Computer Vision Engineer

inDrive, Zerotech

Industry: IT & Software

Specialization: Computer Vision, Natural Language Processing

Yerevan, Armenia

$-

Tech Stack: Python, Pandas, Scikit-Learn, Keras, Linux

Expert’s cases:

Selected and adapted the most suitable object detection architecture for devices with limited computational capabilities from available options
Developed a unique dataset, automated the annotation process, and reduced the model's error rate by 40% compared to the baseline
Successfully quantized a previously non-quantizable model and tailored it for specialized hardware
Optimized neural network architectures for computationally weak device, resulting in a 27% inference speedup with only a 3% decrease in accuracy
Trained neural networks for human pose estimation tasks with custom augmentations, resulting in accurate and efficient models
Customized and trained face recognition neural models, achieving an inference speedup of up to 1160%
Developed and implemented a custom algorithm for object tracking, enabling precise and efficient tracking of objects in various environments
Implemented a neural model to accurately segment document fields, streamlining the processing of important information
Developed a neural model for optical character recognition (OCR), improving the accuracy of text recognition in scanned documents
Conducted extensive research on liveness detection on the photos to prevent fraudulent activity, utilizing neural nets to increase accuracy
Developed a microservices for computer vision tasks, utilizing Python and Kafka to streamline the processing of visual data
Created a synthetic dataset for 3D face modeling from single photos using software rendering techniques
Designed and trained neural models to remove shadows and highlights from digital photos
Developed and trained a neural model for correcting white balance and light levels in photos
Created an algorithm for quick and accurate determination of dominant colors, outperforming the K-means algorithm
Implemented and trained neural networks for object segmentation tasks
Devised and implemented an innovative approach for storing hidden text in neural network weights

Computer Vision Engineer

inDrive, Zerotech

Industry: IT & Software

Specialization: Computer Vision, Natural Language Processing

$-

Tech Stack: Python, Pandas, Scikit-Learn, Keras, Linux

Expert’s cases:

Selected and adapted the most suitable object detection architecture for devices with limited computational capabilities from available options

Developed a unique dataset, automated the annotation process, and reduced the model's error rate by 40% compared to the baseline

Successfully quantized a previously non-quantizable model and tailored it for specialized hardware

Optimized neural network architectures for computationally weak device, resulting in a 27% inference speedup with only a 3% decrease in accuracy

Trained neural networks for human pose estimation tasks with custom augmentations, resulting in accurate and efficient models

Customized and trained face recognition neural models, achieving an inference speedup of up to 1160%

Developed and implemented a custom algorithm for object tracking, enabling precise and efficient tracking of objects in various environments

Implemented a neural model to accurately segment document fields, streamlining the processing of important information

Developed a neural model for optical character recognition (OCR), improving the accuracy of text recognition in scanned documents

Conducted extensive research on liveness detection on the photos to prevent fraudulent activity, utilizing neural nets to increase accuracy

Developed a microservices for computer vision tasks, utilizing Python and Kafka to streamline the processing of visual data

Created a synthetic dataset for 3D face modeling from single photos using software rendering techniques

Designed and trained neural models to remove shadows and highlights from digital photos

Developed and trained a neural model for correcting white balance and light levels in photos

Created an algorithm for quick and accurate determination of dominant colors, outperforming the K-means algorithm

Implemented and trained neural networks for object segmentation tasks

Devised and implemented an innovative approach for storing hidden text in neural network weights