I'm new to AWS and looking to develop a vehicle detection and classification pipeline from scratch. My plan is to take one image per second, upload it to the cloud, and conduct batch inference since real-time analytics aren't needed. I'm budget-conscious with this project, and I'd appreciate any insights or guidance on similar implementations. I've been advised to consider the following cost-saving methods: sending a tarball from the NVR to S3 for inferencing, using Amazon spot instances, and opting for Graviton instances instead of standard EC2. Any tutorials or blog posts would be extremely helpful!
3 Answers
Honestly, there's too much reliance on AI tools for advice. It’s essential to put some effort into research yourself rather than just asking for confirmation of AI-generated suggestions.
When talking about Graviton, remember that it comes in x86 and ARM versions, so think about what suits your application best. Also, it's crucial to define your compute requirements upfront, such as memory and CPU specs, as this will affect your costs and performance.
Using Graviton is interesting since it's a type of processor for EC2, but consider your processing time. If your tarball processes in under 15 minutes, you might want to try AWS Lambda. You can set up a trigger with an S3 event to automate processing. If it takes longer, ECS with spot instances could be your go-to. ECS is easier for beginners, and using something like Fargate would further simplify things. Just remember, there are different options available depending on your needs.
Related Questions
Biggest Problem With Suno AI Audio
How to Build a Custom GPT Journalist That Posts Directly to WordPress