## 📈 **Evaluation Criteria**
Your submission will be evaluated using **Mean Average Precision (mAP@50)**, a standard metric in object detection (check the sample_submission.csv for more clarity).

### **What is mAP@50?**
- **Mean Average Precision (mAP)** measures the accuracy of object detection models.
- The **IoU (Intersection over Union) threshold is set at 0.5**.
- Higher `mAP@50` means your model performs better in detecting objects.

### [**How is Your Score Calculated?**](https://falcon.duality.ai/secure/documentation/ex2-colab-testing-output?sidebarMode=learn&utm_source=kaggle&utm_medium=kaggle_competition&utm_campaign=competition_2)
- **Predicted bounding boxes** compared to **ground truth labels**.
     - For **images with obstructions**, the bounding box should be placed over the **visible portion** of the soup can.
- The **IoU threshold is set at 0.5**, meaning a detection is considered correct if it overlaps with the true object by at least **50%**.
- Your final score is based on the **average precision across all test images**.

---

## 📥 **Submission Format**
To participate, submit your csv file with **predictions in YOLO format**.  

Each **test image** should have a corresponding "image_id" and a "prediction_string" with predicted bounding boxes.  

### **CSV File Format**
Each prediction should contain predictions in **YOLO format**:
"image_id", "prediction string(object class  class_confidence 
 x_center  y_center  width  height)".

- **image_id** →  image name without the .jpg/.png extension
- **Object class** →  `0` for cheerios, `1` for soup
- **confidence** → Prediction confidence (between `0` and `1`).
- **x_center, y_center, width, height** → Coordinates normalized between `0` and `1`.

#### **Example (`0002.png` for a single detected object)**
"0002","0 0.9435517191886902 0.874369642469618 0.30810502370198567 0.25126071506076386 0.25204394658406576"