Image Recognition for Retail
• Create a model in order to detect products shelve share and quantities.
• The higher the shelve share, the higher the probability of closing sales.
• This measure would give the customer an idea of the amount of products they have visible on
shelves at a given moment in time.
Machine learning operations (MLOps) steps
Take the pictures (High Quality).
Label them using a labelling app called LabelImg.
Included pictures of objects at different angles and under different lighting conditions.
Started with 100 pictures of each class (The different brands and packages of beer).
Trained the algorithm using transfer learning using a pre-trained tensorflow model (SSD MobileNet V2 FPNLite 320x320).
Peformed an evaluation on 20% of the pictures obstaining very high precision results upwards of 95% accuracy.