Comprehensive Solution for Vehicle Image Standardization and Enhancement

The biggest problem for any e-commerce platform in any segment is to standardise the images with good quality. DriveX is no different. Since, we are rapidly growing and our business is dependent on the refurbishment centers. Post refurb, the images of the vehicles are captured and pushed to the database. There are a lot of problems we have to standardise the images for our business

1. Unlike E-commerce, our every bike is unique , so we cannot go with stock images

2. With Growing outlets, the images of the bikes are randomly captured with different backgrounds and angles

3. Most of the times, the sequence how the images are captured is not properly streamlined

4. Cost and time to setup process, follow the SoP is very high considering the type of stakeholders we are working with

5. The images while coming to the website creating bad user experience and low CTRs

So, what have we done ?

Product and engineering wanted to solve the problem with limited dependency on the business, sales or operating verticals.

We came up with a fully automated process to solve the problem along with Cost, Speed and Complexity.

Lets Say, we have a given set of images uploaded in the database for a given vehicle like below -

Preprocessing:

• Resize images to a standard size.

• Normalize the image data.

We can do this using standard image processing tool.

Now, we need to ensure that we have identified the right angle of each of these images for which we have used :

Image Classification technique:

• Use a pre-trained Convolutional Neural Network (CNN) model (e.g., ResNet, VGG) fine-tuned for vehicle image classification.

• Classify images into predefined categories (front view, back view, left view, right view, etc.)

Image Description	File Path
Side View (Right)	/mnt/data/Screenshot 2024-07-20 at 1.01.29 AM.png
Side View (Right, Slightly Angled)	/mnt/data/Screenshot 2024-07-20 at 1.01.37 AM.png
Rear View (Slightly Angled)	/mnt/data/Screenshot 2024-07-20 at 1.01.43 AM.png
Rear View	/mnt/data/Screenshot 2024-07-20 at 1.01.48 AM.png
Rear View (Slightly Angled)	/mnt/data/Screenshot 2024-07-20 at 1.01.53 AM.png
Side View (Left, Slightly Angled)	/mnt/data/Screenshot 2024-07-20 at 1.01.58 AM.png
Side View (Left)	/mnt/data/Screenshot 2024-07-20 at 1.02.04 AM.png
Front View (Slightly Angled)	/mnt/data/Screenshot 2024-07-20 at 1.02.09 AM.png
Front View	/mnt/data/Screenshot 2024-07-20 at 1.02.15 AM.png
Side View (Right)	/mnt/data/Screenshot 2024-07-20 at 1.02.15 AM.png

Arranging the images

As we know the sequence to be followed, I will simply re-arrange the images using pre-defined rules

Vehicle Color Identification

Objective: Identify the primary color(s) of the vehicle.

Steps:

• Color Detection:

• Convert images to the HSV color space for better color segmentation.

• Use color clustering (e.g., K-means clustering) to identify dominant colors.

• Map detected colors to standard color names.

Part	Color	Description
Main Body and Tank	Predominantly Black	The main body and tank of the bike are predominantly black.
Accents	Silver/Gray	There are silver/gray accents, particularly on the tank and engine area.
Exhaust Pipes	Chrome/Silver	The exhaust pipes are chrome/silver.
Minor Details	Orange (Turn Signals)	There are also some minor details in orange, like the turn signals.

Based on the picked color, we need to define the color which matches the body

Vehicle Body Color	Background Color
Black	White
	Dark Gray
	Red
Silver/Gray	Dark Blue
	White
	Dark Gray
Chrome/Silver	Dark Green
	Dark Blue
	Black
Orange	Blue
	White
	Dark Gray

These inputs will go the prompt to change the background of the vehicle for example :

If we have enough images provided by the uploader, we can also create a

360-Degree View by

• Arrange images in sequence.

• Using MoviePy.

Category	Metric	Impact
Customer Engagement Metrics	Click-Through Rate (CTR)	Higher CTR on vehicle listings
Customer Engagement Metrics	Time on Page	Increased time users spend on each listing page
Customer Engagement Metrics	Bounce Rate	Reduced bounce rates
Conversion Metrics	Conversion Rate	Higher conversion rates
Conversion Metrics	Lead Generation	Increased number of inquiries and leads generated
Operational Efficiency Metrics	Processing Time	Reduced time required to prepare images for listings
Operational Efficiency Metrics	Manual Effort	Decreased manual effort in processing and standardizing images
Operational Efficiency Metrics	Error Rate	Reduced error rates in image categorization and standardization
Sales Metrics	Sales Volume	Increased overall sales volume
Sales Metrics	Average Selling Price	Higher selling prices
User Satisfaction Metrics	Customer Satisfaction Score (CSAT)	Improved overall customer satisfaction
User Satisfaction Metrics	Net Promoter Score (NPS)	Higher likelihood of recommendations, increasing NPS
Website Performance Metrics	Page Load Time	Improved page load times
Website Performance Metrics	Mobile Performance	Enhanced mobile user experience
Brand Perception Metrics	Brand Trust	Increased trust in the brand
Brand Perception Metrics	Brand Recognition	More recognizable and reliable brand image