Author: Amit Yadav

Date: 2021-10-14 11:00:00

mail: amitech90@gmail.com

project: Car logo classification

Final output

Video is also in the folder.

Architecture used: ResNet50, VGG19

Dataset: data   

            |__ train

            |__ test

Car Brands:

'Audi.common', 'BMW.common', 'Chevrolet.common', 'Datsun.common', 'Fiat.common', 'Ford.common', 'Honda.common', 'Hyundai.common', 'ISUZU.common', 'Jaguar.frontal', 'Jaguar.rear', 'Jeep.common', 'Kia.common', 'Kia.new', 'Mahindra.common', 'Maruti-Suzuki.common', 'Mercedes-Benz.common', 'MG-Motor.common', 'Mitsubishi.common', 'Nissan.common', 'Renault.common', 'Skoda.common', 'Tata.common', 'Tata.text', 'Toyota.common', 'unknown', 'Volkswagen.common', 'Volvo.frontal'

----------------------------------------------------------------------------------------------------------------------------------------------

ResNet50 Architecture:

Model: "resnet50"

code in: car_logo_resnet.ipynb

workflow:

    1. Load the images and visualize them

    2. Found a standard ratio and make it as standard input size for the model

    3. Data Augmentation

    4. Data flow pipeline

    5. Model building (used transfer learning)

    6. Model training

    7. Model evaluation

For 20 Epochs

Trained for 50 epoch but model peaked so earlystopped at 43 with patience 10

Output in folder: results_on_resnet.csv

------------------------------------------------------------------------------------------------------------------------------------------------

VGG19 Architecture:

Model: "vgg19"

code in car_logo_vgg19.ipynb

workflow:

    1. Load the images and visualize them

    2. Found a standard ratio and make it as standard input size for the model

    3. Data Augmentation

    4. Data flow pipeline

    5. Model building (used transfer learning)

    6. Model training

    7. Model evaluation

For 50 epochs

Confusion matrix for 50 epochs

Output in folder: results_on_vgg.csv

----------------------------------------------------------------------------------------------------------------------------------------------