filipsPL / tf_consensus_score

Calculating consensus scores from multiple TensorFlow Lite classificaitons

tf_consensus_score

This module contains a collection of functions for calculating consensus scores from multiple TensorFlow Lite classificaitons.

┌─────────────┐
│             │
│  Image      │       ┌─────────────────────────────────┐      ┌──────────────────────┐
│             ├──────►│                                 │      │                      │
└─────────────┘       │                                 │      │                      │
                      │                                 │      │ Output:              │
┌─────────────┐       │      tf_consensus_score.py      ├─────►│                      │
│tfl model 1  ├──────►│                                 │      │ cat 0.9255041480     │
├─────────────┤       │                                 │      │                      │
│tfl model 2  │       │                                 │      │                      │
├─────────────┤       └─────────────────────────────────┘      └──────────────────────┘
│tfl model n  │
└─────────────┘

Usage:

import cv2
from tf_consensus_score import *

# Initialize a dictionary for local model files
model_files = {}

# Add model files to the dictionary
model_files = [
    "animals1.tflite",
    "animals2.tflite"
]

# Define the path where models are located
models_path = "models/"

# Read the image the cv2 way
image = cv2.imread("cat.jpg", cv2.IMREAD_COLOR)

# Calc consensus score and return the highest scored class
category_name, score = calc_consensus(image, model_files, models_path)
print(category_name, score)

# cat 0.9255041480

Installation

Just download tf_consensus_score.py and put in the program directory.

Functions

calc_probabilities_for_image(image, local_model_file)

Calculate probabilities for a given image using a specific model file.

Arguments

• image: The input image to be classified.
• local_model_file: The path to the local TensorFlow Lite model file.

Returns

A list of classification categories and their corresponding probabilities.

calculate_consensus_scores(ProbabilitiesForImage)

Calculate consensus scores based on a list of probabilities for multiple models.

Arguments

• ProbabilitiesForImage: A list of probabilities for each model.

Returns

A list of consensus scores for each category based on the input probabilities.

return_best_consensus_category(consensus)

Find the category with the highest consensus score.

Arguments

• consensus: A list of consensus scores for each category.

Returns

A tuple containing the best category name and its score.

calc_probabilities_for_all_models(image, local_model_files, models_path)

Calculate probabilities for all models in a list.

Arguments

• image: The input image to be classified.
• models_path: The path where model files are located.
• local_model_files: A list of model file names for a specific category.

Returns

A list of probabilities for each model in the input list.

calc_consensus(image, local_model_files, models_path)

Calculate the consensus category and score for a given image.

Arguments

• image: The input image to be classified.
• models_path: The path where model files are located.
• local_model_files: A list of model file names for a specific category.

Returns

A tuple containing the consensus category name and its score based on the input probabilities.