository contains all code for predicting/detecting and evaulating the model.

This repository combines elements from:

• https://github.com/zhaoyanglijoey/yolov3
• https://github.com/ultralytics/yolov3

Installation

To install all required libaries:

pip install -r requirements.txt

Predictions

Several different weights and configs are available at: https://drive.google.com/open?id=1DjeNxdaF7AW3Nu54_3oRw_1SeYJtOvNL. Some also have the testing data.

Pre trained weights

Name	Classes	Test data
3 classes	cardboard, garbage_bags and containers	Yes
cigarettes	cigarette	Yes
12 classes	container_small, garbage_bag, cardboard, matras, christmas_tree, graffiti, pole, face_privacy_filter and license_plate_privacy_filter, construction_toilet, construction_container, construction_shed	No

Run predictions

To run predictions, download the cfg and weights from https://drive.google.com/open?id=1DjeNxdaF7AW3Nu54_3oRw_1SeYJtOvNL and put them in the correct folders.

Then for example run the following the make a prediction on a file using CPU:

python detector_garb.py -i samples/input5_frame11.jpg -o output

Or to realtime detect on your webcam using GPU: (CUDA must be installed)

python detector_garb.py -i 0 --webcam --video -o ./webcam_output/ --cuda

Docker

To run code in docker

docker-compose build
docker-compose up

Test

For testing download data from: https://drive.google.com/drive/folders/1DjeNxdaF7AW3Nu54_3oRw_1SeYJtOvNL

The garbage bags, containers and cardboard dataset contains 804 images and label files. A smaller dataset with annotations of cigarettes is also available.

To run test execute the following code:

python test.py

Class	Images	Targets	P	R	mAP	F1
all	115	579	0.242	0.941	0.875	0.376
container_small	115	180	0.38	0.989	0.979	0.549
garbage_bag	115	223	0.212	0.964	0.875	0.348
cardboard	115	176	0.122	0.869	0.77	0.231

The model with 12 classes has been trained on a larger collection. The test results are below.

Class	Images	Targets	P	R	mAP	F1
all	179	706	0.263	0.873	0.811	0.392
container_small	179	142	0.521	0.972	0.97	0.678
garbage_bag	179	114	0.266	0.965	0.936	0.417
cardboard	179	78	0.132	0.962	0.89	0.232
matras	179	8	0.467	0.875	0.875	0.609
kerstboom	179	10	0.278	1	1	0.435
graffiti	179	73	0.185	0.932	0.885	0.308
amsterdammertje	179	52	0.325	0.942	0.911	0.483
face_privacy_filter	179	87	0.139	0.782	0.599	0.237
license_plate_privacy_filter	179	103	0.186	0.845	0.713	0.304
construction_toilet	179	7	0.211	0.571	0.524	0.308
construction_container	179	21	0.173	0.905	0.842	0.29

Training

For training a new model look at:

https://github.com/maartensukel/yolov3-garbage-object-detection-training

This is the training loss of 1600 images with 12 classes: