Humans can easily detect and identify objects present in an image. The human visual system is fast and accurate and can perform complex tasks like identifying multiple objects and detect obstacles with little conscious thought. With the availability of large amounts of data, faster GPUs, and better algorithms, we can now easily train computers to detect and classify multiple objects within an image with high accuracy. In this blog, we will explore terms such as object detection, object localization, loss function for object detection and localization, and finally explore an object detection algorithm known as “You only look once” (YOLO).

Object Localization

An image classification or image recognition model simply detect the probability of an object in an image. In contrast to this, object localization refers to identifying the location of an object in the image. An object localization algorithm will output the coordinates of the location of an object with respect to the image. In computer vision, the most popular way to localize an object in an image is to represent its location with the help of bounding boxes. Fig. 1 shows an example of a bounding box.

bounding box, object detection, localization, self driving cars, computer vision, deep learning, classfication — Fig 1. Bounding box representation used for object localization

A bounding box can be initialized using the following parameters:

bx, by : coordinates of the center of the bounding box
bw : width of the bounding box w.r.t the image width
bh : height of the bounding box w.r.t the image height

Defining the target variable

The target variable for a multi-class image classification problem is defined as:

Loss Function

Since we have defined both the target variable and the loss function, we can now use neural networks to both classify and localize objects.

Machine learning challenge, ML challenge

Object Detection

An approach to building an object detection is to first build a classifier that can classify closely cropped images of an object. Fig 2. shows an example of such a model, where a model is trained on a dataset of closely cropped images of a car and the model predicts the probability of an image being a car.

Now, we can use this model to detect cars using a sliding window mechanism. In a sliding window mechanism, we use a sliding window (similar to the one used in convolutional networks) and crop a part of the image in each slide. The size of the crop is the same as the size of the sliding window. Each cropped image is then passed to a ConvNet model (similar to the one shown in Fig 2.), which in turn predicts the probability of the cropped image is a car.

After running the sliding window through the whole image, we resize the sliding window and run it again over the image again. We repeat this process multiple times. Since we crop through a number of images and pass it through the ConvNet, this approach is both computationally expensive and time-consuming, making the whole process really slow. Convolutional implementation of the sliding window helps resolve this problem.

Convolutional implementation of sliding windows

Before we discuss the implementation of the sliding window using convents, let’s analyze how we can convert the fully connected layers of the network into convolutional layers. Fig. 4 shows a simple convolutional network with two fully connected layers each of shape (400, ).

convolutional sliding window, sliding window, 1d convolution, yolo, object detection — Fig 4. Sliding windows mechanism

A fully connected layer can be converted to a convolutional layer with the help of a 1D convolutional layer. The width and height of this layer are equal to one and the number of filters are equal to the shape of the fully connected layer. An example of this is shown in Fig 5.

full connected layer to 1d convolution, 1 d convolution, full connected layers, dense layers — Fig 5. Converting a fully connected layer into a convolutional layer

We can apply this concept of conversion of a fully connected layer into a convolutional layer to the model by replacing the fully connected layer with a 1-D convolutional layer. The number of the filters of the 1D convolutional layer is equal to the shape of the fully connected layer. This representation is shown in Fig 6. Also, the output softmax layer is also a convolutional layer of shape (1, 1, 4), where 4 is the number of classes to predict.

full convolutional networks , converting dense layers to convolutional layers, computer vision, object detection, object localization — Fig 6. Convolutional representation of fully connected layers.

Now, let’s extend the above approach to implement a convolutional version of sliding window. First, let’s consider the ConvNet that we have trained to be in the following representation (no fully connected layers).

Let’s assume the size of the input image to be 16× 16× 3. If we’re to use a sliding window approach, then we would have passed this image to the above ConvNet four times, where each time the sliding window crops a part of the input image of size 14× 14× 3 and pass it through the ConvNet. But instead of this, we feed the full image (with shape 16× 16 × 3) directly into the trained ConvNet (see Fig. 7). This results in an output matrix of shape 2 × 2 × 4. Each cell in the output matrix represents the result of a possible crop and the classified value of the cropped image. For example, the left cell of the output (the green one) in Fig. 7 represents the result of the first sliding window. The other cells represent the results of the remaining sliding window operations.

