Flask app with RF-DETR, live uploads, and cached results to save credits

[davidcarroll]

Show RF: RF-DETR Flask Detector

Hi everyone, I just finished my first project with Roboflow, RF-DETR Flask Detector, a Flask web app for object detection using RF-DETR through the serverless inference API.

You can upload your own image and get detections back, or browse a gallery of sample images from the COCO dataset with results already loaded. Since I’m on the free plan, credits were a real constraint, so I added a caching layer using AWS CloudFront to avoid repeat requests. That made the app faster and a lot cheaper to run.

Live demo: What's in the picture?

Code: GitHub - david-carroll-aws/roboflow-detection-demo: Flask web app for AI-powered object detection and recognition, using Roboflow's RF-DETR transformer model via the serverless inference API, with cached results and live image uploads · GitHub

Since this is my first time working with Roboflow, I’d love feedback. Is anyone else caching inference results, or handling it a different way? Any tips for getting more out of RF-DETR would be great.

Cheers!