Hans Gaiser
Computer Vision Engineer at The Robot Engineers
Delft, South Holland, Netherlands
Summary
🤩
Rockstar🎓
Top SchoolHans Gaiser is a Computer Vision Engineer based in Delft with 13 years of experience building production-grade perception systems for robotics and automation. He has driven vision solutions at Fizyr—specializing in automated picking—and worked concurrently as a robot vision engineer and freelancer, showing a practical focus on deployable systems. Hans contributes to prominent open-source projects such as keras-retinanet and Keras itself, where he improved object detection anchors, focal loss integration, and augmentation determinism—work that bridges research models and real-world training pipelines. His background includes leading robot software frameworks and hands-on roles in industrial robotics, giving him a rare blend of algorithmic depth and systems engineering. Comfortable across the stack from model design to deployment, he often surfaces pragmatic improvements (like reshaping unknown tensors and weight-loading flexibility) that ease real-team workflows. He brings academic roots from TU Delft and a track record of making complex vision tech reliably production-ready.
13 years of coding experience
3 years of employment as a software developer
TU Delft
Dutch, English, German
Stackoverflow
376reputation
8kreached
3answers
1question
Github Skills (16)
retinanet10
neural-network10
object-detection10
keras10
computer-vision10
machine-learning10
deeplearning-ai10
deep-learning10
tensorflow10
loss10
artificial-neural-networks10
python10
data-science10
modeling9
trainings9
Programming languages (21)
Github contributions (5)
fizyr/keras-retinanet
Aug 2017 - Jun 2022
Keras implementation of RetinaNet object detection.
Role in this project:
ML Engineer
ML EngineerContributions:5 releases, 7 reviews, 634 commits in 4 years 10 months
Contributions summary:Hans's primary contributions center around the implementation and enhancement of a RetinaNet object detection model using Keras. They added setup files, generated anchors and anchor targets, and incorporated classification loss functions, including focal loss, to optimize model performance. Furthermore, the user integrated features like softmax activation and class probability scaling to improve the accuracy of classification scores and included training scripts demonstrating the model's usage.
deep-learningretinanetcomputer-visionobject-detectiontensorflow
keras-team/keras
Jul 2017 - Dec 2017
Deep Learning for humans
Role in this project:ML Engineer
ML EngineerContributions:8 commits, 21 PRs, 70 comments in 5 months
Contributions summary:Hans contributed significantly to the Keras library, focusing on improving its functionality and usability. They implemented a seed parameter for image augmentation methods, fixed a typo, and added functionality to reshape unknown tensors. Moreover, the user added datetime formatting for ETA time and allowed custom losses to have mismatching target shapes. They also introduced an option to load weights with mismatch and allowed callbacks to override their model.
pythondata-sciencedeep-learningneural-networksmachine-learning
Find and Hire Top DevelopersWe’ve analyzed the programming source code of over 60 million software developers on GitHub and scored them by 50,000 skills. Sign-up on Prog,AI to search for software developers.