Convolutional sliding window, fully convolutional network, sliding window, object detection, object localization, yolo, rcnn, computer vision, perception, self driving cars — Fig 7. Convolutional implementation of the sliding window

Note that the stride of the sliding window is decided by the number of filters used in the Max Pool layer. In the example above, the Max Pool layer has two filters, and as a result, the sliding window moves with a stride of two resulting in four possible outputs. The main advantage of using this technique is that the sliding window runs and computes all values simultaneously. Consequently, this technique is really fast. Although a weakness of this technique is that the position of the bounding boxes is not very accurate.

The YOLO (You Only Look Once) Algorithm

A better algorithm that tackles the issue of predicting accurate bounding boxes while using the convolutional sliding window technique is the YOLO algorithm. YOLO stands for you only look once and was developed in 2015 by Joseph Redmon, Santosh Divvala, Ross Girshick, and Ali Farhadi. It’s popular because it achieves high accuracy while running in real time. This algorithm is called so because it requires only one forward propagation pass through the network to make the predictions.

The algorithm divides the image into grids and runs the image classification and localization algorithm (discussed under object localization) on each of the grid cells. For example, we have an input image of size 256 × 256. We place a 3× 3 grid on the image (see Fig. 8).

YOLO algorithm, you only look once, Joseph Redmon, Computer vision, pattern recognition, Real time object detection — Fig. 8 Grid (3 x 3) representation of the image

Next, we apply the image classification and localization algorithm on each grid cell. For each grid cell, the target variable is defined as

Do everything once with the convolution sliding window. Since the shape of the target variable for each grid cell is 1 × 9 and there are 9 (3 × 3) grid cells, the final output of the model will be:

The advantages of the YOLO algorithm is that it is very fast and predicts much more accurate bounding boxes. Also, in practice to get more accurate predictions, we use a much finer grid, say 19 × 19, in which case the target output is of the shape 19 × 19 × 9.

Conclusion

With this, we come to the end of the introduction to object detection. We now have a better understanding of how we can localize objects while classifying them in an image. We also learned to combine the concept of classification and localization with the convolutional implementation of the sliding window to build an object detection system. In the next blog, we will go deeper into the YOLO algorithm, loss function used, and implement some ideas that make the YOLO algorithm better. Also, we will learn to implement the YOLO algorithm in real time.

Have anything to say? Feel free to comment below for any questions, suggestions, and discussions related to this article. Till then, keep hacking with HackerEarth.

I Used AI to Build a "Simple Image Carousel" at VibeCodeArena. It Found 15+ Issues and Taught Me How to Fix Them.

‍

My Learning Journey

I wanted to understand what separates working code from good code. So I used VibeCodeArena.ai to pick a problem statement where different LLMs produce code for the same prompt. Upon landing on the main page of VibeCodeArena, I could see different challenges. Since I was interested in an Image carousal application, I picked the challenge with the prompt "Make a simple image carousel that lets users click 'next' and 'previous' buttons to cycle through images."

‍

Within seconds, I had code from multiple LLMs, including DeepSeek, Mistral, GPT, and Llama. Each code sample also had an objective evaluation score. I was pleasantly surprised to see so many solutions for the same problem. I picked gpt-oss-20b model from OpenAI. For this experiment, I wanted to focus on learning how to code better so either one of the LLMs could have worked. But VibeCodeArena can also be used to evaluate different LLMs to help make a decision about which model to use for what problem statement.

The model had produced a clean HTML, CSS, and JavaScript. The code looked professional. I could see the preview of the code by clicking on the render icon. It worked perfectly in my browser. The carousel was smooth, and the images loaded beautifully.

But was it actually good code?

I had no idea. That's when I decided to look at the evaluation metrics

What I Thought Was "Good Code"

A working image carousel with:

Clean, semantic HTML
Smooth CSS transitions
Keyboard navigation support
ARIA labels for accessibility
Error handling for failed images

It looked like something a senior developer would write. But I had questions:

Was it secure? Was it optimized? Would it scale? Were there better ways to structure it?

Without objective evaluation, I had no answers. So, I proceeded to look at the detailed evaluation metrics for this code

What VibeCodeArena's Evaluation Showed

The platform's objective evaluation revealed issues I never would have spotted:

Security Vulnerabilities (The Scary Ones)

No Content Security Policy (CSP): My carousel was wide open to XSS attacks. Anyone could inject malicious scripts through the image URLs or manipulate the DOM. VibeCodeArena flagged this immediately and recommended implementing CSP headers.

