CODEWITHSUNDEEP
pythonmachine-learningclassificationconv-neural-network
Paul Gwamanda
Paul Gwamanda

Reputation: 302

Image classification: Best approach to training the model

Given a model that has to classify 10 table-cloth items (spoons, forks, cups, plate etc,) and must be tested on an image of a table with all the table-cloth items in it (test_model_accuracy,) which is the best approach for training:

Upvotes: 2

Views: 175

Answers (1)

Giovanni Patruno
Giovanni Patruno

Reputation: 773

The strategy that you will choose depends mainly on the structure of the CNN that you are going to create.

If you train a model that is able to recognize if an image contains a spoon or a fork, you will not be able to test on a table with several table-cloth items (e.g. both a fork and a spoon) because the network will try to answer if in the image there is a spoon or a fork indeed.

Anyway, it is still possible to train the network to classify several features (strategy "A") but in that case you need a model that is able to do Multi-label classification.

Finally, I would suggest to go for the "B" strategy because, in my humble opinion, it fits good the application domain.

Hope this answer is clear and helpful!

Cheers.

Upvotes: 1

Related Questions