What is e-waste? E-waste is electronic products that are near or at the end of their useful life and have been damaged. Computer, TV, copier, fax, etc. are among the products that we use on a daily basis.

Applying the best method to dispose of electronic devices is a challenge that the world has been facing since the 1970s, and today, electronic devices that are thrown away are much more than before, which are referred to as electronic waste.

In other words, e-waste is any used and broken electronic equipment that is thrown away and is very dangerous due to the toxic chemicals that naturally arise from the metals inside it when buried.

List of common e-waste items Home Appliances

microwave Home entertainment devices Electric ovens Heaters Fan Communication devices and information technology

Mobile phones Smart phones Desktop computers Computer monitor Laptop Hard disks Home entertainment devices

DVDs Televisions Video game systems Fax machines Copiers Printers Electronic Services

Massage chairs Heating pads Remote control TV remotes Electric cable Lamps Smart lights Night lights treadmill Smart watches Heart monitors Diabetes testing equipment Office and medical equipment

Copiers/printers IT servers wire and cable Wi-Fi dongles Dialysis machines Imaging equipment Telephone and central systems Audio and video equipment Network hardware (eg server, switch, hub, etc.) Power strips and power supplies Uninterruptible power supply (UPS systems) power distribution systems (PDU) autoclave shock device Today, with the advancement of technology and the high speed of introducing new products to the market, even healthy devices become obsolete after a while.

Think of the many VCR players that were replaced by the introduction of DVD players, and now DVD players with

Blu-ray players have been replaced.

The Environmental Protection Agency estimates that up to 60 million tons of e-waste end up in landfills in the United States alone each year. E-waste is the starting point for the introduction of toxic substances into the environment

While the use of electronic devices is not dangerous for us in terms of having toxic substances, most electronic devices contain some types of toxic substances, including beryllium, cadmium, mercury and lead, which pose serious environmental risks to soil, water, air and life. They bring our beast with them.

The more e-waste and metals in landfills, the more these toxic substances seep into the groundwater.

The entry of these substances into the water severely harms the wildlife and causes the wildlife to become ill due to lead, arsenic, cadmium and other metals poisoning due to the high concentration of these minerals. What to do with electronic waste? Fortunately, there is a proven solution! E-waste recycling is very useful and efficient.

Recovering parts inside devices that are still valuable and providing recycled metals to manufacturers who can make new products

Using these parts is a very good solution.

It can be said that almost all electronic waste contains some kind of recyclable material, this includes materials such as plastic, glass and metals, etc., and for this reason, it may seem a little funny that these devices are called "garbage" in While it can be very useful.

Technological innovators continue to create electronic devices that make our lives easier and more convenient in every way imaginable, so in response to the question, "What is e-waste?" A good answer would be: "It depends."

Electronic waste generation worldwide stood at roughly 62 million metric tons in 2022. Several factors, such as increased spending power, and the availability of electronics, has fueled e-waste generation in recent decades, making it the fastest growing waste stream worldwide. This trend is expected to continue, with annual e-waste generation forecast at 82 million metric tons in 2030.

How much e-waste do people produce?

Globally, e-waste generation per capita averaged 7.8 kilograms in 2022. However, this differs greatly depending on the region. While Asia produces the most e-waste worldwide in volume, Europe and Oceania were the regions with the highest e-waste generation per capita, at 17.6 and 16.1 kilograms respectively.

E-waste disposal

In 2022, the share of e-waste formally collected and recycled worldwide stood at 22.3 percent. Meanwhile, around 48 million metric tons are estimated to have been collected informally, with 29 percent of this value being disposed as residual waste, most likely ending up in landfills. Due to the hazardous materials that are often used in electronics, improper e-waste disposal is a growing environmental concern worldwide.

The dataset contains year- and state-wise total quantity of electronic waste (E-waste) which is collected and processed.