Missing Input Validation: The platform pointed out that while the code handles image errors, it doesn't validate or sanitize the image sources. A malicious actor could potentially exploit this.

Hardcoded Configuration: Image URLs and settings were hardcoded directly in the code. The platform recommended using environment variables instead - a best practice I completely overlooked.

SQL Injection Vulnerability Patterns: Even though this carousel doesn't use a database, the platform flagged coding patterns that could lead to SQL injection in similar contexts. This kind of forward-thinking analysis helps prevent copy-paste security disasters.

Performance Problems (The Silent Killers)

DOM Structure Depth (15 levels): VibeCodeArena measured my DOM at 15 levels deep. I had no idea. This creates unnecessary rendering overhead that would get worse as the carousel scales.

Expensive DOM Queries: The JavaScript was repeatedly querying the DOM without caching results. Under load, this would create performance bottlenecks I'd never notice in local testing.

Missing Performance Optimizations: The platform provided a checklist of optimizations I didn't even know existed:

No DNS-prefetch hints for external image domains
Missing width/height attributes causing layout shift
No preload directives for critical resources
Missing CSS containment properties
No will-change property for animated elements

Each of these seems minor, but together they compound into a poor user experience.

Code Quality Issues (The Technical Debt)

High Nesting Depth (4 levels): My JavaScript had logic nested 4 levels deep. VibeCodeArena flagged this as a maintainability concern and suggested flattening the logic.

Overly Specific CSS Selectors (depth: 9): My CSS had selectors 9 levels deep, making it brittle and hard to refactor. I thought I was being thorough; I was actually creating maintenance nightmares.

Code Duplication (7.9%): The platform detected nearly 8% code duplication across files. That's technical debt accumulating from day one.

Moderate Maintainability Index (67.5): While not terrible, the platform showed there's significant room for improvement in code maintainability.

Missing Best Practices (The Professional Touches)

The platform also flagged missing elements that separate hobby projects from professional code:

No 'use strict' directive in JavaScript
Missing package.json for dependency management
No test files
Missing README documentation
No .gitignore or version control setup
Could use functional array methods for cleaner code
Missing CSS animations for enhanced UX

The "Aha" Moment

Here's what hit me: I had no framework for evaluating code quality beyond "does it work?"

The carousel functioned. It was accessible. It had error handling. But I couldn't tell you if it was secure, optimized, or maintainable.

VibeCodeArena gave me that framework. It didn't just point out problems, it taught me what production-ready code looks like.

My New Workflow: The Learning Loop

This is when I discovered the real power of the platform. Here's my process now:

Step 1: Generate Code Using VibeCodeArena

I start with a prompt and let the AI generate the initial solution. This gives me a working baseline.

Step 2: Analyze Across Several Metrics

I can get comprehensive analysis across:

Security vulnerabilities
Performance/Efficiency issues
Performance optimization opportunities
Code Quality improvements

This is where I learn. Each issue includes explanation of why it matters and how to fix it.

Step 3: Click "Challenge" and Improve

Here's the game-changer: I click the "Challenge" button and start fixing the issues based on the suggestions. This turns passive reading into active learning.

Do I implement CSP headers correctly? Does flattening the nested logic actually improve readability? What happens when I add dns-prefetch hints?

I can even use AI to help improve my code. For this action, I can use from a list of several available models that don't need to be the same one that generated the code. This helps me to explore which models are good at what kind of tasks.

For my experiment, I decided to work on two suggestions provided by VibeCodeArena by preloading critical CSS/JS resources with <link rel="preload"> for faster rendering in index.html and by adding explicit width and height attributes to images to prevent layout shift in index.html. The code editor gave me change summary before I submitted by code for evaluation.

Step 4: Submit for Evaluation

After making improvements, I submit my code for evaluation. Now I see:

What actually improved (and by how much)
What new issues I might have introduced
Where I still have room to grow

Step 5: Hey, I Can Beat AI

My changes helped improve the performance metric of this simple code from 82% to 83% - Yay! But this was just one small change. I now believe that by acting upon multiple suggestions, I can easily improve the quality of the code that I write versus just relying on prompts.

Each improvement can move me up the leaderboard. I'm not just learning in isolation—I'm seeing how my solutions compare to other developers and AI models.

So, this is the loop: Generate → Analyze → Challenge → Improve → Measure → Repeat.

