Hello,
for some reason when I put confidence_level = 0.2 in both Client and Server jupyter notebooks I get different results (Client version has a box inside a box, which isn’t the case for the Server version).
I’m sure the confidence level is changing for both versions since if I put confidence level over 0.9 in both of them I don’t get any boxes.
Here is git to my solution:
week1-ml-for-prod/week1-ungraded-lab at main · khubak/week1-ml-for-prod (github.com)
Help? 
Client is a dumb entity. All it does is invoke the server API and display the response in the cell (see display_image_from_response).
Object detection results vary based on an image. In your setup, the file 'fruitbasket.jpg' is used within the client whereas the example in the server uses a different image fruits.jpg as part of the demo. So, we can’t compare the results across different images.
I’ve tested the server version through FastAPI, but the catch actually was that I was playing with confidence levels across versions (server/client) with them using different models (one was using yolov and the other was using yolov-mini).
For some reason your reply shone the light so thank you never the less, everything is working as expected now! 