Team GrBrGr, Case Kaufland
This paper presents a machine learning based approach for solving the business problem of identifying from pictures the products chosen by the Kaufland customers. These pictures are all taken from the same angle and typically show one or multiple products from the same category in a bag which makes the background and the bag recurrent elements.
Here we explored the method of transfer learning – using already trained and very deep NN like InceptionV3, InceptionResnetV2, VGG19, Resnet50 with combinations of retraining and no retraining of the existing layers. We solved this multiclassification problem of predicting the probabilities of each class of products by adding different final custom layers and we obtain the best result of 85% accuracy on a validation set of 20% (which was never seen by the training model).
This result was achieved with the model VGG19 which distinguished itself not only for providing the best categorical accuracy but also for training speed, execution speed once deployed and reduced resource consumption.
I will like to see the reported accuracy against the provided test data set by Kaufland, not just on 20% of the training set.
Since datascience is experimental science I very much appreciate the approach by the team to compare wide range of CNN architectures!
Very well done!
Hello Tony, the evaluation against the test dataset provided this morning is present at the bottom with the header “Test Dataset”, the results are identical to that of the holdout (the confusion matrices are also displayed) with a top-3 accuracy of 0.9732 (we had 0.9735 on our 20% holdout)
I love the thorough botanical analysis, it definitely underlines some main problems that need to be addressed before any human or computer vision can take place.
Congratulations for the Great job! 🙂
A very valuable is your investigation across different models, their comparison and conclusions.
And also, appreciate the very well and systematically documented presentation! 🙂