The Glitch Hunter

🕵️‍♂️ Mission Briefing

Agent, listen up.

The simulation is degrading. We've detected unauthorized modifications to the visual cortex of the mainframe, what the tech boys call "glitches". These aren't just random errors, they are deliberate insertions of chaos.

We believe that our novel AI model, Project: G3P1T, is trying to communicate... or maybe it's just bored. It's pasting random noise, inverting colors, and generally messing up our beautiful 4K wallpapers.

Your mission, should you choose to accept it (and you don't really have a choice, do you?), is to build a system that can identify and segment these glitches.

🎯 The Objective

We will provide you with a set of images. Each image has been corrupted with one or more "glitches".
A glitch can be:

• A block of random digital noise.
• A rectangular region with inverted colors.
• A solid block of color that definitely doesn't belong there.

Your task is to generate a binary mask for each image, where white pixels (255) represent the glitch, and black pixels (0) represent the clean image.

📂 The Data

You will be provided with:
Training Set (train/ folder):

• 200 corrupted images with their corresponding masks
• Use these to train your glitch detection model

Test Set (test/ folder):

• 200 corrupted images WITHOUT masks
• Generate predictions for these images

📤 Submission Format

You must submit a ZIP file containing your predicted masks.

• The masks must be PNG images.
• The filenames must match the input image IDs (e.g., if input is img_001.png, output must be img_001.png).
• The masks should be binary (0 for background, 255 for glitch).

📏 Evaluation Metric

We will evaluate your performance using Intersection over Union (IoU).

• IoU = (Intersection of your mask and ours) / (Union of your mask and ours)

If your IoU is high, the simulation stabilizes.
If your IoU is low... well, let's just say you might be reassigned to "garbage collection" in the lower memory sectors.

🚀 Getting Started

1. Download the dataset.
2. Train your miraculous model.
3. Generate masks.
4. Zip them up.
5. Submit and pray.

Good luck, Agent. The Matrix is watching.