The facial_depth_estimation from liuguoyou

Facial Depth Estimation: 3D and 2D-Synthetic-Data---Group_C3I-Team---

Modeling of the Synthetic Dataset: Pixel-accurate facial depth generation process
Evaluating State-of-Art models for Single Image Depth Estimation
An Encoder-decoder based Facial Depth Estimation Model
Hybrid Loss Function

Single Frame Image Facial Depth Estimation

This is the project page for our research:

An Efficient Encoder-Decoder Model for Portrait Depth Estimation from Single Images trained on Pixel-Accurate Synthetic Data

High-Accuracy Facial Depth Models derived from 3D Synthetic Data
https://ieeexplore.ieee.org/document/9180166

Methodology for Building Synthetic Datasets with Virtual Humans
https://ieeexplore.ieee.org/abstract/document/9180188

Learning 3D Head Pose From Synthetic Data: A Semi-Supervised Approach
https://ieeexplore.ieee.org/abstract/document/9369299

Accurate 2D Facial Depth Models Derived from a 3D Synthetic Dataset https://ieeexplore.ieee.org/abstract/document/9427595

We will also update latest progress and available sources to this repository~

Note

This repository contains PyTorch implementations of FaceDepth, UNet-Simple, BTS, DenseDepth.

Dataset Generation Process

Setup & Requirements

To run this project, install it locally using pip install...:

$ pip install keras, Pillow, matplotlib, opencv-python, scikit-image, sklearn, pathlib, pandas, -U efficientnet,
$ pip install https://www.github.com/keras-team/keras-contrib.git, torch, torchvision

$ Python >= 3.6
Pytorch >= 1.6.0
Ubuntu 16.04
CUDA 9.2
cuDNN (if CUDA available)

Pretrained model

download the pre-trained model and keep in the FaceDepth directory:

https://nuigalwayie-my.sharepoint.com/:u:/g/personal/f_khan4_nuigalway_ie/EepkuVajAhdIjZoQm5Weyx4BjXcEZy-uw5OWxxMXq1WJPA?e=rv3aSY

Prepare Dataset for training & Testing

We prepared the dataset for training and testing
contact me on the following email: [email protected] for the complete dataset, I will provide the download link

Random sample frames with high-resolutions RGB images and their corresponding ground truth depth with differentvariations

Samples from the generated synthetic data with different variation of head pose

Live Demo

We attach live demo implementation for Pytorch.
Note change the model check point name if different

$ cd FaceDepth
$ python live_demo.py

Testing

First make sure that you have some images (RGB and depth) or you can use our test images: (rgb_syn_test) and (gt_syn_test).

$ cd FaceDepth
$ make a folder and keep your RGB images in that  
$ make a folder and keep your gt depth images in that
$ Change the path in the Facedepth_test.py
$ Name a folder that you want to save the predicted results (images)

Once the preparation steps completed, you can test using following commands.

$ cd FaceDepth
$ python Facedepth_test.py

Training

Once the dataset download process completed, please make sure unzip all the data into a new folder and follow the following steps:

$ cd FaceDepth
$ Download the .csv or .txt files
$ Change the paths in the train.py

Once the preparation steps completed, you can train using following commands.

$ cd FaceDepth
$ python train.py --batch_size 4 --epochs 25

Training BTS and DenseDepth methods for our synthetic facial depth dataset.

We referred to BTS and DenseDepth:

$ https://github.com/cogaplex-bts/bts

$ https://github.com/ialhashim/DenseDepth

Training UNet_sample on synthetic facial depth dataset.

We referred to Facial_depth_estimation:

$ python train.py --batch_size 6 --epochs 25

Results

Properties of the studied methods

Qualitative results of a facial monocular depth estimation methods

Citation

If you find this work useful for your research, please consider citing our paper:

@INPROCEEDINGS{9427595,
  author={Khan, Faisal and Basak, Shubhajit and Corcoran, Peter},
  booktitle={2021 IEEE International Conference on Consumer Electronics (ICCE)}, 
  title={Accurate 2D Facial Depth Models Derived from a 3D Synthetic Dataset}, 
  year={2021},
  volume={},
  number={},
  pages={1-6},
  doi={10.1109/ICCE50685.2021.9427595}}
  
  @INPROCEEDINGS{9180166,
  author={Khan, Faisal and Basak, Shubhajit and Javidnia, Hossein and Schukat, Michael and Corcoran, Peter},
  booktitle={2020 31st Irish Signals and Systems Conference (ISSC)}, 
  title={High-Accuracy Facial Depth Models derived from 3D Synthetic Data}, 
  year={2020},
  volume={},
  number={},
  pages={1-5},
  doi={10.1109/ISSC49989.2020.9180166}}
  
  @ARTICLE{9369299,
  author={Basak, Shubhajit and Corcoran, Peter and Khan, Faisal and Mcdonnell, Rachel and Schukat, Michael},
  journal={IEEE Access}, 
  title={Learning 3D Head Pose From Synthetic Data: A Semi-Supervised Approach}, 
  year={2021},
  volume={9},
  number={},
  pages={37557-37573},
  doi={10.1109/ACCESS.2021.3063884}}
  
  @INPROCEEDINGS{9180188,
  author={Basak, Shubhajit and Javidnia, Hossein and Khan, Faisal and McDonnell, Rachel and Schukat, Michael},
  booktitle={2020 31st Irish Signals and Systems Conference (ISSC)}, 
  title={Methodology for Building Synthetic Datasets with Virtual Humans}, 
  year={2020},
  volume={},
  number={},
  pages={1-6},
  doi={10.1109/ISSC49989.2020.9180188}}

liuguoyou / facial_depth_estimation Goto Github PK

facial_depth_estimation's Introduction