Automatic Brand LOGO detection using Python

Published on 16 May 2022. Written by Sneha Kashyap

Signature and Brand logo act as a significant content for numerous documents specially for scanned documents. Brand recognition in pictures and videos is the key drawback in an exceedingly very large choice of applications, like infringement detection, discourse advertises placement, vehicle brand for intelligent traffic-control systems, machine-controlled computation of brand-related statistics on social media, etc. Historically, Brand logo recognition has been self-addressed with key point-based detectors and descriptors. This technique for brand recognition uses deep learning. Our recognition pipeline consists of a brand region proposal followed by a framework of Python known as PyTorch specifically trained for brand classification, whether or not they are exactly localized.

Read more..

SLNOTE


Project Description

In deep learning, it’s all regarding making, training, and deploying a network.

PyTorch:

The deep learning framework of Python is termed as Pytorch. It comes with Autograd- associate automotive vehicle reckon gradients. It’s equipped with tools to form and train deep learning simply and expeditiously. It conjointly supports GPU (Graphic Process Unit). It supports UNIX system, Mac, and Windows.

Tensor:

Tensor is a prime system of PyTorch. It’s Multi-dimensional matrix, virtually like Numpy’s ndarrays however able to run on GPU to accelerate computing.

Modules used in this project

Logo detection framework has many modules that are:

SLLATEST
Data assortment

• Import needed knowledge library.
• Outline many directories.
• Produce load_datasets utility perform to load knowledge set.

Prepare knowledge for Network:

• Produce list_image_paths utility perform to scan relative image file paths from the document into list variables.
• Add train_logo_relpaths and [*fr1] val_logo_relpaths to train_relpaths.
• Add train_logo_relpaths and conjointly the partner of val_logo_relpaths to val_relpaths.
• Use datasets.ImageFolder( ) that most popular directory structure as dataset/classes/img.jpg .
• Produce prepare_datasets utility perform to repeat image files in line with the lists of relative ways to the most popular directory structure.
• Import torch and torchvision libraries.
• Outline data_transforms that size, convert to tensor, and normalize inputs by mean and variance of employment dataset.
• The datasets created victimization torchvision.datasets.ImgFolder with argu-dataset directories and data_transform.
• Produce data loaders victimization DataLoader
• Now, creation of imshow utility perform to show an image.

Create Network:

• Produce our network by taxonomic category nn.Module.
• Outline redness technique by produce conv1 layer victimization nn.Conv2d with arguments-3 in-channels, half-dozen out-channels, five x 5-pixel filter with stride=1 as default, conv2 layer with arguments–6 in-channels, sixteen out channels, and same size filter.
• Produce pool layer victimization nn.MaxPool2d with arguments-2 x 2-pixel filter, stride=2.
• Produce totally connected layer fc1 victimization nn.Linear.
• Before instantiating our network, use torch.device to notice GPU if it’s obtainable otherwise use central processing unit. Then set our network to the device. Train the Network
• Import torch.optim
• Outline criterion or loss perform victimization nn.CrossEntropyLoss that already incorporates softmax perform with entropy loss.
• Produce train_val perform to educate network on coaching dataset and measure network on the validation dataset.
• Now, the model of information set is used for eval mode for evaluating.
• Train and validate the network.

Evaluate

• Analysis of the check series of information set victimization check performs that is exclusive virtually like validate perform.
• When the analysis of the data accuracy of the info is corrected by the framework.

Software Requirements:

• Front end: Python3
• Back end: Python 3.6.0 interpreter
• Graphical User Interface (GUI) Toolkit
• PyTorch Module

Hardware Components:

• Processor – Intel core i3 or above
• Hard Disk – 128 GB
• Memory – 1 GB RAM.

Advantages:

• This technique helps the corporate or user’s to observe their promoting efforts.
• Effective way to measure complete brand awareness.

Disadvantages:

• This framework wants an immense network to install.
• Sometimes get confused in detecting the fizzy logo or signature.

SLDYK
Kit required to develop Automatic Brand LOGO detection using Python:

• No kit required

Technologies you will learn by working on Automatic Brand LOGO detection using Python:

• Python

Any Questions?