Every iteration makes me better at both evaluating AI code and writing better prompts.

‍

‍

What This Means for Learning to Code with AI

This experience taught me three critical lessons:

1. Working ≠ Good Code

AI models are incredible at generating code that functions. But "it works" tells you nothing about security, performance, or maintainability.

The gap between "functional" and "production-ready" is where real learning happens. VibeCodeArena makes that gap visible and teachable.

2. Improvement Requires Measurement

I used to iterate on code blindly: "This seems better... I think?"

Now I know exactly what improved. When I flatten nested logic, I see the maintainability index go up. When I add CSP headers, I see security scores improve. When I optimize selectors, I see performance gains.

Measurement transforms vague improvement into concrete progress.

3. Competition Accelerates Learning

The leaderboard changed everything for me. I'm not just trying to write "good enough" code—I'm trying to climb past other developers and even beat the AI models.

This competitive element keeps me pushing to learn one more optimization, fix one more issue, implement one more best practice.

How the Platform Helps Me Become A Better Programmer

VibeCodeArena isn't just an evaluation tool—it's a structured learning environment. Here's what makes it effective:

Immediate Feedback: I see issues the moment I submit code, not weeks later in code review.

Contextual Education: Each issue comes with explanation and guidance. I learn why something matters, not just that it's wrong.

Iterative Improvement: The "Challenge" button transforms evaluation into action. I learn by doing, not just reading.

Measurable Progress: I can track my improvement over time—both in code quality scores and leaderboard position.

Comparative Learning: Seeing how my solutions stack up against others shows me what's possible and motivates me to reach higher.

What I've Learned So Far

Through this iterative process, I've gained practical knowledge I never would have developed just reading documentation:

How to implement Content Security Policy correctly
Why DOM depth matters for rendering performance
What CSS containment does and when to use it
How to structure code for better maintainability
Which performance optimizations actually make a difference

Each "Challenge" cycle teaches me something new. And because I'm measuring the impact, I know what actually works.

The Bottom Line

AI coding tools are incredible for generating starting points. But they don't produce high quality code and can't teach you what good code looks like or how to improve it.

VibeCodeArena bridges that gap by providing:

✓ Objective analysis that shows you what's actually wrong
✓ Educational feedback that explains why it matters
✓ A "Challenge" system that turns learning into action
✓ Measurable improvement tracking so you know what works
✓ Competitive motivation through leaderboards

My "simple image carousel" taught me an important lesson: The real skill isn't generating code with AI. It's knowing how to evaluate it, improve it, and learn from the process.

The future of AI-assisted development isn't just about prompting better. It's about developing the judgment to make AI-generated code production-ready. That requires structured learning, objective feedback, and iterative improvement. And that's exactly what VibeCodeArena delivers.

Here is a link to the code for the image carousal I used for my learning journey

#AIcoding #WebDevelopment #CodeQuality #VibeCoding #SoftwareEngineering #LearningToCode

‍

Revolutionizing Mobile Talent Hiring: The HackerEarth Advantage

The demand for mobile applications is exploding, but finding and verifying developers with proven, real-world skills is more difficult than ever. Traditional assessment methods often fall short, failing to replicate the complexities of modern mobile development.

Introducing a New Era in Mobile Assessment

At HackerEarth, we're closing this critical gap with two groundbreaking features, seamlessly integrated into our Full Stack IDE:

Now, assess mobile developers in their true native environment. Our enhanced Full Stack questions now offer full support for both Java and Kotlin, the core languages powering the Android ecosystem. This allows you to evaluate candidates on authentic, real-world app development skills, moving beyond theoretical knowledge to practical application.

Say goodbye to setup drama and tool-switching. Candidates can now build, test, and debug Android and React Native applications directly within the browser-based IDE. This seamless, in-browser experience provides a true-to-life evaluation, saving valuable time for both candidates and your hiring team.

Assess the Skills That Truly Matter

With native Android support, your assessments can now delve into a candidate's ability to write clean, efficient, and functional code in the languages professional developers use daily. Kotlin's rapid adoption makes proficiency in it a key indicator of a forward-thinking candidate ready for modern mobile development.

Breakup of Mobile development skills ~95% of mobile app dev happens through Java and Kotlin — This chart illustrates the importance of assessing proficiency in both modern (Kotlin) and established (Java) codebases.

