Pretty bad imo. Tried three objects: Diet Coke can, bottle of cleaning fluid, desktop cactus. Didn’t identify any even remotely close. Was just like a random guess.

OP here: I shouldn't have added the word 'etc' to the list of things it classified, but have changed the readme to show a full list of objects that the demo works with. My intent was not to make a classifier for any and all objects, to serve as a demonstration of what someone could use the app for.

I'm going to suggest it's pretty bad form to call this "Fast.ai Camera App" when it's not a Fast.ai product.

Hi! I can see where you're coming from; however, since this product is initially intended to tightly work in conjunction with Render, which uses the fast.ai name in it's repo (https://github.com/render-examples/fastai-v3) and the fact that this model (to my limited knowledge) only works with models produced by fast.ai, making a clear naming distinction was crucial so that someone creating models using another library would realize they couldn't use this. If you have a suggestion for how to make that distinction in a clear and concise manner without coming across as an official product I would definitely be willing to hear you out!

Well I don't see why the fact it uses a fast.ai model is particularly relevant for the name.

You don't mention PyToch or ResNet of Python or any of the million of other libraries it needs.

Right, but this library is designed only to be used with a fast.ai model (and the hope is that fast.ai practitioners will find it helpful). I suppose you could create your own backend that worked with another library if you wanted to take the time, but that's not how it's designed.

EDIT: Regardless, I've updated some of the language on the readme to be more clear that this is not an official fast.ai product (and made it more clear in the header that it uses fast.ai models), because I do think your criticism had some validity and I really do appreciate the feedback.

Agreed, even though they use a model from Fast.ai, the name is very misleading. With that title you'd expect an official release with a rubber stamp at the bottom.

Yes, and better quality.

I wasn’t familiar with the that fastai model until now. I took 6 photos and the classification was an utter failure.

Pretty neat, it recognized a chair even though it was obscured with clothes. Granted it does not recognize some heavily specific objects (as it says in the description) but it is a start.

this can be done in browser with tensorflow.js + a pretrained mobilenet in a couple dozen lines of code

example: https://jott.live/html/tfjsmobilenet code: https://jott.live/code/tfjsmobilenet

> After the picture is taken, it's sent to a fast.ai CNN model running on render

Why does the model have to run on render? Can it be run on other servers?

Good question! Like I mentioned in the readme, I'm hoping to add a flask server soon; however, render is recommended on the fast.ai website and is the simplest way to get a model running (https://course.fast.ai/deployment_render.html), so I thought connecting my app to render would be the most helpful way for fast.ai practitioners to get started.