Model Card: Fine-Tuned InceptionV3 & Xception for Human Decomposition Image Classification

These CNN models were developed for the classification of human decomposition images into various stage of decay categories, including fresh, early decay, advanced decay, and skeletonized (Megyesi et al., 2005).

Model Details

Model Description

Developed by: Anna-Maria Nau
Funded by: National Institute of Justice
Model type: CNNs for Image Classification
Base Model: InceptionV3 and Xception pretrained on ImageNet
Transfer Learning Method: Two-step transfer learning: (1) freeze all pre-trained convolutional layers of the base model and train newly added classifier layers on custom dataset and (2) unfreeze all layers, and fine-tune model end-to-end on custom dataset.

Model Sources

Paper :
- Stage of Decay Estimation Exploiting Exogenous and Endogenous Image Attributes to Minimize Manual Labeling Efforts and Maximize Classification Performance
- Towards Automation of Human Stage of Decay Identification: An Artificial Intelligence Approach

Dataset

Dataset Name: Human Decomposition Dataset
Source: The dataset used in this study was obtained from the Forensic Anthropology Center (FAC) at the University of Tennessee, Knoxville, but due to privacy considerations, it is not available for public access. Please reach out to obtain access.
Classes: fresh (1), early decay (2), advanced decay (3), and skeletonized (4) based on Megyesi et al's scoring method.

Usage

The stage of decay classification is bodypart specific (i.e., head, torso, or limbs), so make sure to pick the correct bodypart model.

from tensorflow.keras.models import load_model
import numpy as np
from tensorflow.keras.preprocessing.image import img_to_array, load_img

# Load the entire model
model = load_model('path_to_your_model')  # e.g. head/inceptionV3 to perform stage of decay classfication of head images

# Load and preprocess an image
img = load_img('path_to_image.jpg', target_size=(299, 299))  # adjust size as per model input
img = img_to_array(img)  # convert to numpy array
img = np.expand_dims(img, axis=0)  # add batch dimension
img = img / 255.0  # normalize pixel values if needed

# Make predictions
predictions = model.predict(img)

# Use argmax to get the class label 
predicted_class = np.argmax(predictions, axis=1)