I did this project as part of my research assistantship at Cylab under the guidance of Professor MariosSavvides. In this project my objective was to identify out of stock and wrongly placed products using aisle planograms.

I used RetinaNet for product and label detection in the planogram images. I cropped images based on detection. Using Cosine Similarity, I cleaned the dataset.

Finally, I utilized EfficientNet model (also tried ResNetCutmix model) for product identification and classification using Ringloss.

Once the product id was identified then based on the label information product was classified as plug if it was a wrongly placed product on shelf.

I achieved detection precision and recall of more than 95%. For product classification, I achieved top 1% validation accuracy of 85% and top 5% validation accuracy of 96%.