Competition Overview

Competition Overview

For full text in indonesian, see petunjuk-teknis.md.

Task

Classify waste images into 3 categories:

• Recyclable — non-electronic recyclable waste (bottles, cans, paper, cardboard, glass)
• Electronic — e-waste, working or broken (phones, laptops, keyboards, cables, etc.)
• Organic — biodegradable material (leaves, fruit, veg, food scraps)

Dataset

• Train: 26,527 labeled images, one subfolder per class
• Test: 1,458 unlabeled images, filenames ordered 1–1458 matching template.csv
• Download: https://bit.ly/datasetbdc2026

Rules

• Only visual image content may be used as features — no metadata, no external labels
• No external labeled image data allowed for training
• Pretrained backbones OK (EfficientNet, ResNet, ConvNeXt, ViT, etc.) as long as they weren't trained on the competition's train/test data — must be documented in the final report
• Test data may only be used for final inference, never for training/validation/tuning
• Preprocessing/augmentation allowed on train data only

Submission

• Format: CSV, columns id, predicted
• Codes: 0 = Recyclable, 1 = Electronic, 2 = Organic
• Template: https://bit.ly/submissionbdc2026
• Max 3 submissions per team; highest score counts
• Submission window opens 8 July 2026
• Metric: Macro-averaged F1-score
• Deadline: 30 July 2026, 16:00 WIB, no extensions

Selection

• Up to 3 teams per university advance
• Ties broken by earliest submission
• 22 teams total selected for semifinal
• Selected teams must prove work wasn't done manually (video proof required)