ForensicFit is created to preprocess scanned images from different tears and generate a database that will be used in different machine-learning approaches. This package will prepare the data in four different approaches, which will be shown in the tutorial sections. For more details please see Lang, Logan, Pedram Tavadze, Meghan Prusinowski, Zachary Andrews, Cedric Neumann, Tatiana Trejos, and Aldo H. Romero. "Using convolutional neural networks to support examiners in duct tape physical fit comparisons." Forensic Science International 353 (2023): 111884, at https://doi.org/10.1016/j.forsciint.2023.111884. For more detailed documentation please visit https://romerogroup.github.io/ForensicFit/. ForensicFit can be installed from PyPI https://pypi.org/project/forensicfit/.

To access the trined models please follow the following steps:

1. Create a virtual envronment python -m venv venv

   Activate the environment

   For ubunutu

   source venv/bin/activate

   For windows

   venv\Scripts\activate.bat

2. Clone repo git clone git@github.com:romerogroup/ForensicFit.git

3. Go into repo cd ForensicFit

4. After cloning the repo. Install the ML Requirements pip install -r requirements_ml.txt pip install .

5. Download the processed data from into ForensicFit/data/processed.

6. Download the the pretrained models: (This must be executed from the ForensicFit directory)

from forensicfit.machine_learning.download_models import download_models
download_models()

6. If training from scratch. Call the following function (This must be executed from the ForensicFit directory)

from forensicfit.machine_learning.trainer_tensorflow import train
train()

7. Evaluate the dataset. (This must be executed from the ForensicFit directory)

from forensicfit.machine_learning.evaluate import evaluate

evaluate()

8. If you want to predict on a single image do the following: (This must be executed from the ForensicFit directory)

import os
from forensicfit.machine_learning.predict import predict
processed_dir=os.path.join("data", "processed", "normal_split", "match_nonmatch_ratio_0.3")
processed_back_dir = os.path.join(processed_dir,'back','test','match')
processed_front_dir = os.path.join(processed_dir,'front','test','match')
image_front_path=os.path.join(processed_front_dir,'HQC_001_R-HQC_003_L.jpg')
image_back_path=os.path.join(processed_back_dir,'HQC_002_L-HQC_004_R.jpg')
prediction=ml.predict(image_front_path=image_front_path,image_back_path=image_back_path)
print(prediction)