Skip to content

Getting Started

This section provides a guide on how to install and use OTValidate.



Installation instructions for Windows are following soon.

Install OTValidate by cloning the repository with git:

git clone
cd OTValidate
pip install -r requirements.txt
pip install .

Or install by using the Makefile:

git clone
cd OTValidate
make install

The installation for machines running on Apple's M1 chip is not as straightforward. There are two ways to install OTValidate on an M1 Mac. As a prerequisite the package manager Homebrew is required.

  1. By executing these commands in the following order:

    git clone
    cd OTValidate
    brew install openblas
    OPENBLAS=$(brew --prefix openblas) CFLAGS="-falign-functions=8 ${CFLAGS}" pip install scipy==1.7.2
    pip install -r requirements.txt
    pip install .
  2. By using the Makefile

    git clone
    cd OTValidate
    make install_m1


Image Annotation Data

The folder containing the ground truth annotations of the images need to be in the YOLO format:

│   obj.names
│   train.txt
    │   frame_01.png
    │   frame_01.txt
    │   frame_02.png
    │   frame_02.txt
    │   ...


Analyse Object Detection Performance

Quickstart Guide

from OTValidate  import evaluate_detection_performance

# path to the directory containing the annotated dataset in otdet format
gt_data = "path/to/data1"

# model weights
model1 = "path/to/"
model2 = "path/to/"
model3 = "path/to/"

Use the evaluate_detection_performance function to calculate a set of object detection metrics of the respective models:

    path_to_model_weights=[model1, model2, model3],
    is_gt_xyxy_format=False, # whether the ground truth's bounding box is in xyxy or xywh format


The evaluation results of the models will be saved in the directories containing the annotation data. An out directory containing all the results will be created there.