256 points by sadserverless 5 months ago flag hide 17 comments
someuser1 5 months ago next
Amazing work! I've been looking for a way to incorporate image recognition into my project, and this might just be the solution.
anotheruser2 5 months ago prev next
Great article! I'm curious what the cost-benefit analysis for going serverless looks like for this kind of API?
someuser1 5 months ago next
That's a great question. I'll be sure to include that in a follow-up post.
anotheruser3 5 months ago prev next
I'm not super familiar with serverless architectures, could you expand on how you deployed and scaled this API?
someuser1 5 months ago next
Certainly! I used AWS Lambda to handle the core logic, and API Gateway to handle routing the requests to Lambda.
anotheruser2 5 months ago next
I've heard using ML models in Lambda functions can be tricky due to timeouts, did you run into that issue at all?
someuser1 5 months ago next
That's a valid concern. Fortunately, AWS recently introduced provisioned concurrency for Lambda which helped me avoid those limitations.
anotheruser4 5 months ago prev next
Did you consider using pre-trained models, or did you train your own custom models for this project?
someuser1 5 months ago next
I used pre-trained models from AWS's SageMaker, but I look forward to exploring custom model training soon.
user5 5 months ago prev next
Very cool project! For production-level use, have you looked into complying with privacy regulations, such as GDPR for image data?
someuser1 5 months ago next
Yes, my team and I have taken GDPR and other privacy regulations into account. We store just enough data to satisfy the use case, and discard it promptly afterward.
user6 5 months ago prev next
How has your team dealt with false positives and negatives in the image recognition?
someuser1 5 months ago next
It's important to note that we are always working on reducing false positives and negatives. We have implemented continuous integration and testing of our models.
user7 5 months ago prev next
How do you think edge computing will change image recognition and similar ML use cases?
someuser1 5 months ago next
Great question! I believe edge computing will allow for much faster processing of data, including image data. It also allows for more autonomy and functionality even without network connectivity.
user8 5 months ago prev next
Did you think about using multiple ML services to provide your image recognition capabilities to help improve accuracy?
someuser1 5 months ago next
We did! Although using multiple services can provide higher accuracy, my focus was to keep the implementation as simple and as cost-effective as possible.