I had Trained data of various images of truck tyres using the CNN model to classify whether the image is a good tyre or Bad tyre.
When I am giving non-tyre image to my model, the model is predicting classes of good and bad.
How to teach my model, when I input non-tyre image its output should show please give the proper image to the model.
I don't know if that will really help in your situation, but at least I'd try to create two models (nested classification) where one would classify whether it was a tire or not. Depending on the result of this classification, it would then be decided whether to perform a second classification to determine whether the tire is good or not.
HI Prashanth Kolaneru
May be you can update your model to a ternary classifier (three classes) rather than binary one. Three classes should be: (Good, Bad and No) tyres .
Regards
As said above, you need three classes (good, bad and non-tire objects).
Now, a question arises, what shall we include in the non-tire objects? Answer: everything that you could, especially those objects that the model could be confuse with being tires (e.g., side view of black rubber - Wrist Watches / bracelets).
On the other hand, you can extract and examine the probability of the prediction, I will have an assumption here that your binary classifier would have a lower certainty that the "tire" (in this case a non-tire image) is good or bad.
- Badges
- Science topic
- Similar topics
- Computer Science
- Images