# 🧠 Dataset Description

This dataset is designed for an object detection competition using the YOLO (You Only Look Once) format. It includes images and corresponding annotation files, split into training and test sets. Participants must train a model to detect and classify objects in the test images and submit their predictions in a specific CSV format.

## 📁 Files

- **dataset_kaggle.zip**  
  Contains the training data. Inside:
  - `images` — training images (`.jpg`)
  - `labels` — YOLO annotation files (`.txt`)
  - `classes.txt` — list of class names (one per line)
  - `notes.json` — Metadata

- **test_set_without_label**  
  Contains the test images. Inside:
  - `images/` — test images without labels

- **starter-notebook.ipynb**  
  A short notebook to help you submit your solution.

- **sample_submission.csv**  
  A template showing the expected submission format. Each row matches one image and contains predicted objects as a flat string in YOLO format.

## 🧾 Data Format

Each `.txt` file in `labels/` corresponds to one image and contains lines like:

<class_id> <x_center> <y_center> <width> <height>

- `class_id` refers to the index in `classes.txt` (starting from 0)
- All coordinates are **normalized** between `0` and `1`
- `x_center` and `y_center` are the center of the bounding box
- `width` and `height` are the size of the box relative to the image


**Example:**

1 0.45 0.50 0.20 0.15
0 0.70 0.75 0.10 0.12

## 🎯 Submission Format

Your final submission must be a CSV file with two columns:

Id,PredictionString
test_image_001.jpg,"0 0.5 0.5 0.2 0.2 2 0.7 0.7 0.1 0.1"
test_image_002.jpg,""

Where `PredictionString` contains zero or more objects for the image, each represented as:
<class_id> <x_center> <y_center> <width> <height>

Multiple objects are space-separated in a single string.

## ❓ FAQ

**What files do I need?**  
Use the training archive to train your model and the test archive to generate predictions.

**What am I predicting?**  
You are predicting bounding boxes and classes for all objects in each test image.

**What format should my submission be in?**  
A single CSV file like `sample_submission.csv`, with predicted bounding boxes in YOLO format.

**What is `classes.txt`?**  
A list of class names, one per line. The line number = `class_id`.


Good luck!