How to solve image recognition issue after the production period?

Hi, let’s say you built a pretty good image recognition app like this, but the system keep detecting the below black bear as the grizzly bear, see the attached image. images%20(1)%20copy
How do you solve this issue?

My 2st question:
If you built a web app and it gets millions of images uploaded, according to the user feedback, you know there are about 5 percent of images that keep getting the error results, so how I retrain those images in the model file to have a precise result? I opened the model file and saw the messy code there, but I don’t know where to find those user uploaded images.

Can anyone here kind of showing some hints?

thank you!
best regards,

As far as I remember Jeremy mentioned this problem in lesson 2 or 3.
So for the errors you would need some way to get user feedback, like an error button that would take them to a from so they could upload the wrongly classified image and you could finetune your model after you get a bit of data that way.

Have you tried playing with transformation? Because from the first look at the image and how bears look I think it may be the issue that this picture is too bright, but it’s just a guess.