A TensorFlow implementation of the Mnemonic Descent Method.

Last update: Oct 07, 2022

Overview

MDM

A Tensorflow implementation of the Mnemonic Descent Method.

Mnemonic Descent Method: A recurrent process applied for end-to-end face alignment
G. Trigeorgis, P. Snape, M. A. Nicolaou, E. Antonakos, S. Zafeiriou.
Proceedings of IEEE International Conference on Computer Vision & Pattern Recognition (CVPR'16).
Las Vegas, NV, USA, June 2016.

Installation Instructions

Menpo

We are an avid supporter of the Menpo project (http://www.menpo.org/) which we use in various ways throughout the implementation.

Please look at the installation instructions at:

http://www.menpo.org/installation/

TensorFlow

Follow the installation instructions of Tensorflow at and install it inside the conda enviroment you have created

https://www.tensorflow.org/versions/r0.9/get_started/os_setup.html#installing-from-sources

but use

git clone https://github.com/trigeorgis/tensorflow.git

as the TensorFlow repo. This is a fork of Tensorflow (#ff75787c) but it includes some extra C++ ops, such as for the extraction of patches around the landmarks.

Pretrained models

Disclaimer: The pretrained models can only be used for non-commercial academic purposes.

A pretrained model on 300W train set can be found at: https://www.doc.ic.ac.uk/~gt108/theano_mdm.pb

Training a model

Currently the TensorFlow implementation does not contain the same data augmnetation steps as we did in the paper, but this will be updated shortly.

    # Activate the conda environment where tf/menpo resides.
    source activate menpo
    
    # Start training
    python mdm_train.py --datasets='databases/lfpw/trainset/*.png:databases/afw/*.jpg:databases/helen/trainset/*.jpg'
    
    # Track the train process and evaluate the current checkpoint against the validation set
    python mdm_eval.py --dataset_path="./databases/ibug/*.jpg" --num_examples=135 --eval_dir=ckpt/eval_ibug  --device='/cpu:0' --checkpoint_dir=$PWD/ckpt/train
    
    python mdm_eval.py --dataset_path="./databases/lfpw/testset/*.png" --num_examples=300 --eval_dir=ckpt/eval_lfpw  --device='/cpu:0' --checkpoint_dir=$PWD/ckpt/train
    
    python mdm_eval.py --dataset_path="./databases/helen/testset/*.jpg" --num_examples=330 --eval_dir=ckpt/eval_helen  --device='/cpu:0' --checkpoint_dir=$PWD/ckpt/train
    
    # Run tensorboard to visualise the results
    tensorboard --logdir==$PWD/ckpt

A TensorFlow implementation of the Mnemonic Descent Method.

Related tags

Overview

MDM

Installation Instructions

Menpo

TensorFlow

Pretrained models

Training a model

Owner

A sample pytorch Implementation of ACL 2021 research paper "Learning Span-Level Interactions for Aspect Sentiment Triplet Extraction".

Video Representation Learning by Recognizing Temporal Transformations. In ECCV, 2020.

Code for reproducing key results in the paper "InfoGAN: Interpretable Representation Learning by Information Maximizing Generative Adversarial Nets"

RM Operation can equivalently convert ResNet to VGG, which is better for pruning; and can help RepVGG perform better when the depth is large.

Ascend your Jupyter Notebook usage

A quantum game modeling of pandemic (QHack 2022)

Implementation of "Debiasing Item-to-Item Recommendations With Small Annotated Datasets" (RecSys '20)

Source code for "Taming Visually Guided Sound Generation" (Oral at the BMVC 2021)

Approaches to modeling terrain and maps in python

My Body is a Cage: the Role of Morphology in Graph-Based Incompatible Control

Notebook and code to synthesize complex and highly dimensional datasets using Gretel APIs.

Official implementation of the PICASO: Permutation-Invariant Cascaded Attentional Set Operator

Code for KHGT model, AAAI2021

交互式标注软件，暂定名 iann

Python wrapper of LSODA (solving ODEs) which can be called from within numba functions.

Code for the paper "Offline Reinforcement Learning as One Big Sequence Modeling Problem"

Repo for code associated with Modeling the Mitral Valve.

Implementation of PyTorch-based multi-task pre-trained models

Code and real data for the paper "Counterfactual Temporal Point Processes", available at arXiv.

Large Scale Multi-Illuminant (LSMI) Dataset for Developing White Balance Algorithm under Mixed Illumination