Team solutions

Posted vsenderov, valan, hvrigazov, cecopld, mitzev, tonypetrov, pepe Mentors: tony6 CommentsPosted in Image recognition, Learn, Team solutionsImage recognition, Learn, Team solutions

Team Cherry. The Kaufland Case. Fast and Accurate Image Classification Architecture for Recognizing Produce in a Real-Life Groceries’ Setting

Our best model (derived from VGG) achieved 99.46% top3 accuracy (90.18% top1) with processing time during training of 0.006 s per image on a single GPU Titan X (200s / epoch with 37 000 images).

The teams vision is for the team members to see where they stand compared to others in terms of ideas and approaches to computer vision and to learn new ideas and approaches from the other team-mates and the mentors.

Therefore the team is pursuing a pure computer vision approach to solving the Kaufland and/or the ReceiptBank cases.

Posted georgi-kostadinov, stavri, pepe, mitzev, tonypetrov, pepe Mentors: tony18 CommentsPosted in Team solutionsTeam solutions

Using Convolutional Neural Networks for Real-Time Product Recognition in Smart Scales – Imagga’s Solution to the Kaufland Case

Many big retailers offer in store a rapidly growing variety of fresh produce from local and global sources, such as fruit and vegetables that need to be weighed quickly and effortlessly to mark their quantity and respective price. Smart scales that use image recognition to optimise user experience and allow additional features, such as e.g. cooking recipes can provide a new solution to this problems. The solution we provide to the Kaufland case includes training a Convolutional Neural Network (CNN) with GoogLeNet architecture on the original Kaufland data set and fine-tuning it with a Custom Training Set we have created, achieving the following results (Kaufland Case Model #13): training accuracy: Top-1: 91%, Top-5: 100%; validation accuracy: Top-1: 86.1% , Top-5: 99%, and TEST dataset accuracy of: Top-1: 86.1%, Top-5: 99.2%. We have also created another model (Kaufland Case Model #14) by combining similar categories, achieving: training accuracy: Top-1: 96%, Top-5: 100%; validation accuracy: Top-1: 92.5%, Top-5: 100%, and TEST dataset accuracy: Top-1: 91.3%, Top-5: 100%. All trainings were done on our NVIDIA DGX Station training machine using BVLC Caffe and the NVIDIA DIGITS framework. In our article we show visualisations of our numerous trainings, provide an online demo with the best classifiers, which can be further tested. During the final DSS Datathon event we plan to show a live food recognition demo with one of our best models running on a mobile phone . Demo URL: http://norris.imagga.com/demos/kaufland-case/

Posted madalina, stephan, pepe Mentors: hl5 CommentsPosted in Image recognition, Learn, Team solutionsImage recognition, Learn, Team solutions

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.

Posted mboyanov, hbuyukliev, anna-maria-angelova, ktokmakov, va1io, bernhard, tomislavk1 CommentPosted in SNA, Team solutionsSNA, Team solutions

404 Telenor Social Media Not Found

Social networks are characterized by the links between the nodes. We employed six different link analysis algorithms to rank the nodes in the network by their importance. For the task of leader detection, the best link analysis algorithm proved to be the vanilla PageRank. Out of the top 2000 nodes, 1725 are leaders, achieving a precision of 0.8625 and a recall ot 0.8792.
We also explored an alternative solution based on embeddings. We trained a Skipgram model where we set the context of a given node to its neighbourhood. To avoid the assumption for word order made by the model we repeated and reshuffled longer neighbourhoods. These embeddings were then used to train a softmax classifier, achieving a macro-f1 score of 0.6667.
We futher provide a visualization tool which can be used to explore the graph manually.