File Fragment Classification

This repo is official implementation of the paper "Light-Weight File Fragments Classification Using Depthwise Separable Convolutions". The Datasets used in this paper can be accessed through the following link FFT-75.

Training

First, download all the dataset, decompress the folders and put them in dataset folder. data/<dataset folders>

RUN Script

Run this script for inference time

!python inference_time.py --data_folder ./data/ --results_folder ./results/

Train

python train.py --data_folder ./data/ --results_folder ./results/ --scenario 4 --size 4096 --batch_size 128 --lr 0.002

This command will start a training session using the data under the data/size_scenario directory. See ./train --help for a description of command line arguments.

The ressults are saved under the results/scenario/size directory.

Evaluation

python eval.py --data_folder ./data/ --results_folder ./results/ --scenario 0

This command will load previously trained model for each scenario [1, 6] and evaluate the model on the validation dataset.

It will then generate confusion matrix images and store them in eps format under the results folder.

The results directory will be structured as following:

├── results
   |   ├── 1
   |   |   ├── 4096
   |   |   └── 512
   |   |
   |   ├── 2
   |   |   ├── 4096
   |   |   └── 512
   |   |
   |   |── 3
   |   |   ├── 4096
   |   |   └── 512
   |   |
   |   ├── 4
   |   |   ├── 4096
   |   |   └── 512
   |   |
   |   ├── 5
   |   |   ├── 4096
   |   |   └── 512
   |   |
   |   ├── 6
   |       ├── 4096
   |       └── 512

2. Requirements

python>=3.6
torch>=1.5.0
torchvision
torchsummary
numpy
scipy
matplotlib