Note:

The blank cells in the dataset represent no data being reported by the respective states

Sustainable management of electronic waste is critical to achieving a circular-economy and minimizing environment and public health risks. The objective of this study was to investigate the use of pyrolysis as a possible technique to recover valuable materials and energy from different components of e-waste as an alternative approach for limiting their disposal to landfills. The study includes investigating the potential impact of thermal processing of e-waste.Thermogravimetric (TG) analysis and differential thermogravimetric analysis (DTG) of e-waste components were used to better understand the mass loss characteristics of the pyrolysis process up to 700 oC. The changes in e-waste chemical components during pyrolysis were considered using Fourier-transform infrared (FTIR) spectrometry and X-ray fluorescence (XRF) techniques. The energy recovery from pyrolysis was made in a horizontal tube furnace under anoxic and isothermal condition of selected temperatures of 300, 400 and 500 oC. Critical and valuable metals were recovered from electronic components. Pyrolysis produced liquid and gas mixtures organic compounds that can be used as fuels, but the process also emitted particulate matter and semi-volatile organic products, and the remaining ash contained leachable pollutants. Furthermore, toxicity leaching characteristic profile of e-waste and partly oxidized products were conducted to measure the levels of pollutants leached before and after pyrolysis at selected temperatures. The results of this study contribute to the development of alternative approaches to practical recycling that could especially help reduce plastic pollution and recover materials of value from e-waste. Additionally, this information may be used to assess the risk of exposure of workers to emissions semi-formal recycling centers.

This dataset is associated with the following publication: Sahle-Demessie, E., B. Mezgebe, J. Dietrich, Y. Shan, S. Harmon, and C.C. Lee. Material recovery from electronic waste using pyrolysis: Emissions measurements and risk assessment. Journal of Environmental Chemical Engineering. Elsevier B.V., Amsterdam, NETHERLANDS, 9(1): 104943, (2021).

This dataset provides detailed information on the quantity of various precious metals extracted from e-waste of different electronic devices. The metals analyzed include Gold, Aluminum, Silver, Carbon, Platinum, Rhodium, Nickel, Tin, and Lithium. Each entry in the dataset corresponds to a specific electronic device, such as smartphones, gaming consoles, and laptops, with their respective metal contents measured in grams.

Columns:

Item Name:

Type: String Description: Name of the electronic item. This helps in identifying the specific device being referred to. Category:

Type: Categorical (e.g., Cat1, Cat2, Cat3, Cat4) Description: Category of the electronic device. It classifies the item into a broader group, which can impact recovery rates. Brand Name:

Type: Categorical (e.g., Panasonic, Sony, Lenovo, etc.) Description: Brand of the device. Different brands might use different materials and have varying recovery rates. Device Age:

Type: Integer Description: Age of the device in years. This can influence the amount of recoverable materials as devices may degrade over time. Device Condition:

Type: Categorical (e.g., Broken, Average, Good) Description: Condition of the device at the time of recycling. Affects the amount and quality of recoverable materials. Device Type:

Type: Categorical (e.g., Consumer Electronics, Appliance, IT Equipment) Description: Type of electronic device. Different types of devices have different material compositions and recovery rates. Year of Manufacture:

Type: Integer Description: Year the device was manufactured. Older devices may contain different materials compared to newer ones. Gold (g):

Type: Float Description: Amount of gold (in grams) present in the device. Aluminum (g):

Type: Float Description: Amount of aluminum (in grams) present in the device. Silver (g):

Type: Float Description: Amount of silver (in grams) present in the device. Carbon (g):

Type: Float Description: Amount of carbon (in grams) present in the device. Platinum (g):

Type: Float Description: Amount of platinum (in grams) present in the device. Rhodium (g):

Type: Float Description: Amount of rhodium (in grams) present in the device. Nickel (g):

Type: Float Description: Amount of nickel (in grams) present in the device. Tin (g):

Type: Float Description: Amount of tin (in grams) present in the device. Lithium (g):

Type: Float Description: Amount of lithium (in grams) present in the device. Material Recovery Rate:

Type: Float Description: The percentage of material recovered from the device. This is the target variable for predictive modeling.

Overview

The goal of this project was to create a structured dataset which can be used to train computer vision models to detect electronic waste devices, i.e., e-waste or Waste Electrical and Electronic Equipment (WEEE). Due to the often-subjective differences between e-waste and functioning electronic devices, a model trained on this dataset could also be used to detect electronic devices in general. However, it must be noted that for the purposes of e-waste recognition, this dataset does not differentiate between different brands or models of the same type of electronic devices, e.g. smartphones, and it also includes images of damaged equipment.

