Logo Detection

Model by Modzy

This model detects popular brand logos within images. This model can be used to search, edit, or count the number of logos that occur in a set of images or video frames.

  • Description

    Product Description

    PERFORMANCE METRICS:

    36.7% Mean Average Precision

    This model was trained on an 80% subset of the Logos in the Wild dataset. This large-scale dataset contains web collected images with annotations for 871 brands. This model was tested on 20% of the images and achieves a mean Average Precision (mAP) score of 0.367.

    A higher precision score indicates that the majority of labels predicted by the model for different classes are accurate. Further information here.

    OVERVIEW:

    This model is based on the YOLOv3 deep neural network. YOLOv3 is a single-shot object detection architecture, only requiring a single forward pass of the input through the network in order to make detection predictions. In order to detect objects of different sizes, YOLOv3 detects objects at 3 different scales at different branches throughout its architecture. The prediction branch at each scale is responsible for predicting candidate bounding boxes and object classes. These candidate bounding boxes are then consolidated using a technique known as Non-Maximum Suppression to generate the final output.

    TRAINING:

    This model was trained on a subset of the Logos in the Wild dataset, which consists of labeled bounding boxes and their corresponding brand names in 6,809 “in the wild” images. Each image contains varying numbers of logo classes and instances. Roughly 80% of these images were used for training. The model was trained for 55 epochs on an NVIDIA Tesla V100 GPU, using Adam optimization with an initial learning rate 0.001.

    VALIDATION:

    Validation was performed on roughly 20% of the dataset.

    INPUT SPECIFICATION

    The input(s) to this model must adhere to the following specifications:

    Filename Maximum Size Accepted Format(s)
    image
    20M .png, .jpg, .jpeg

    OUTPUT DETAILS

    This model will output the following:

    Filename Maximum Size Format
    results.json 10M .json

    The output file (“results.json”) will contain detected logo bounding boxes. Each bounding box will contain the corresponding logo class name, confidence score, and top left/bottom right x,y coordinates defining the box. This model can detect the following logos:

    H&M ABUS Accenture Adidas
    Aldi Allianz Allianz Aral
    Audi BEM Bionade BMW
    Bosch Bosch Budweiser Burger King
    Burger King Canon Cartier Caterpillar
    Chanel Chanel Coca-Cola Colgate
    CVS Pharmacy Esso Esso FedEx
    Gillette Heineken Heineken Honda
    HSBC HSBC Huawei Huawei
    Hyundai Intel Kellog’s Kia
    Kraft Lay’s Lexus L’Oréal
    Marlboro McDonald’s McDonald’s Mercedes-Benz
    Nescafé Nike Nissan NIVEA
    Pampers Pepsi Pepsi Pepsi
    Philips Pizza Hut Pizza Hut Puma
    Red Bull Red Bull Reebox Rolex
    Rolex Samsung Santander Shell
    Shell Starbucks Starbucks Subway
    T-Mobile Target Target Telekom
    Toyota Toyota Uniqlo Uniqlo
    UPS Visa Volkswagen Wells Fargo
    Zara