Streamlining Your Assessment Workflow

The integrated mobile emulator fundamentally transforms the assessment process. By eliminating the friction of fragmented toolchains and complex local setups, we enable a faster, more effective evaluation and a superior candidate experience.

Old Fragmented Way vs. The New, Integrated Way — Visualize the stark difference: Our streamlined workflow removes technical hurdles, allowing candidates to focus purely on demonstrating their coding and problem-solving abilities.

Quantifiable Impact on Hiring Success

A seamless and authentic assessment environment isn't just a convenience, it's a powerful catalyst for efficiency and better hiring outcomes. By removing technical barriers, candidates can focus entirely on demonstrating their skills, leading to faster submissions and higher-quality signals for your recruiters and hiring managers.

A Better Experience for Everyone

Our new features are meticulously designed to benefit the entire hiring ecosystem:

For Recruiters & Hiring Managers:

Accurately assess real-world development skills.
Gain deeper insights into candidate proficiency.
Hire with greater confidence and speed.
Reduce candidate drop-off from technical friction.

For Candidates:

Enjoy a seamless, efficient assessment experience.
No need to switch between different tools or manage complex setups.
Focus purely on showcasing skills, not environment configurations.
Work in a powerful, professional-grade IDE.

Unlock a New Era of Mobile Talent Assessment

Stop guessing and start hiring the best mobile developers with confidence. Explore how HackerEarth can transform your tech recruiting.

A New Era of Code

Vibe coding is a new method of using natural language prompts and AI tools to generate code. I have seen firsthand that this change makes software more accessible to everyone. In the past, being able to produce functional code was a strong advantage for developers. Today, when code is produced quickly through AI, the true value lies in designing, refining, and optimizing systems. Our role now goes beyond writing code; we must also ensure that our systems remain efficient and reliable.

From Machine Language to Natural Language

I recall the early days when every line of code was written manually. We progressed from machine language to high-level programming, and now we are beginning to interact with our tools using natural language. This development does not only increase speed but also changes how we approach problem solving. Product managers can now create working demos in hours instead of weeks, and founders have a clearer way of pitching their ideas with functional prototypes. It is important for us to rethink our role as developers and focus on architecture and system design rather than simply on typing c

The Promise and the Pitfalls

I have experienced both sides of vibe coding. In cases where the goal was to build a quick prototype or a simple internal tool, AI-generated code provided impressive results. Teams have been able to test new ideas and validate concepts much faster. However, when it comes to more complex systems that require careful planning and attention to detail, the output from AI can be problematic. I have seen situations where AI produces large volumes of code that become difficult to manage without significant human intervention.

AI-powered coding tools like GitHub Copilot and AWS’s Q Developer have demonstrated significant productivity gains. For instance, at the National Australia Bank, it’s reported that half of the production code is generated by Q Developer, allowing developers to focus on higher-level problem-solving . Similarly, platforms like Lovable or Hostinger Horizons enable non-coders to build viable tech businesses using natural language prompts, contributing to a shift where AI-generated code reduces the need for large engineering teams. However, there are challenges. AI-generated code can sometimes be verbose or lack the architectural discipline required for complex systems. While AI can rapidly produce prototypes or simple utilities, building large-scale systems still necessitates experienced engineers to refine and optimize the code.

The Economic Impact

The democratization of code generation is altering the economic landscape of software development. As AI tools become more prevalent, the value of average coding skills may diminish, potentially affecting salaries for entry-level positions. Conversely, developers who excel in system design, architecture, and optimization are likely to see increased demand and compensation.
Seizing the Opportunity

Vibe coding is most beneficial in areas such as rapid prototyping and building simple applications or internal tools. It frees up valuable time that we can then invest in higher-level tasks such as system architecture, security, and user experience. When used in the right context, AI becomes a helpful partner that accelerates the development process without replacing the need for skilled engineers.

This is revolutionizing our craft, much like the shift from machine language to assembly to high-level languages did in the past. AI can churn out code at lightning speed, but remember, “Any fool can write code that a computer can understand. Good programmers write code that humans can understand.” Use AI for rapid prototyping, but it’s your expertise that transforms raw output into robust, scalable software. By honing our skills in design and architecture, we ensure our work remains impactful and enduring. Let’s continue to learn, adapt, and build software that stands the test of time.

Ready to streamline your recruitment process? Get a free demo to explore cutting-edge solutions and resources for your hiring needs.

Introduction to Object Detection

Explore this post with: