Mainak1792/VIBEID

VIBEID: A STRUCTURAL VIBRATION-BASED SOFT BIOMETRIC DATASET AND BENCHMARK FOR PERSON IDENTIFICATION

VIBeID: A Structural VIBration-based Soft Biometric Dataset and Benchmark for Person IDentification

This repository provides a script to download Pre-processed VIBeID datasets, create DataLoaders for training and testing, and train a ResNet-18 and ResNet-50 model using PyTorch.

Requirements

• Python 3.x
• pip (Python package installer)
• Kaggle API key (kaggle.json) [optional]

Arguments

--kaggle_dataset

• Description: Kaggle dataset identifier in the format mainakml/dataset-name.
• Type: str
• Default: 'mainakml/vibeid-a-4-1'
• Example: --kaggle_dataset yourusername/yourdataset
• Note: This argument specifies which dataset to download from Kaggle.

--output_dir

• Description: Directory to download and unzip the Kaggle dataset.
• Type: str
• Default: 'vibeid-a-4-1/VIBeID_A_4_1'
• Example: --output_dir /path/to/output_dir
• Note: The script will create this directory if it does not exist and will store the downloaded dataset here.

--batch_size

• Description: Batch size for the DataLoader.
• Type: int
• Default: 16
• Example: --batch_size 16
• Note: This determines the number of samples that will be propagated through the network at once.

--num_workers

• Description: Number of worker threads to use for loading the data.
• Type: int
• Default: 2
• Example: --num_workers 4
• Note: This is used to speed up data loading by using multiple threads.

--num_epochs

• Description: Number of epochs to train the model.
• Type: int
• Default: 50
• Example: --num_epochs 30
• Note: One epoch means that each sample in the dataset has had an opportunity to update the internal model parameters once.

--model

• Description: Model type to use for training.
• Type: str
• Choices: ['resnet18', 'resnet50']
• Default: resnet18
• Example: --model resnet50
• Note: Specifies which ResNet model architecture to use.

--num_classes

• Description: Number of output classes for the model.
• Type: int
• Default: 15
• Example: --num_classes 15/30/40/100
• Note: This should match the number of classes in your dataset.

--three_all

• Description: Fine tune last 3 layers or all layers
• Type: int
• Default: 0
• Example: --three_all 0/1
• Note: 0:Fine tune last 3 layers, 1: Fine tune all layers.

Step-by-Step guide

Convert Signal to CWT images

• Run spec_maker.py python spec_maker.py --file_path "A2_2_30p.mat" --notebook_path "folder_to_save_CWT _images"
• Run train test file python train_test.py --data_dir "CWT_image_folder_name" --output_dir "folder_to_save" --test_size 0.2

Person Identification using Deep learning

Quick Run

• Run Multi-class Classification (Single Image) - single_run_demo.ipynb
• Run Multi-class Classification (Multi Image)- multi_run_demo.ipynb
• Run Domain Adaptation demo - domain_adaptation_demo.ipynb

STEP 1: Install Libraries:

python install_libraries.py

STEP 2: Download the Datasets

You can download the datasets from the Kaggle (dataset is public)

1. vibeid-a1 A1
2. vibeid-a2 A2
3. vibeid-a3 A3
4. vibeid-a4 A4

OR run

python kaggle_dataset_download.py --kaggle_dataset "mainakml/dataset link"

Quick Run python kaggle_dataset_download.py --kaggle_dataset "mainakml/vibeid-a-4-1"

change the dataset link as your requirement

1. mainakml/vibeid-a1
2. mainakml/vibeid-a2
3. mainakml/vibeid-a3
4. mainakml/vibeid-a-4-1

STEP 3: Quick Run

python single_run.py --output_dir C:\Users\mainak\Documents\GitHub\VIBEID\VIBeID_A_4_1 --batch_size 16 --num_epochs 100 --model resnet18 --num_classes 15

STEP 4: Run dataset as per your requirement

single_image_run

python single_run.py --output_dir "add dataset link which contains train and test" --batch_size 16 --num_epochs 100 --model resnet18 --num_classes 15/30/40/100

multi_image_run

python multi_run.py --output_dir "add dataset link which contains train and test" --batch_size 16 --num_epochs 100 --model resnet18 --num_classes 15/30/40/100

Domain Adaptation using Deep learning

• Pretrained models are available in the folder
• change the path to the target test and val directory
• update three_all parameter python domain_run.py --model_path resnet_18_RGB_A3.1_100.pth --target_train_dir "add path to test"\test --target_test_dir "add path to val"\val --three_all 0/1

---