The structure of this dataset is based on the UNU-KEYS classification Wang et al., 2012, Forti et al., 2018. Each class in this dataset has a tag containing its corresponding UNU-KEY. This dataset structure has the following benefits: 1. It allows the user to easily classify e-waste devices regardless of which e-waste definition their country or organization uses, thanks to the correlation between the UNU-KEYS and other classifications such as the HS-codes or the EU-6 categories, defined in the WEEE directive; 2. It helps dataset contributors focus on adding e-waste devices with higher priority compared to arbitrarily chosen devices. This is because electronic devices in the same UNU-KEY category have similar function, average weight and life-time distribution as well as comparable material composition, both in terms of hazardous substances and valuable materials, and related end-of-life attributes Forti et al., 2018. 3. It gives dataset contributors a clear goal of which electronic devices still need to be added and a clear understanding of their progress in the seemingly endless task of creating an e-waste dataset.

This dataset contains annotated images of e-waste from every UNU-KEY category. According to Forti et al., 2018, there are a total of 54 UNU-KEY e-waste categories.

Description of Classes

At the time of writing, 22. Apr. 2024, the dataset has 19613 annotated images and 77 classes. The dataset has mixed bounding-box and polygon annotations. Each class of the dataset represents one type of electronic device. Different models of the same type of device belong to the same class. For example, different brands of smartphones are labelled as "Smartphone", regardless of their make or model. Many classes can belong to the same UNU-KEY category and therefore have the same tag. For example, the classes "Smartphone" and "Bar-Phone" both belong to the UNU-KEY category "0306 - Mobile Phones". The images in the dataset are anonymized, meaning that no people were annotated and images containing visible faces were removed.

The dataset was almost entirely built by cloning annotated images from the following open-source Roboflow datasets: [1]-[91]. Some of the images in the dataset were acquired from the Wikimedia Commons website. Those images were chosen to have an unrestrictive license, i.e., they belong to the public domain. They were manually annotated and added to the dataset.

Cite This Project

This work was done as part of the PhD of Dimitar Iliev, student at the Faculty of German Engineering and Industrial Management at the Technical University of Sofia, Bulgaria and in collaboration with the Faculty of Computer Science at Otto-von-Guericke-University Magdeburg, Germany.

If you use this dataset in a research paper, please cite it using the following BibTeX: @article{iliev2024EwasteDataset, author = "Iliev, Dimitar and Marinov, Marin and Ortmeier, Frank", title = "A proposal for a new e-waste image dataset based on the unu-keys classification", journal = "XXIII-rd International Symposium on Electrical Apparatus and Technologies SIELA 2024", year = 2024, volume = "23", number = "to appear", pages = {to appear} note = {under submission} }

Contribution Guidelines

Image Collection

1. Choose a specific electronic device type to add to the dataset and find its corresponding UNU-KEY. * The chosen type of device should have a characteristic design which an object detection model can learn. For example, CRT monitors look distinctly different than flat panel monitors and should therefore belong to a different class, regardless that they are both monitors. In contrast, LED monitors and LCD monitors look very similar and are therefore both labelled as Flat-Panel-Monitor in this dataset.
2. Collect images of this type of device. * Take note of the license of those images and their author/s to avoid copyright infringement. * Do not collect images with visible faces to protect personal data and comply with GDPR regulations. * Images can

E-waste generation worldwide has nearly doubled since 2010, from **** million metric tons to roughly ** million tons in 2022. Electronic waste is one of the fastest growing waste streams, with global e-waste generation projected to reach ** million metric tons by 2030. What makes up electronic waste? In 2022, small equipment, such as vacuum cleaners, microwaves, toasters, and electric kettles made up the largest share of global electronic waste generation, at more than **** million metric tons. Another ** million metric tons of large equipment waste was also generated that year. Although still accounting for less than one percent of e-waste generated worldwide, the growth in solar PV capacity worldwide has seen photovoltaic panels as a growing waste stream. Where is electronic waste generated? China is by far the largest e-waste generating country worldwide, with more than ** million metric tons generated in 2022. In fact, Asia accounted for nearly half of all e-waste generated that year. Nevertheless, when it comes to e-waste generation per capita, four of the top five countries were located in Europe, with Norway leading the ranking at **** kilograms per inhabitant.

E-Waste Dataset

Overview

The E-Waste Dataset is a collection of images representing electronic waste items categorized into distinct classes. The dataset is designed for tasks such as image classification, object detection, and other computer vision applications. Electronic waste, or e-waste, is a growing concern globally, and this dataset aims to contribute to the development of technology-driven solutions for its management and recycling.

