Dataset Card for AIDOVECL

We introduce an annotated AI-generated dataset of eye-level vehicle images using outpainting, offering versatile generation of diverse vehicle classes in varied contexts with pretrained models.

Dataset Details

Dataset Description

Our dataset consists of AI-generated, eye-level vehicle images created by detecting and cropping vehicles from manually chosen seed images. These vehicles are then outpainted onto larger canvases to replicate diverse real-world scenarios. The resulting images are meticulously annotated, offering high-quality ground truth data. Through the use of sophisticated outpainting techniques and stringent image quality assessments, we ensure exceptional visual fidelity and contextual accuracy.

• Curated by: Amir Kazemi, Qurat ul ain Fatima, Volodymyr Kindratenko, Christopher W Tessum.
• Funded by [optional]: The authors acknowledge support from the University of Illinois National Center for Supercomputing Applications (NCSA) Faculty Fellowship Program, the Zhejiang University and University of Illinois Dynamic Research Enterprise for Multidisciplinary Engineering Sciences (DREMES), and the University of Illinois Discovery Partners Institute (DPI). This work utilizes resources supported by the National Science Foundation’s (NSF) Major Research Instrumentation program, grant #1725729, as well as the University of Illinois at Urbana-Champaign.
• Shared by [optional]: NA
• Language(s) (NLP): NA
• License: Creative Commons Attribution 4.0 International

Dataset Sources [optional]

• Repository: https://huggingface.co/datasets/amir-kazemi/aidovecl/tree/main
• Paper [optional]: [Under Review]
• Demo [optional]: https://github.com/amir-kazemi/aidovecl

Uses

Direct Use

This dataset is intended for several key applications: Autonomous Driving: Enhancing the training of machine learning models for better vehicle detection and classification in diverse and complex urban environments. Urban Planning: Assisting in the analysis and design of urban infrastructure by providing detailed and annotated eye-level vehicle images. Environmental Monitoring: Enabling better monitoring and management of urban traffic and its environmental impact through comprehensive visual data. By addressing the scarcity of annotated data with high-quality, AI-generated images, this dataset aims to improve the performance and generalization of models in these domains.

Out-of-Scope Use

While the dataset offers substantial improvements in data availability and quality, it is not intended for certain uses: Malicious Activities: Any use of this dataset for unlawful activities, including surveillance without consent or privacy infringement, is strictly prohibited. Unrelated Fields: The dataset is tailored for applications in autonomous driving, urban planning, and environmental monitoring. Its use in unrelated fields where the context and annotations do not align with the dataset’s design may yield poor performance and unreliable results.

Dataset Structure

Refer to the readme.md at https://github.com/amir-kazemi/aidovecl.

Dataset Creation

Curation Rationale

Publicly available vehicle recognition datasets have driven progress in the field but often fall short in key areas, including lacking contextual richness, offering limited vehicle classifications, and presenting suboptimal viewing angles. These shortcomings underscore the necessity for customizable vehicle datasets tailored to meet the specific requirements of contemporary vehicle recognition applications.

Source Data

Data Collection and Processing

Refer to the manuscript currently under review.

Who are the source data producers?

Refer to the manuscript currently under review.

Annotations [optional]

Annotation process

Self-annotated using a method currently under review.

Who are the annotators?

Self-annotated using a method currently under review.

Personal and Sensitive Information

NA

Bias, Risks, and Limitations

Our dataset currently cannot outpaint multiple vehicles in a single image due to the limitations of the inpainting model, which struggles with generating coherent scenes with multiple vehicles. This also stems from the challenge of selecting seed images that naturally correlate. Consequently, our dataset lacks rich occlusion scenarios crucial for training robust detection systems. Additionally, our reliance on pretrained detection and inpainting models means that objects not included in the original model training may be missed or appear unrealistic. We also face a shortage of seed images, especially for certain vehicle classes like vans, leading to class imbalances. Addressing these issues is crucial for improving our dataset's effectiveness and the accuracy of detection models.

Recommendations

Users should be made aware of the risks, biases and limitations of the dataset. More information needed for further recommendations.

Citation [optional]

BibTeX:

[Under Review]

APA:

[Under Review]

Glossary [optional]

NA

More Information [optional]

NA

Dataset Card Authors [optional]

Amir Kazemi

Dataset Card Contact

kazemi2@illinois.edu