DinoDental: Benchmarking DINOv3 as a Unified Vision Encoder for Dental Image Analysis
arXiv:2603.28297v1 Announce Type: new Abstract: The scarcity and high cost of expert annotations in dental imaging present a significant challenge for the development of AI in dentistry. DINOv3, a state-of-the-art, self-supervised vision foundation model pre-trained on 1.7 billion images, offers a promising pathway to mitigate this issue. However, its reliability when transferred to the dental domain, with its unique imaging characteristics and clinical subtleties, remains unclear. To address this, we introduce DinoDental, a unified benchmark designed to systematically evaluate whether DINOv3 — Kun Tang, Xinquan Yang, Mianjie Zheng, Xuefen Liu, Xuguang Li, Xiaoqi Guo, Ruihan Chen, Linlin Shen, He Meng
View PDF HTML (experimental)
Abstract:The scarcity and high cost of expert annotations in dental imaging present a significant challenge for the development of AI in dentistry. DINOv3, a state-of-the-art, self-supervised vision foundation model pre-trained on 1.7 billion images, offers a promising pathway to mitigate this issue. However, its reliability when transferred to the dental domain, with its unique imaging characteristics and clinical subtleties, remains unclear. To address this, we introduce DinoDental, a unified benchmark designed to systematically evaluate whether DINOv3 can serve as a reliable, off-the-shelf encoder for comprehensive dental image analysis without requiring domain-specific pre-training. Constructed from multiple public datasets, DinoDental covers a wide range of tasks, including classification, detection, and instance segmentation on both panoramic radiographs and intraoral photographs. We further analyze the model's transfer performance by scaling its size and input resolution, and by comparing different adaptation strategies, including frozen features, full fine-tuning, and the parameter-efficient Low-Rank Adaptation (LoRA) method. Our experiments show that DINOv3 can serve as a strong unified encoder for dental image analysis across both panoramic radiographs and intraoral photographs, remaining competitive across tasks while showing particularly clear advantages for intraoral image understanding and boundary-sensitive dense prediction. Collectively, DinoDental provides a systematic framework for comprehensively evaluating DINOv3 in dental analysis, establishing a foundational benchmark to guide efficient and effective model selection and adaptation for the dental AI community.
Subjects:
Computer Vision and Pattern Recognition (cs.CV)
Cite as: arXiv:2603.28297 [cs.CV]
(or arXiv:2603.28297v1 [cs.CV] for this version)
https://doi.org/10.48550/arXiv.2603.28297
arXiv-issued DOI via DataCite (pending registration)
Submission history
From: Xinquan Yang [view email] [v1] Mon, 30 Mar 2026 11:23:57 UTC (8,618 KB)
Sign in to highlight and annotate this article
Conversation starters
Daily AI Digest
Get the top 5 AI stories delivered to your inbox every morning.
More about
researchpaperarxiv
Personalised care, youth mental health, and digital technology: A value sensitive design perspective and framework
Digital health is typically driven, in part, by the principle of personalised care. However, the underlying values and associated ethical design considerations at the intersection of personalised care, youth mental health, and digital technology are underexplored. Through a value sensitive design lens, this work aims to contribute a prototype conceptual framework for the ethical design and evaluation of personalised youth digital mental health technology, which comprises three values–personalisation, empowerment, and autonomy–and 15 design norms as fundamental yet non-exhaustive ethical criteria. Furthermore, it provides illustrative applications of the framework by applying it to (1) the proactive design of two exemplary digital mental health technologies to draw out emerging ethical cons
Decoding AI ethics: What do ethical tools tell us about ethics?
This study performs a systematic content analysis to map out the landscape of eighty-seven ethical tools concerning the type of ethical principles addressed, the approach promoted (technological vs. socio-technical vs. societal), and the kind of assessment endorsed (self-assessment versus assessment by others). The findings reveal a skewed representation and emphasis of specific ethical principles across these tools. Furthermore, ethical tools primarily advocate for a technological or socio-technical approach. In contrast, only a few ethical tools emphasize the importance of a societal approach. Additionally, the research highlights a reliance on self-assessments within these tools, raising concerns about the objectivity and comprehensiveness of ethical evaluations in AI development. Our f
Disruptive technologies, engineered concepts, and normative guidance
Socially disruptive technologies can induce normative disorientation. This occurs as they disrupt established concepts that have traditionally provided normative guidance. A notable example of such technology-induced conceptual disruption is the advent of ventilator technology. Patients who lost brain stem activity and autonomous ventilation, yet remained alive through ventilator support, created a state of uncertainty: they were considered “dead” in terms of (autonomous) ventilation and brain activity, but “alive” in terms of cardiac function. This descriptive ambiguity led to normative disorientation, particularly among clinicians and patients’ relatives. In response, conceptual engineering and the introduction of the new concept of “brain death” have been identified as critical steps to
Knowledge Map
Connected Articles — Knowledge Graph
This article is connected to other articles through shared AI topics and tags.
More in Research Papers
Disruptive technologies, engineered concepts, and normative guidance
Socially disruptive technologies can induce normative disorientation. This occurs as they disrupt established concepts that have traditionally provided normative guidance. A notable example of such technology-induced conceptual disruption is the advent of ventilator technology. Patients who lost brain stem activity and autonomous ventilation, yet remained alive through ventilator support, created a state of uncertainty: they were considered “dead” in terms of (autonomous) ventilation and brain activity, but “alive” in terms of cardiac function. This descriptive ambiguity led to normative disorientation, particularly among clinicians and patients’ relatives. In response, conceptual engineering and the introduction of the new concept of “brain death” have been identified as critical steps to
Blocking the Internet Archive Won’t Stop AI, But It Will Erase the Web’s Historical Record
<div class="field field--name-body field--type-text-with-summary field--label-hidden"><div class="field__items"><div class="field__item even"><p>Imagine a newspaper publisher announcing it will no longer allow libraries to keep copies of its paper. </p> <p><span>That’s effectively what’s begun happening online in the last few months. The Internet Archive—the world’s largest digital library—has preserved newspapers since it went online </span><a href="https://en.wikipedia.org/wiki/Wayback_Machine"><span>in the mid-1990s</span></a><span>. The Archive’s mission is to preserve the web and make it accessible to the public. To that end, the organization operates the Wayback Machine, which now contains </span><a href="https://blog.archive.org/trillion/"><span>more than one trillion archived web p
Report: 95% of Organisations do not Have Full Trust in Cybersecurity Vendors
The report examines one of cybersecurity’s most urgent and overlooked necessities: trust. Sophos, a global security solutions firm, have published their findings from a vendor-agnostic study, based on responses from 5,000 organisations across 17 countries. The Cybersecurity Trust Reality 2026 report reveals a critical challenge facing CISOs: Trust in cybersecurity vendors is fragile, difficult to […] The post Report: 95% of Organisations do not Have Full Trust in Cybersecurity Vendors appeared first on DIGIT .
Discussion
Sign in to join the discussion
No comments yet — be the first to share your thoughts!