AIOZ AI Pothole Detection Challenge Overview
Potholes aren’t just annoying. They cause accidents, damage tires, and cost cities millions every year.
The Pothole Detection Challenge puts you in the driver’s seat: Build a computer-vision model that automatically finds potholes in real road photos and draws precise boxes around each one.
What Makes This Challenge Tough and Exciting
You won’t be training on clean, uniform footage. The images come from:
- City streets, highways, and country roads
- Daytime, nighttime, rainfall, glare, shadows, etc.
- Dash-cams, roadside CCTV, and even drone footage
A winning model must:
- Read subtle texture under harsh lighting
- Ignore cracks, patches, and stains that aren’t real potholes
- Stay quiet on perfectly smooth asphalt (no false alarms!)
What You Get: Data & Labels
- Training set: Thousands of real road photos + one small text file per image. Each line in the label file represents a single pothole: 1 0.45 0.62 0.18 0.12
(class 1 = pothole | rest = normalized center-x, center-y, width, height)
- Test set: Just images. Labels are hidden to ensure fair scoring.
Two Ways to Submit
1. Quick public leaderboard score
Upload a simple CSV with two columns: id, predict_label
Example row (two potholes detected in one image):
000123,1 0.98 0.45 0.32 0.18 0.12 1 0.87 0.71 0.55 0.20 0.09
All detections go into the same cell, separated by spaces.
2. Final private ranking (what really counts)
Upload your full code and model weights. The system runs your code on a hidden test set to prevent anyone from overfitting the public leaderboard.
How Winners are Chosen: Scoring
mAP@0.5 = Mean Average Precision at 50% overlap
- Your model gives every detection a confidence score (0–1)
- A detection is correct only if the box drawn overlaps a real pothole by at least 50%
- Final ranking is based on the private test set only
Timeline: Mark Your Calendar
| Phase | Dates |
|---|---|
| Registration | Dec 3 – Dec 16, 2025 |
| Training | Dec 17 – Dec 30, 2025 |
| Validation | Dec 31, 2025 – Jan 13, 2026 |
| Final Submission | Jan 14 – Jan 20, 2026 |
| Judging | Jan 21 – Feb 3, 2026 |
| Results | Feb 4, 2026 |
Reward Structure:
Special Winner: 2,000 AIOZ Tokens
Top 5 Ranked Winners: 500 AIOZ Tokens for each
In total: 4,500 AIOZ Tokens
Ready to Make Roads Safer?
This challenge is a real civic problem you can help to solve.
From harsh lighting to multi-angle robustness, every hurdle you overcome brings us closer to smarter, safer streets.
Registration opened: December 3rd, 2025.
Jump in now: https://aiozai.network/challenges/pothole-detection-challenge-5598
Let’s patch the planet, one pothole at a time!
Related Content
Lightweight Text Generation with SmolLM-135M: Fast, Compact, and Capable
Text generation remains one of the most widely used AI capabilities. From drafting articles and composing captions to structuring short narratives and writing stories, creators and builders are constantly seeking models that can deliver high-quality text with minimal computational resources. SmolLM-135M introduces compact and efficient text generation that makes high-quality language synthesis more accessible and practical for real-world applications. About SmolLM-135M SmolLM-135M is a light
Archer Image Generator: Authentic Archer-Style Artwork Made Simple
Now available on AIOZ AI—the collaborative marketplace powered by AIOZ DePIN—Archer Image Generator is a specialized text-to-image model designed to produce illustrations with sharp lines, flat shading, and the punchy, animated look fans associate with the TV show Archer. Trained on screenshots from the series alongside AI-generated images and user-contributed content, it captures the show’s unique look and feel by including “Archer style” tokens in your prompts. Whether you’re a fan of the ser
Bring Your Photos To Life With Cartoonize Image Diffusion
Now available on AIOZ AI V1—the collaborative marketplace powered by AIOZ DePIN—the Cartoonize Image Diffusion model transforms real photos into vibrant, stylized cartoons using simple, natural-language instructions. This customized diffusion model builds on Stable Diffusion 1.5 with instruction-tuning techniques from FLAN and the conditional editing approach of InstructPix2Pix, enabling direct & high-fidelity cartoonization without per-image fine-tuning. It excels at interpreting textual promp
XFeat: Accelerated Features for Lightweight Image Matching
Now available on AIOZ AI—the collaborative marketplace powered by AIOZ DePIN—the XFeat model delivers fast, lightweight, and accurate feature detection and matching for images captured from different viewpoints. Designed for efficiency, XFeat extracts discriminative keypoints and descriptors before performing rapid correspondence matching. This method makes it well-suited for resource-constrained environments where speed and reliability matter. Hosted on AIOZ AI using the PyTorch framework, XF
Color Harmonization: Smarter Color Control For More Coherent Designs
Now available on AIOZ AI—the collaborative marketplace powered by AIOZ DePIN—the Color Harmonization model transforms images by adjusting and enhancing color balance according to harmony principles, creating visually captivating and aesthetically balanced compositions. This computational model applies selected harmony templates to align colors, improving coherence while preserving details and visual impact. Based on the work of Amir Hossein Kargaran and implemented in PyTorch, it excels in ima
Git over SSH And Git LFS: A Major Update on AIOZ AI
AIOZ AI is rolling out a powerful new capability: full support for Git over SSH (Secure Shell) with Git LFS (Large File Storage). Developers and creators can now manage source code and large AI assets - model weights, datasets, media files - directly on the AIOZ AI platform with speed, security, and zero friction. This is version control built for modern AI workflows. Why This Update Matters AI projects are large, complex, and resource-heavy. Traditional Git isn’t built for multi-gigabyte fi