LiteHPE: Head Pose Estimation

Features

• 6D Rotation Matrix Representation
• High Performance.
• Easy Integration
• Customizability

*** The project is structured into four main files:**

• main.py: The entry point of the project, orchestrating the training, evaluation, and prediction processes.
• nets.py: Contains the definitions of neural network models or architectures.
• datasets.py: Manages dataset handling, including loading, preprocessing, and augmentations.
• util.py: Provides utility functions for data manipulation, visualization, logging, and other support tasks.

Performance Metrics

The model achieved the following Mean Absolute Error (MAE) metrics across different pose angles:

Results

Backbone	Epochs	Pitch	Yaw	Roll	Params (M)	FLOPS (M)	Pretrained weights
MobileOne_s0	120	5.4	4.2	3.8	2.1	275	model
MobileOne_s1	120	5.2	4.0	3.7	4.8	825	model
MobileOne_s2	120	4.9	3.7	3.4	7.8	1299	model
MobileOne_s3	120	4.7	3.6	3.3	10.1	1896	model
MobileOne_s4	120	4.6	3.5	3.2	14.8	2978	model

Installation

1. Clone the repository
2. Create a Conda environment using the environment.yml file:

conda env create -f environment.yml

3. Activate the Conda environment:

conda activate HPE

Preparing the Dataset

1. Download the 300W-LP, AFLW2000 Datasets:
2. Download the dataset from the official project page.
3. Place the downloaded dataset into a directory named 'Datasets'.

Training the Model

To initiate the training process, use the following command:

• Configure your dataset path in main.py for training
• Configure Model name (default A2) in main.py for training
• Run the below command for Single-GPU training

python main.py --train

• Run the below command for Multi-GPU training $ is number of GPUs

bash main.sh $ --train

Testing the Model

Configure your dataset path in main.py for testing Run the below command:

bash main.sh $ --test

Inference

• Configure your video path in main.py for visualizing the demo
• Run the below command:

python main.py --demo