# Dumpster Diving — Data

来源：https://www.kaggle.com/competitions/garbage-segmentation/data

采集时间：2026-10-19T07:19:45Z

This dataset brings together real-world and household waste images to create a challenging and diverse environment for image segmentation models. It contains annotations for five common waste categories.


**Components**
- TACO: A publicly available dataset featuring waste items labeled in varied environments, It includes naturally occurring occlusions, cluttered backgrounds, and lighting variations.

- Self-Collected Images: Images gathered from Estonian households to capture local waste management conditions

**Image Details**
- Image size: 512 × 512 pixels
- Images: Standard RGB images
- Masks: Color-coded segmentation masks or JSON annotations

Each image is paired with a corresponding mask identifying object locations and classes.

**Waste Categories**
1. Paper/Cardboard: Newspapers, magazines, cartons, and packaging materials
2. Plastic: Bottles, containers, bags, wrappers
3. Metal: Aluminum cans, tin containers, foil
4. Glass: Bottles, jars, broken glass
5. Other / Mixed Municipal Waste: Organic waste, composites, contaminated packaging, etc.


**Files Provided**
- train - Folder containing the train images
- train_json - JSON file that has the image_id & annotations for each train image
- test - Folder containing the images for which you have to predict the segmentation
- test-without_segmentation.json - JSON file inside which contains the image_id for each test image
- sample_submission.csv - a sample submission file in the correct format, only for the first test image