Manpreet is available for hire

Manpreet Singh Minhas

Verified Expert in Engineering

Artificial Intelligence Engineer and Developer

Location

Waterloo, ON, Canada

Toptal Member Since

August 17, 2021

Manpreet is a computer vision and deep learning engineer and developer who has taken many AI-based ideas from the research stage to production. He regularly shares his industry and research expertise in deep learning and computer vision by authoring technical articles on Medium's Towards Data Science. Manpreet's industry experience is backed by a master's degree in computer vision and deep learning, he is passionate about technology, and he loves to build cool stuff.

Deep Learning PyTorch TensorFlow Scikit-learn Python 3 Pandas NumPy OpenCV Git Jupyter SciPy Visual Studio Code (VS Code)Computer Vision Machine Learning Amazon Web Services (AWS)Anomaly Detection CNN

Portfolio

Fugro

Computer Vision, Amazon SageMaker, Jupyter, Pandas, NumPy, SciPy, Scikit-learn...

University of Waterloo

PyTorch, Anomaly Detection, Convolutional Neural Networks (CNN), Pandas...

Experience

Python 3 - 4 years Deep Learning - 4 years Scikit-learn - 4 years PyTorch - 4 years Convolutional Neural Networks (CNN) - 4 years OpenCV - 3 years Computer Vision - 3 years Git - 3 years

Availability

Part-time

Preferred Environment

Python 3, Git, Visual Studio Code (VS Code), Ubuntu, PyTorch, GitHub, Amazon Web Services (AWS)

The most amazing...

...thing I've researched and developed is AnoNet, a weakly supervised, fully convolutional network for anomaly detection in textured surfaces.

Work Experience

Computer Vision and Deep Learning Research Engineer

2020 - PRESENT

Fugro

Developed pavement distress detection and classification from end-to-end in PyTorch. Integrated the algorithms into Fugro's C# application by creating a C++ DLL interface.
Created bird's-eye-view projection to get a top-down view algorithm and a Python GUI application for processing data from SQL databases. Implemented multi-processing to speed up processing by 50%.
Incorporated deep learning-based object detection and tracking algorithms to reduce manual processing costs by approximately 35%.
Automated the testing and deployment of packages by introducing GitHub Actions to the team's workflow.
Applied text recognition on traffic signs to automate categorization based on the MUTCD standard.

Technologies: Computer Vision, Amazon SageMaker, Jupyter, Pandas, NumPy, SciPy, Scikit-learn, TensorFlow, PyTorch, Convolutional Neural Networks (CNN), Anomaly Detection, Machine Learning, OpenCV, Python 3, Object Detection, Amazon Web Services (AWS), Algorithms, SQL, Deep Learning, Automated Testing, Python, C#, C++, DLL, GitHub, Artificial Intelligence (AI), Data Science, Image Processing, Amazon S3 (AWS S3)

Research Associate

2018 - 2019

University of Waterloo

Researched, developed, and published an article titled AnoNet: Weakly Supervised Anomaly Detection in Textured Surfaces Using CNNs.
Wrote and published a CNN-based autoencoder architecture for semi-supervised anomaly detection.
Created and published a technique for anomaly detection in images using transfer learning.
Developed and presented defect detection using deep learning from minimal annotations.

Technologies: PyTorch, Anomaly Detection, Convolutional Neural Networks (CNN), Pandas, TensorFlow, NumPy, SciPy, Scikit-learn, Matplotlib, Jupyter, Visual Studio Code (VS Code), Ubuntu, Python 3, Python, Deep Learning, Machine Learning, Artificial Intelligence (AI), Computer Vision, Data Science

Experience

AnoNet: Weakly Supervised Anomaly Detection in Textured Surfaces Using CNNs

https://arxiv.org/abs/1911.10608

A CNN architecture capable of learning to detect the actual shape of anomalies in textured surfaces from weakly labeled data and a ﬁlter bank-based initialization technique that led to stellar performance. AnoNet achieves state-of-the-art performance with an average F1 score of 0.91 and an AUROC value of 0.92 across four challenging datasets.

Automatic Redaction of Video Recordings Using Deep Learning

A deep learning system using MTCNN and ResNet50 trained on the VGGFace2 dataset for the task of the automated redaction of video recordings. The system blurs all faces except that of the person of interest in the input video.

Bird's-eye-view Projection to Get a Top-Down View

A bird's-eye projection algorithm for getting top-down images via inverse perspective mapping. The projection is done such that it allows for the measurement of physical dimensions of the objects directly and geo-references the pixel locations. I also developed a GUI allowing the processing of data from an SQL server and implemented parallel processing to increase the processing speed by 50%.

Semi-supervised Anomaly Detection Using Autoencoders

https://github.com/msminhas93/anomaly-detection-using-autoencoders

Researched and developed a semi-supervised anomaly detection technique based on convolutional autoencoders. It can be trained using only the normal (non-anomalous) samples. The approach led to an impressive average F1 score of 0.885 on two challenging datasets with limited training samples.

Road Crack Segmentation Using Transfer Learning

https://github.com/msminhas93/DeepLabv3FineTuning

Trained a DeepLabv3 model for segmenting road cracks from a right-of-way camera using transfer learning. The dataset contained urban road surface images with cracks as defects. The model achieved an AUROC value of 0.842 on the test dataset.

Defect Detection Using Deep Learning from Minimal Annotations

https://www.insticc.org/node/TechnicalProgram/visigrapp/2020/presentationDetails/91680

Formulated the automatic defect detection task as the problem of anomaly detection in which samples that deviate from the normal or defect-free samples need to be identified. Developed a network-based transfer learning approach. Achieved high performance from limited data samples with an average F1 score and AUROC values of 0.8914 and 0.9766, respectively.

Education

2018 - 2019

Master's Degree in Computer Vision and Deep Learning

University of Waterloo - Waterloo, ON, Canada

2011 - 2015

Bachelor's Degree in Electronics and Telecommunication

University of Mumbai - Mumbai, Maharashtra, India

Certifications

MARCH 2019 - PRESENT

Deep Learning Specialization (Five Courses)

DeepLearning.AI

Skills

Libraries/APIs

PyTorch, TensorFlow, Scikit-learn, OpenCV, Pandas, NumPy, SciPy, Matplotlib

Tools

TensorBoard, Scikit-image, Git, Jupyter, Amazon SageMaker, GitHub

Languages

Python 3, C++, SQL, Python, C#

Paradigms

Anomaly Detection, Siamese Neural Networks, Automated Testing, Data Science

Platforms

Visual Studio Code (VS Code), Amazon Web Services (AWS), Ubuntu

Storage

Amazon S3 (AWS S3)

Other

Convolutional Neural Networks (CNN), Deep Learning, Neural Networks, Computer Vision, Image Processing, Object Detection, Machine Learning, Variational Autoencoders, Graphical User Interface (GUI), Algorithms, DLL, Artificial Intelligence (AI)

Collaboration That Works

How to Work with Toptal

Toptal matches you directly with global industry experts from our network in hours—not weeks or months.

Share your needs

Discuss your requirements and refine your scope in a call with a Toptal domain expert.

Choose your talent

Get a short list of expertly matched talent within 24 hours to review, interview, and choose from.

Start your risk-free talent trial

Work with your chosen talent on a trial basis for up to two weeks. Pay only if you decide to hire them.

Top talent is in high demand.

Start hiring