What is the application?
It specifies the breed of cat by pictures using your device's camera or image gallery.
How does it works?
The photo is fed to the input of the neural network (at the moment the EfficientNetV2 architecture is used) and at its output a hypothesis is formed about what breed of cat is shown in this photo. The new version of the classifier has become less playful and only reacts to photos of real cats. Drawn cats, cartoons, toys, dogs, other animals, photos of people - the neural network most often ignores.
What is recognition accuracy?
The system is trained to recognize 62 cat breeds from 13,000 photographs. In this version of the application, the accuracy of recognition of cat breeds was 63% on 2 thousand photos from the test sample (not used in training the classifier) and 86% on all available photos. The training database of cat photos is being supplemented and improved, so the number of breeds and the quality of their recognition will increase in new releases.
Goals For Future.
It will be added to supplement the training set of cat photos your examples and thus continuously expand the number of cat breeds and recognition accuracy. The purpose of the project to create an expert system able to recognize the photos all known breeds of cats.
It specifies the breed of cat by pictures using your device's camera or image gallery.
How does it works?
The photo is fed to the input of the neural network (at the moment the EfficientNetV2 architecture is used) and at its output a hypothesis is formed about what breed of cat is shown in this photo. The new version of the classifier has become less playful and only reacts to photos of real cats. Drawn cats, cartoons, toys, dogs, other animals, photos of people - the neural network most often ignores.
What is recognition accuracy?
The system is trained to recognize 62 cat breeds from 13,000 photographs. In this version of the application, the accuracy of recognition of cat breeds was 63% on 2 thousand photos from the test sample (not used in training the classifier) and 86% on all available photos. The training database of cat photos is being supplemented and improved, so the number of breeds and the quality of their recognition will increase in new releases.
Goals For Future.
It will be added to supplement the training set of cat photos your examples and thus continuously expand the number of cat breeds and recognition accuracy. The purpose of the project to create an expert system able to recognize the photos all known breeds of cats.
Show More
