loginabhay/yolo_model_evaluator

Evaluates the yolo model and creates metrices required for detection model evaluation

YOLO Model Evaluator

Effortlessly run object detection on CPU using only OpenCV with this YOLO model evaluation tool. This tool is designed to evaluate the performance of YOLO models by generating essential metrics, including confusion matrix, average precision, recall, and mean average precision (mAP). It provides a comprehensive analysis of your detection model's accuracy. Additionally, the tool offers the capability to generate annotations from the detection model and save them as text files in YOLO format. Simplify your model evaluation process with this straightforward and efficient solution.

Installation:

1. Clone the repo

https://github.com/loginabhay/YOLO_Model_Evaluator.git
cd YOLO_Model_Evaluator

2. Setup a Virtual Environment

This is assuming you have navigated to the YOLO_Model_Evaluator root after cloning it.

NOTE: This is tested under python3.10. For other python versions, you might encounter version conflicts.

# install required packages from pypi
python3 -m venv <your env name>
source <your env name>/bin/activate

3. Intall Requirement to Run the Project

pip install -r requirements.txt

4. Running the Project

• The project is divided in two major
  • Generating the prediction or annotation from the given images
  • Comparing the prediction with the ground truth in order to evaluate the metrices

4.1 Generating Prediction

Set of arguments used are given below:

Argument	Description	Flag
-i	Path to the input image folder	Required
-o	Path to the output folder	Default='./image_predictions'
-c	Path to YOLO Model CFG file	Required
-m	Path to YOLO Model weight file	Required
-conf	Change for the confidence value	Default=0.3
-n	Change for the network input shape	Default=416
-nms	Change for the nms value	Default=0.5
-ca	Creates annotations instead of predictions	Default=False

Example command

python3 yolo_img_infer_write_prediction.py -i <path to input folder>
                    -c <path to input cfg> -m <path to input weights>

4.2 Evaluating Metrices

Set of arguments used are given below:

Argument	Description	Flag
-ip	Path to the input prediction folder	Required
-o	Path to the output folder	Default='./generated_metrices'
-ig	Path to the ground truth folder	Required
-s	Save the generated metrices in the csv format	Default=True
-conf	Change for the confidence value	Default=0.3
-nm	Path to the classes names file	Required
-nms	Change for the nms value	Default=0.5

Example command

python3 Evaluation_matrix_yolo.py -ip <path to input prediction folder>
                    -ig <path to input ground truth folder> -nm <path to classes names>