Content

The dataset is organized into three main folders:

• Train: Contains images for training the machine learning models.
• Test: Contains images for evaluating the performance of the models.
• Validation: Contains images for fine-tuning and validating the models.

Each of these folders is further divided into classes, with each class representing a specific type of electronic waste item.

Classes

The dataset comprises various classes of electronic waste, including but not limited to:

1. PCB (Printed Circuit Board)
2. Player
3. Battery
4. Microwave
5. Mobile
6. Mouse
7. Printer
8. Television
9. Washing Machine
10. Keyboard

Data Sources

The images in this dataset were collected from diverse sources, including open datasets, image repositories, and proprietary sources. Efforts were made to ensure a representative and diverse collection of electronic waste items.

Inspiration

The inspiration behind creating this dataset is to foster research and innovation in the field of computer vision and machine learning, specifically addressing challenges related to the identification and recycling of electronic waste. By providing a standardized dataset, we aim to encourage collaboration among researchers and developers working on solutions to mitigate the environmental impact of e-waste.

License

The E-Waste Dataset is released under [Apache 2.0]

This dataset is designed to support AI-driven e-waste tracking, smart disposal, and recycling solutions. It contains essential details on electronic products, including expiry dates, pricing trends, and other relevant attributes. The data can be used for predictive analytics, market trend analysis, and sustainability-focused machine learning models.

A current listing of NYS Registered Electronic Waste Recycling Facilities. Electronic waste types accepted vary from facility to facility.

E waste detection Dataset for detecting e waste components

Just **** percent of electronic waste generated worldwide was documented as formally collected and recycled in 2022. Europe had the highest collection and recycling rate, at **** percent. Many European countries export e-waste - often illegally - to developing countries. Several of these countries are in Africa, where the collection and recycling rate of e-waste was just *** percent. In 2022, around ** million metric tons of e-waste were estimated to be disposed as residual waste, with another ** million metric being handled in low income countries with insufficient management infrastructure.

Overview

The goal of this project was to create a structured dataset which can be used to train computer vision models to detect electronic waste devices, i.e., e-waste or Waste Electrical and Electronic Equipment (WEEE). Due to the often-subjective differences between e-waste and functioning electronic devices, a model trained on this dataset could also be used to detect electronic devices in general. However, it must be noted that for the purposes of e-waste recognition, this dataset does not differentiate between different brands or models of the same type of electronic devices, e.g. smartphones, and it also includes images of damaged equipment.

The structure of this dataset is based on the UNU-KEYS classification Wang et al., 2012, Forti et al., 2018. Each class in this dataset has a tag containing its corresponding UNU-KEY. This dataset structure has the following benefits: 1. It allows the user to easily classify e-waste devices regardless of which e-waste definition their country or organization uses, thanks to the correlation between the UNU-KEYS and other classifications such as the HS-codes or the EU-6 categories, defined in the WEEE directive; 2. It helps dataset contributors focus on adding e-waste devices with higher priority compared to arbitrarily chosen devices. This is because electronic devices in the same UNU-KEY category have similar function, average weight and life-time distribution as well as comparable material composition, both in terms of hazardous substances and valuable materials, and related end-of-life attributes Forti et al., 2018. 3. It gives dataset contributors a clear goal of which electronic devices still need to be added and a clear understanding of their progress in the seemingly endless task of creating an e-waste dataset.

This dataset contains annotated images of e-waste from every UNU-KEY category. According to Forti et al., 2018, there are a total of 54 UNU-KEY e-waste categories.

Description of Classes

At the time of writing, 22. Apr. 2024, the dataset has 19613 annotated images and 77 classes. The dataset has mixed bounding-box and polygon annotations. Each class of the dataset represents one type of electronic device. Different models of the same type of device belong to the same class. For example, different brands of smartphones are labelled as "Smartphone", regardless of their make or model. Many classes can belong to the same UNU-KEY category and therefore have the same tag. For example, the classes "Smartphone" and "Bar-Phone" both belong to the UNU-KEY category "0306 - Mobile Phones". The images in the dataset are anonymized, meaning that no people were annotated and images containing visible faces were removed.

