Car_Brand_Classification

This is a Image Classification Technique and classified with Deep Learning which classifies car which belongs to any of the 10 classes Those classes are Audi , Lamborghini , Mercedes , KIA , Suzuki , Tata , Ford , Lexus , Honda , Mahindra

The front end is developed with the help of Gradio which provides an Interface which is readily available for Data Scientists which avoids using HTML,CSS,JavaScript and this is mainly useful for POC purpose and this classification is done using Creating the Architecture from scratch and then shifted to the transfer learning techniques such as InceptionV3 and VGG16 for getting better prediction

The below page is the Front End for the Application which is developed using Gradio

The next image is the continution image here we have the various classes of cars as the sample images

Now we will test on one of the image and see the prediction from both the Transfer Learning Architectures The top one is InceptionV3 and the bottom one is the VGG16 and here we ouput the top 3 classes classified for the cars

Testing and getting the results

Here we ca see that both the architectures gave us the right predictions for the car but there are some errors and these two techniques gives us good performance and also helps us to identify the classes of the cars accurately

Front End Tool
Gradio

IDE
Jupyter Notebook

Deep Learning Framewrok
Tensorflow

         

Work Flow for the Project

Data Collection

All these images are being scraped from the web using the Simple Image download module from python and these images are of two extensions jpeg and png and these images are provided by various websites

Data Preprocessing

The data is split into training and test and the training data is applied with various tranformations such as horizontal flip , zoom in , zoom out , shear range and scaling etc
The test data is scaling as we cant apply any transformations to the test data

Model Creation

Total Three Architectures are created one is the base model and which is created from scratch and other two are the transfer learning techniques such as InceptionV3 and VGG16 which improved the performance of the model

The accuarcies of the models used for classification :

Architectures             Accuracies
1.Base Model                     91%
2.InceptionV3                     98%
3.VGG16                              98%

If you like my project and want to know more about it I provided my LinkedIn Profile happy to connect

My linkedIn profile