Episode 045 - Computer Vision on AWS with Francesco Pochetti
Jul 15, 2022
auto_awesome
Francesco Pochetti, Senior Machine Language Engineer at Bolt and an AWS Machine Learning Hero, discusses his transition from chemist to computer vision engineer, the value of diverse learning in machine learning, and challenges of computer vision on AWS such as face matching and OCR. They also cover topics like network bandwidth challenges, deploying models on SageMaker, optimizing neural networks, and face blurring for privacy.
Using AWS services like AWS recognition and Amazon Textract in image-based verifications is crucial for ensuring user requirements are met.
Deploying deep learning computer vision models effectively requires optimization frameworks like TensorRT and overcoming challenges related to size of images and network bandwidth.
Deep dives
Building a Verification Platform at Bolt
At Bolt, the speaker is building a verification platform to ensure users meet certain requirements, such as having a valid driving license. This process involves image-based verifications, including face matching and optical character recognition (OCR) to read and verify information from ID documents. AWS services like AWS recognition and Amazon Textract play a key role in these verifications. The speaker also highlights the importance of understanding the business problem being solved and the need for human review fallback using SageMaker Ground Truth.
Challenges in Computer Vision Deployment
The speaker mentions that training deep learning computer vision models is not a challenge thanks to pre-trained models. However, the real challenge lies in deploying these models effectively. The speaker discusses the inherent slowness of deep learning models and the need for optimization frameworks like TensorRT to achieve real-time processing. The size of high-resolution images and the limitations of network bandwidth also pose deployment challenges. Despite these challenges, services like AWS SageMaker simplify the deployment of models.
Privacy and the Importance of Responsible AI
Privacy and responsible AI are key concerns highlighted by the speaker. The importance of considering the ethical implications and potential harm of AI models is emphasized. Privacy, especially in relation to AI and facial recognition, is a top concern that needs to be addressed. The speaker stresses the need for responsible development and understanding the impact of AI on society.
In this episode, Dave chats with Francesco Pochetti, Senior Machine Language
Engineer at Bolt, and an AWS Machine Learning Hero. Francesco covers his career start as a
chemist, his journey into a career of Data Science, and how Computer Vision technology is
handling some of the most difficult Machine Learning problems today.
Francesco on Twitter: https://twitter.com/Fra_Pochetti
Dave on Twitter: https://twitter.com/thedavedev
Francesco’s Website: https://francescopochetti.com/
Francesco’s LinkedIn: https://www.linkedin.com/in/francescopochetti/
Francesco’s GitHub: https://github.com/FraPochetti
[BLOG] Blurry faces: Training, Optimizing and Deploying a segmentation model on Amazon
SageMaker with NVIDIA TensorRT and NVIDIA Triton:
https://francescopochetti.com/blurry-faces-a-journey-from-training-a-segmentation-model-to-deploying-tensorrt-to-nvidia-triton-on-amazon-sagemaker/
[BLOG] Machine Learning and Developing inside a Docker Container in Visual Studio Code:
https://francescopochetti.com/developing-inside-a-docker-container-in-visual-studio-code/
[BLOG] Deploying a Fashion-MNIST web app with Flask and Docker:
https://francescopochetti.com/deploying-a-fashion-mnist-web-app-with-flask-and-docker/
[BLOG] IceVision meets AWS: detect LaTeX symbols in handwritten math and deploy with Docker
on Lambda:
https://francescopochetti.com/icevision-meets-aws-detect-latex-symbols-in-handwritten-math-and-deploy-with-docker-on-lambda/
[DOCS] Amazon Rekognition: https://aws.amazon.com/rekognition/
[DOCS] Amazon SageMaker: https://aws.amazon.com/sagemaker/
[DOCS] Amazon Textract: https://aws.amazon.com/textract/
[DOCS] Deploy fast and scalable AI with NVIDIA Triton Inference Server in Amazon SageMaker:
https://aws.amazon.com/blogs/machine-learning/deploy-fast-and-scalable-ai-with-nvidia-triton-inference-server-in-amazon-sagemaker/
[GIT] Nvidia Triton Inference Server:
https://github.com/triton-inference-server/server/
[GIT] Blurry faces: Training, Optimizing and Deploying a segmentation model on Amazon
SageMaker with NVIDIA TensorRT and NVIDIA Triton:
https://github.com/FraPochetti/KagglePlaygrounds/tree/master/triton_nvidia_blurry_faces
Subscribe:
Amazon Music:
https://music.amazon.com/podcasts/f8bf7630-2521-4b40-be90-c46a9222c159/aws-developers-podcast
Apple Podcasts: https://podcasts.apple.com/us/podcast/aws-developers-podcast/id1574162669
Google Podcasts:
https://podcasts.google.com/feed/aHR0cHM6Ly9mZWVkcy5zb3VuZGNsb3VkLmNvbS91c2Vycy9zb3VuZGNsb3VkOnVzZXJzOjk5NDM2MzU0OS9zb3VuZHMucnNz
Spotify:
https://open.spotify.com/show/7rQjgnBvuyr18K03tnEHBI
TuneIn:
https://tunein.com/podcasts/Technology-Podcasts/AWS-Developers-Podcast-p1461814/
RSS Feed:
https://feeds.soundcloud
Get the Snipd podcast app
Unlock the knowledge in podcasts with the podcast player of the future.
AI-powered podcast player
Listen to all your favourite podcasts with AI-powered features
Discover highlights
Listen to the best highlights from the podcasts you love and dive into the full episode
Save any moment
Hear something you like? Tap your headphones to save it with AI-generated key takeaways
Share & Export
Send highlights to Twitter, WhatsApp or export them to Notion, Readwise & more
AI-powered podcast player
Listen to all your favourite podcasts with AI-powered features
Discover highlights
Listen to the best highlights from the podcasts you love and dive into the full episode