Building Scalable AI Computer Vision Applications
Building Scalable AI Computer Vision Applications


Computer vision has enabled enterprises to leverage image and video data to gain actionable insights that can help to generate new revenue opportunities and improve efficiency. Recent advancements in AI and machine learning have opened the door for millions of CV use cases to be built. However, the process of getting these applications into production has been slow due to the lack of unified workflows/platforms and the fragmentation of available tools within the open-source. Furthermore, existing tools have only catered towards data scientists and not general application developers.

This talk will introduce a standardized process and workflow that can be followed to build any computer vision application. This process can help to accelerate development for both application developers and data scientists. We will elaborate on the following:
How to prepare image or video datasets for ML model creation
How to efficiently annotate images/videos to create training/validation datasets
Determine if you should use a pre-trained and apply transfer learning or use a customized model
The need for benchmarking across multiple models, datasets, and hardware targets
What model/system metrics really matter for applications
How to extract poor-performing data for relabelling/retraining
The considerations when deploying on the cloud, edge devices, or edge servers
Monitoring model and data quality/performance


Dr. Nitin Gupta helped found Dori AI with the vision of enabling the applied AI market with a platform that accelerates the adoption of AI/ML across major vertical industries. He has 10+ years of experience leading product definition and architecture in edge, mobile, and computer vision-based systems. Prior to Dori, Nitin was a technical product lead at Google responsible for commercializing Daydream ML/CV-based systems for VR/AR software and hardware product lines.

In addition to that, Nitin led Pebble firmware/graphics development and commercialized wearable products including the Pebble, Pebble Time, Pebble Time Steel, and Pebble Time Round. He started his career at Qualcomm where he led a number of efforts within QCT to architect, patent and commercialize boot and UEFI software for some of the first Android, iOS, and Windows-based smartphones. He also holds a Ph.D. in Computer Science from the University of Manchester where he was advised by the co-founder of ARM.