Computer Vision

We use deep learning and other state-of-the-art algorithms to gain valuable insights from images and videos as well as to manipulate and generate new data using GAN networks. Our custom trained neural networks can also be optimized for reducing hardware costs in cloud or edge computing (model distillation).
CAF logo
P&G logo
Classification
orange circle
Driver safety applications
orange circle
Brand detection
Object detection and tracking
orange circle
In-store retail analytics
orange circle
Security applications
Image segmentation
orange circle
Industrial fault detection
orange circle
Robotic surgery
Image generation and manipulation
orange circle
Manipulate facial features
orange circle
Generate anonymized datasets
orange circle
Deepfakes
Human pose detection and identity validation
orange circle
Identity validation
orange circle
Workout monitoring
orange circle
Elderly care
Related work
construction
Mining driver safety
Safety is a top priority in mining operations and all precautions must be taken in order to avoid accidents. Distractions while driving can be a big risk, especially when conducting a 24/7 operation. The client therefore wanted to detect and be able to take immediate action if drivers were eating, using the phone or just not following best practices while driving, such as not having both hands on the steering wheel.
We designed and implemented a driver safety application for mining vehicles that operate on NVIDIA Jetson Edge devices using Microsoft’s IoT Hub. We detect when a driver is eating, driving without both hands, etc using computer vision techniques and send the corresponding alerts through an onsite private 4G network. Technologies leveraged include Python, Tensorflow, Docker, NVIDIA Jetson, React, Azure, Microsoft IoT Hub, Azure DevOps.
Ageing Generative Adversarial Networks
Our client needed to generate images changing a person's age but preserving identity, similar to FakeApp. These applications are very useful when running health prevention campaigns to show visually the effects different lifestyles can have on a person.
younger age
Younger
real age
Real
older age
Older
Based on the work of companies like NVIDIA, we implemented an ageing system for human facial photos. It is based on generative adversarial networks utilizing state of the art papers and techniques. Our system receives the photo of a person and alters the person’s age while preserving their unique visual identity. We solved challenges such as developing the production deployment of a scalable solution, optimized for minimum processing time yet while minimizing cloud costs for our client.
Technologies leveraged include Python, Tensorflow, Pytorch, Docker, AWS.
Some of the architectures we've worked with
StyleGAN
StyleGAN
StyleGAN is a powerful NVIDIA architecture that provides state-of-the-art results in image generation. We have experience using it to generate both deepfakes and synthetic datasets.
AlphaPose
AlphaPose
When it is necessary to estimate the pose of multiple people, AlphaPose is an excellent alternative. We have experience working with different types of applications and use cases.
Pix2pix
Pix2pix
We have experience training Pix2Pix models to perform image transformations, for example on faces, obtaining realistic and very good quality results.
GAN
alphapose
pix2pix
Shape
Get in touch with one of our specialists. Let's discover how can we help you.
Training, developing and delivering machine learning models into production
Got a project?
Let’s talk