The dataset was almost entirely built by cloning annotated images from the following open-source Roboflow datasets: [1]-[91]. Some of the images in the dataset were acquired from the Wikimedia Commons website. Those images were chosen to have an unrestrictive license, i.e., they belong to the public domain. They were manually annotated and added to the dataset.

Cite This Project

This work was done as part of the PhD of Dimitar Iliev, student at the Faculty of German Engineering and Industrial Management at the Technical University of Sofia, Bulgaria and in collaboration with the Faculty of Computer Science at Otto-von-Guericke-University Magdeburg, Germany.

If you use this dataset in a research paper, please cite it using the following BibTeX: @article{iliev2024EwasteDataset, author = "Iliev, Dimitar and Marinov, Marin and Ortmeier, Frank", title = "A proposal for a new e-waste image dataset based on the unu-keys classification", journal = "XXIII-rd International Symposium on Electrical Apparatus and Technologies SIELA 2024", year = 2024, volume = "23", number = "to appear", pages = {to appear} note = {under submission} }

Contribution Guidelines

Image Collection

1. Choose a specific electronic device type to add to the dataset and find its corresponding UNU-KEY. * The chosen type of device should have a characteristic design which an object detection model can learn. For example, CRT monitors look distinctly different than flat panel monitors and should therefore belong to a different class, regardless that they are both monitors. In contrast, LED monitors and LCD monitors look very similar and are therefore both labelled as Flat-Panel-Monitor in this dataset.
2. Collect images of this type of device. * Take note of the license of those images and their author/s to avoid copyright infringement. * Do not collect images with visible faces to protect personal data and comply with GDPR regulations. * Images can

♻️ E-Waste Classification Dataset (37 Classes) - Public Subset

📌 Context

This dataset was developed as part of the research paper "Hybrid Framework for E-Waste Classification using YOLO11 and Large Language Model Validation," presented at the 2025 International Conference on Advanced Technologies in Energy and Informatic (ICATEI).

📦 Dataset Versioning (v1.0)

Please Note: This repository contains a curated public subset of the complete dataset used in our original research. We have released this specific subset (Version 1.0) to provide the academic community with a high-quality, balanced foundation for reproducing our core methodology and testing baseline YOLO architectures. Future updates may expand the dataset.

📊 Dataset Details

The dataset is structured in standard YOLO annotation format (.txt bounding boxes) and is ready for immediate training. * Number of Classes: 37 distinct e-waste categories * Dataset Split: Pre-split into 60-20-20, 70-20-10, and 80-10-10 for immediate training.

📸 Data Collection Methodology

This is a hybrid dataset: * Manual Photography: Real-world photos captured by our research team across various lighting conditions, backgrounds, and angles to ensure robustness. * Publicly Sourced Images: Curated from public sources and search engines (under Fair Use for academic research) to expand underrepresented classes and ensure comprehensive 37-class coverage.

⚖️ Dataset Balancing Strategy

Because real-world data contains a natural class imbalance (e.g., highly common items like Cables vs rare items like Projectors), we implemented an oversampling algorithm during preprocessing to protect model training from minority class neglect. The training set was balanced using a cyclic duplication pipeline targeting the median class instance count.

📄 License and Citation

License: CC BY-NC 4.0 (Attribution-NonCommercial 4.0 International). This dataset is strictly restricted to academic and non-commercial research.

Citation: If you use this dataset in your research, please cite our conference paper:

David, A., Lase, A. N., Fauzi, F. R., Kallista, M., & Fikri, R. M. (2025). Hybrid Framework for E-Waste Classification using YOLO11 and Large Language Model Validation. 2025 International Conference on Advanced Technologies in Energy and Informatic (ICATEI). Telkom University, Indonesia.

Hardware-ID

A Curated High-Resolution Image Dataset for Electronic Waste Identification & Classification

Overview

Hardware-ID is a curated, high-resolution image dataset focused on the identification and classification of Electronic Waste (E-Waste). As global electronic waste production continues to accelerate, the development of automated recycling and sorting systems is critical for a sustainable circular economy.

This dataset bridges a significant gap in existing computer vision resources by providing granular, annotated data for internal laptop components — parts that are traditionally difficult for models to distinguish.

