basegun-ml 2.0.0

A deeplearning package for the basegun weapon recongition app

Project description

Basegun_ml is a deeplearning package for the basegun weapon recongition app.

Installation

pip install basegun-ml

Usage

Classification

from basegun_ml.classification import get_typology, list_typologies
#After the import the model is already warmed-up for faster inference

#Convert image to bytes
with open("test.jpg", "rb") as file:
    image_bytes = file.read()

#Prediction of the weapon typology
typology,confidence_score,confidence_level=get_typology(image_bytes)

#Obtain the list of the different typologies
list_typologies()

Variables description

typology: it corresponds to the weapon class predicted from the image. The list of typologies can be obtained from the function

confidence_score: it corresponds to the confidence of the class prediction of the algorithm, the closer to 1 to more confident is the prediction

confidence_level: there are 3 level of confidence defined. According to this performance level the basegun user will have more information.

Measure length

from basegun_ml.measure import get_lengths

#Convert image to bytes
with open("test.jpg", "rb") as file:
    image_bytes = file.read()

#Get lengths
weapon_length,barrel_length,confidence_card=get_lengths(image_bytes)

Variables description

weapon_length: it corresponds to the weapon overall length predicted from the image.

barrel_length: it corresponds to the barrel length of the weapon predicted from the image.

confidence_card: it corresponds to the confidence score for the card prediction. A card is used as a reference for the measure module

If the gun is not detected, the exception MissingGun is raised

If the card is not detected, the exception MissingCard is raised

Alarm Model detection

from basegun_ml.ocr import is_alarm_weapon
#After the import the model is already warmed-up for faster inference

#Convert image to bytes
with open("test.jpg", "rb") as file:
    image_bytes = file.read()

#Prediction of the weapon typology
alarm_model = is_alarm_weapon(image_bytes, quality_check=True )

Variables description

alarm_model if the gun is one of the alarm model it returns "alarm weapon from model". If the gun has the PAK marking then alarm_model returns "alarm weapon PAK".

quality_check specify if the quality analysis is run before the text detection

If the image quality is too low, the exception LowQuality is raised

If no text is detected, the exception MissingText is raised

Tests

Tests are available for the classification task and the measure length task

pytest tests/test_classification.py 
pytest tests/test_measure.py
pytest tests/test_OCR.py

Credits

• This project uses the Ultralytics Library
• The oriented bounding box detection is inspired from this YOLOV5 implementation
• The image quality analysis uses Pyiqa
• The OCR tasks are computed using PaddleOCR