Emoji Segmentation

🏞️ Emoji Segmentation 🖼️ + 😀

For this problem, you need to implement a model capable of semantic segmentation
on images with geometric backgrounds containing emojis. The model should generate a binary mask
where pixels corresponding to emojis are 255 (white) and the background is 0 (black).

---

🔹 Dataset

The dataset is provided as images and masks:

• train.csv: contains examples for model training

  Column	Type	Description
  SampleID	string	Unique identifier for each example
  Path	string	Path to the image containing the emoji
  Mask	string	Path to the corresponding binary mask

• test.csv: contains examples for prediction

  Column	Type	Description
  SampleID	string	Unique identifier
  Path	string	Path to the image containing the emoji

---

⚙️ Evaluation Metric

The metric used for this problem is:

• IoU (Intersection over Union): for each image, IoU is computed between the predicted mask and the ground-truth mask.

The total score is calculated as the mean IoU across all test images.

Scoring

1. Compute IoU (Intersection over Union) for each real-predicted mask pair.
2. Compute the mean IoU across all samples.
3. Convert the mean IoU to points:

• mean IoU >= 0.75 → 100 points
• 0.50 ≤ mean IoU < 0.75 → 50–99 points proportionally
• 0.25 ≤ mean IoU < 0.50 → 25–49 points proportionally
• mean IoU < 0.25 → 0–24 points proportionally

---

📨 Submission Format

The submission should be a zip archive containing a directory with the predicted segmentation masks and a CSV file named submission.csv mapping SampleID to predicted mask path (column name must be PredictedMask) for each test image:

Column	Type	Description
SampleID	string	Unique identifier for the test row
PredictedMask	string	Path to the predicted mask image (PNG)

🔹 Example submission.csv

SampleID	PredictedMask
test_0	predictions/test_0_mask.png
test_1	predictions/test_1_mask.png
test_2	predictions/test_2_mask.png

---

🖼️ Examples

Original Image	Mask