---

task_categories: - image-classification - object-detection - image-to-text - text-to-image language: - en - pt tags: - circular-economy - remanufacturing - fine-grained - multimodal - continual-learning - household-appliances size_categories: - 100K<n<1M

TudoBonus Dataset

⚠️ Note on this Release

This is a sample release of the TudoBonus dataset, made available for review purposes in the context of a submission to NeurIPS 2026 (Dataset and Benchmark Track). The full dataset will be released upon paper acceptance.

The current release includes:

• ✅ metadata.csv — full metadata file with all annotated attributes
• ✅ A representative sample of images organized by interaction_id
• ⏳ Full image set — to be released upon acceptance

---

Dataset Summary

TudoBonus is a large-scale multimodal dataset of remanufactured household appliances, designed for fine-grained classification, defect detection, cross-modal retrieval, and continual learning in the Circular Economy domain.

The full dataset comprises:

• 🖼️ 457,000 images across 492 SKUs
• 🏷️ 15 annotated metadata features per product
• 📝 Product descriptions in Portuguese and English
• 📦 16 product categories (refrigerators, washing machines, stoves, dishwashers, and more)
• 🔲 97,162 bounding box annotations for defect detection

---

Dataset Structure

The dataset is organized as follows: The column interaction_id in metadata.csv serves as the primary key linking each record to its corresponding image folder.

---

The column interaction_id in metadata.csv serves as the primary key linking each record to its corresponding image folder.

---

Metadata Fields

Column	Type	Description	Complete (%)
interaction_id	string	Links record to image folder	100.0
sku	string	Stock Keeping Unit identifier	100.0
product_type	categoric	Appliance category	100.0
department	categoric	Product department	100.0
brand	categoric	Product brand	100.0
color	categoric	Product color	100.0
voltage	categoric	Electrical voltage	100.0
height	float	Height in cm	100.0
width	float	Width in cm	100.0
depth	float	Depth in cm	100.0
weight	float	Weight in kg	100.0
description_pt	string	Portuguese description	97.1
original_value	float	Retail price in BRL	96.3
quality_label	categoric	Remanufacturing quality grade	94.2
power	float	Power in watts	75.9
bounding_box	list	Defect bbox [x, y, w, h]	24.1
description_en	string	English description (AI-translated)	24.1
view	categoric	Image viewpoint	24.1
created_at	date	Cataloguing date	100.0
split	categoric	Train/test split flag	100.0

---

Tasks and Baselines

TudoBonus supports the following tasks:

Task	Metric	Best Baseline
SKU Classification	Balanced Accuracy	0.60 (SigLIP2 / DINOv2)
Defect Detection	mAP50-95	18.40 (YOLOv8m)
Text-to-Image Retrieval	Recall@K	CLIP-based
Image-to-Text Retrieval	Recall@K	CLIP-based
Continual Learning	Overall Acc.	0.65 (DINOv2 Frozen Backbone)
Zero-shot Classification	Accuracy	45.21 (ALIGN + description)

---

Loading the Dataset

from datasets import load_dataset

ds = load_dataset("TudoBonus/tudobonus")

Or using the Croissant metadata file:

import mlcroissant as mlc

ds = mlc.Dataset(
    "https://huggingface.co/datasets/TudoBonus/tudobonus"
    "/resolve/main/croissant.json"
)
records = ds.records(record_set="products")
for record in records:
    print(record["sku"], record["interaction_id"])

---

Languages

• Product descriptions: Portuguese (original) and English (AI-translated)
• Metadata: English

---

License

This dataset is released under the Creative Commons Attribution 4.0 International (CC BY 4.0) license, permitting free use for research and commercial purposes with proper attribution.

---

Responsible AI

• Geographic bias: dataset reflects the Brazilian remanufacturing market, with predominantly regional brands
• Brand bias: Electrolux is the dominant brand (~224k images)
• Language bias: English descriptions are AI-translated and may contain subtle inaccuracies
• Intended use: computer vision research, circular economy automation
• Prohibited use: surveillance, personal identification

---

Citation

@dataset{tudobonus2026,
  title     = {TudoBonus: A Large-Scale Multimodal Benchmark for 
               Fine-Grained Classification, Retrieval, and Continual 
               Learning in the Circular Economy},
  year      = {2026},
  note      = {Submitted to NeurIPS 2026. Anonymous submission.}
}