Context

While many hardware datasets focus on exterior device identification, Hardware-ID goes deeper — literally. By documenting the internal modular components of laptops, this dataset facilitates the training of models designed for:

• Automated Dismantling — Assisting robotic systems in identifying part boundaries.
• Sustainable Waste Management — Improving the purity of sorted materials for recycling.
• Inventory Cataloging — Identifying salvaged parts for the secondary market.

Dataset Composition

Target Classes

The dataset includes diverse components from various laptop generations and manufacturers, with a specific focus on:

Optical Drives

DVD-ROMs, Blu-ray drives, and CD-writers.

Input Interfaces

Touchpads, trackpads, and internal ribbon connectors.

Acoustics

Internal speakers, acoustic chambers, and sub-woofers.

Cooling Systems (Optional)

Heat sinks and centrifugal fans.

Technical Methodology

Each image has been manually audited to ensure high data quality. The labeling strategy employed via Label Studio ensures that even overlapping or partially obscured components are annotated with high precision — making the dataset robust for real-world "messy" environments like recycling centers.

Acknowledgments

This dataset was developed as part of a personal initiative to leverage computer vision for environmental sustainability and technical hardware documentation.

(i) Waste electrical and electronic equipment (WEEE), also known as e-waste, such as computers, televisions, fridges and mobile phones, is one the fastest growing waste streams in the EU. WEEE include precious materials the recycling of which should be enhanced. (ii) The indicator is calculated by multiplying the 'collection rate' as set out in the WEEE Directive with the 'reuse and recycling rate' set out in the WEEE Directive; where: o The 'collection rate' equals the volumes collected of WEEE in the reference year divided by the average quantity of electrical and electronic equipment (EEE) put on the market in the previous three years (both expressed in mass unit). o The 'reuse and recycling rate' is calculated by dividing the weight of WEEE that enters the recycling/preparing for re-use facility by the weight of all separately collected WEEE (both in mass unit) in accordance with Article 11(2) of the WEEE Directive 2012/19/EU, considering that the total amount of collected WEEE is sent to treatment/recycling facilities. The indicator is expressed in percent (%) as both terms are measured in the same unit. (iii) EU Member States plus the United Kingdom, Iceland, Liechtenstein and Norway (iv)

Explore the booming E Waste Recycling And Reuse Services Market with a USD 2146.71 Million valuation and 7.3% CAGR through 2034. Discover drivers, trends, and key players in this essential sector. Key drivers for this market are: Limited shelf life of electronic products, High rate of obsolescence. Potential restraints include: High cost of e-waste management, Fewer e-waste collection zones.

The booming e-waste recycling market is projected to reach $50 billion by 2025, growing at an 8% CAGR through 2033. Discover key market drivers, trends, and challenges influencing this rapidly expanding sector, including regional breakdowns and leading companies.

Market size, estimate, forecast and CAGR for the E-waste Management Market Size & Share Report, 2025-2030.

E-Waste Vision 18: Compiled Electronic Waste Image Dataset

Overview This dataset contains electronic-waste images organized into 18 categories for multi-class image classification and benchmarking tasks.

The dataset is intended for: - Computer vision research - Deep learning and transfer learning experiments - Model comparison on e-waste object categories - Educational use in sustainability-focused AI projects

Important Rights and Licensing Notice This dataset is a compiled collection of images gathered from multiple third-party sources.

The uploader does not claim ownership of the original images. Copyright and licensing rights remain with the respective original rights holders.

Because complete source-level license records are not currently available for all files, no new blanket license is granted for the underlying images through this upload.

Use of this compiled dataset is provided for educational and research purposes, at your own risk. Before redistribution, publication, or any commercial use, users must independently verify rights and licensing for each image from its original source.

This dataset is provided as-is, without warranty of any kind.

Class Labels (18) 1. Air-Conditioner
2. Battery
3. heat-sink
4. Keyboard
5. Laptop
6. light bulbs
7. Microchip-IC
8. Microwave
9. Mobile
10. Mouse
11. Passive-Component
12. PCB
13. Printer
14. Refrigerator
15. Resistor
16. Television
17. transistor
18. Washing Machine

Suggested Use Cases - Baseline CNN and transformer-based image classification - Transfer learning on environmental and recycling datasets - Class-wise error analysis with confusion matrices - Educational projects on e-waste identification systems

Limitations - Images are collected from mixed third-party sources - Source quality, resolution, and capture conditions may vary - Potential class imbalance and domain bias may exist - Full per-image source attribution is not currently available