Beer Price Checker

Group Project Members

Alex Li
Azwin Lam
Jason Lee

Objective

• Create a deployable application that enables a quick and effective method for checking & comparing retail prices of beers from supermarkets

Data Sources

Overview

• 3500+ images of beer bottles and cans with their logos
  • Google Images
• 18 brands of beer found from Consumer Council Price Watch website
  • https://online-price-watch.consumer.org.hk/opw/

Beer Image Samples

Consumer Council Data Sample

Data Cleaning

Beer Images

• Images were generally not zoomed in enough on the logos
• Used Inbac for clean and efficient image cropping
  • https://github.com/weclaw1/inbac

Price Data

• Basic reformatting of the dataframe through Pandas for improved ease of use

 

Modelling

• Model done using Tensorflow CNN
• Baseline model accuracy ~85%
• Final model accuracy ~98%

Finalised Model Structure

model = tf.keras.Sequential([data_augmentation])
model.add(Conv2D(input_shape=(img_height,img_width,3),filters=64,kernel_size=(3,3),padding="same", activation="relu", kernel_regularizer=regularizers.l1_l2(l1=1e-5, l2=1e-4)))
model.add(MaxPooling2D(pool_size=2,)) 
model.add(Dropout(0.2))
model.add(Conv2D(kernel_size = 2, filters = 64, activation='relu', kernel_regularizer=regularizers.l1_l2(l1=1e-5, l2=1e-4)))
model.add(Conv2D(kernel_size = 2, filters = 64, activation='relu', kernel_regularizer=regularizers.l1_l2(l1=1e-5, l2=1e-4)))
model.add(MaxPooling2D(pool_size=2))
model.add(Dropout(0.2))
model.add(Conv2D(kernel_size = 2, filters = 128, activation='relu', kernel_regularizer=regularizers.l1_l2(l1=1e-5, l2=1e-4)))
model.add(Conv2D(kernel_size = 2, filters = 128, activation='relu', kernel_regularizer=regularizers.l1_l2(l1=1e-5, l2=1e-4)))
model.add(MaxPooling2D(pool_size=2))
model.add(Dropout(0.2))
model.add(Conv2D(kernel_size = 2, filters = 256, activation='relu', kernel_regularizer=regularizers.l1_l2(l1=1e-5, l2=1e-4)))
model.add(Conv2D(kernel_size = 2, filters = 256, activation='relu', kernel_regularizer=regularizers.l1_l2(l1=1e-5, l2=1e-4)))
model.add(MaxPooling2D(pool_size = 2))
model.add(Dropout(0.2))
model.add(GlobalMaxPooling2D())
model.add(Dense(num_classes, activation = 'softmax', kernel_regularizer=regularizers.l1_l2(l1=1e-5, l2=1e-4)))

Deployment / Production

• Deployed using Streamlit, hosted with Google Cloud Platform VM
• VM Link to App: Beer Price Checker
  • Live deployment until end of June
• Streamlit Share Link to App: Beer Price Checker
• User uploaded photo will output a prediction and display retail prices for the predicted brand

Sample Data Input

Sample Data Output

Future Improvements

• Expand existing dataset
  • Include more beer brands
  • Larger selection of grocery categories (eg. chips, cleaning products)
  • More price sources
• Improve recognition ability of model
  • Expand capabilities to predict poorly taken photographs
• UI development
  • More user friendly way to filter returned results
  • Improve the overall visual experience

Coding Language & Main Libraries

Python Version:
Python 3.7.10 (Google Colab)

Main Libraries Used:
Tensorflow 2.4.1
Inbac 2.1.0
Opencv-python 4.1.2
Streamlit 0.78.0