Pose Transformers: Human Motion Prediction with Non-Autoregressive Transformers

Angel Martínez-González, Michael Villamizar, Jean-Marc Odobez

IEEE/CVF International Conference on Computer Vision, 2021

paper · project page · poster

A new non-autoregressive Transformer architecture to predict 3D human motion and activities.

An Efficient Image-to-Image Translation HourGlass-based Architecture for Object Pushing Policy Learning

Marco Ewerton, Angel Martínez-González, Jean-Marc Odobez

IEEE/RSJ International Conference on Intelligent Robots and Systems, 2021

paper · project page

An object pushing policy based using an Hourglass-based CNN architecture.

Residual Pose: A Decoupled Approach for Depth-Based 3D Human Pose Estimation

Angel Martínez-González, Michael Villamizar, Olivier Canévet, Jean-Marc Odobez

IEEE/RSJ International Conference on Intelligent Robots and Systems, 2020

paper · project page · video

A new 3D pose estimation method that relies in residual pose modeling suitable for multi-person scenarios.

WatchNet++: Efficient and Accurate Depth-Based Network for Detecting People Attacks and Intrusion

Michael Villamizar, Angel Martínez-González, Olivier Canévet, Jean-Marc Odobez

Machine Vision and Applications (MVAP), 2020

paper · video · BibTex

Detecting people from top-view images for attack detection in security airlocks.

Efficient Convolutional Neural Networks for Depth-Based Multi-Person Pose Estimation

Angel Martínez-González, Michael Villamizar, Olivier Canévet, Jean-Marc Odobez

IEEE Transactions on Circuits and Systems for Video Technology, 2020 (accepted Nov. 2019)

paper · data · project page

Leveraging lightweight CNN, domain adaptation and knowledge distillation for 2D pose estimation.

Investigating Depth Domain Adaptation for Efficient Human Pose Estimation

Angel Martínez-González, Michael Villamizar, Olivier Canévet, Jean-Marc Odobez

ECCV Workshop on Human Behavior Understanding, 2018

paper · data

Investigates domain adaptation to close the covariance shift gap from learning with synthetic data.

WatchNet: Efficient and Depth-based Network for People Detection in Video Surveillance Systems

Michael Villamizar, Angel Martínez-González, Olivier Canévet, Jean-Marc Odobez

IEEE International Conference on Advanced Video and Signal-Based Processing (AVSS), 2018

paper

People detection from top-to-down facing images for security systems.

Real-time Convolutional Networks for Depth-based Human Pose Estimation

Angel Martínez-González, Michael Villamizar, Olivier Canévet, Jean-Marc Odobez

IEEE/RSJ International Conference on Intelligent Robots and Systems, 2018

paper

Introducing a new CNN architecture for real time pose estimation from synthetic depth images.

Real Time Face Detection Using Neural Networks

Angel Martínez-González, Victor Ayala Ramirez

IEEE Mexican International Conference on Artificial Intelligence, 2011

paper

Skin color-based searching and face detection with neural networks.

DepthHuman: A tool for depth image synthesis for human pose estimation

We have created the DIH dataset, a large scale dataset of synthetic depth images with annotations for depth-based 2D pose estimation with this tool.

code · dataset

ViZDoom and Reinforcement Learning

Solving navigation tasks in a 3D FPS game environment for autonomous agents with deep reinforcement learning methods.

project page

Montecarlo Localization

When an autonomous drone needs to localize itself in a map, motion models and particle filters come to place to save the day.

code

3D object culling methods for virtual reality gaming (collaboration with GCS)

GameCoder Studios (GCS)