Hibou: A Family of Foundational Vision Transformers for Pathology

Pathology, the microscopic examination of diseased tissue, is critical for diagnosing various medical conditions, particularly cancers. Traditional methods are labor-intensive and prone to human error. Digital pathology, which converts glass slides into high-resolution digital images for analysis by computer algorithms, revolutionizes the field by enhancing diagnostic accuracy, consistency, and efficiency through automated image analysis and large-scale data processing. Foundational transformer pretraining is crucial for developing robust, generalizable models as it enables learning from vast amounts of unannotated data. This paper introduces the Hibou family of foundational vision transformers for pathology, leveraging the DINOv2 framework to pretrain two model variants, Hibou-B and Hibou-L, on a proprietary dataset of over 1 million whole slide images (WSIs) representing diverse tissue types and staining techniques. Our pretrained models demonstrate superior performance on both patch-level and slide-level benchmarks, surpassing existing state-of-the-art methods. Notably, Hibou-L achieves the highest average accuracy across multiple benchmark datasets. To support further research and application in the field, we have open-sourced the Hibou-B model, which can be accessed at https://github.com/HistAI/hibou

Trimanipulation: Evaluation of human performance in a 3-handed coordination task

Many teleoperation tasks require three or more tools working together, which need the cooperation of multiple operators. The effectiveness of such schemes may be limited by communication. Trimanipulation by a single operator using an artificial third arm controlled together with their natural arms is a promising solution to this issue. Foot-controlled interfaces have previously shown the capability to be used for the continuous control of robot arms. However, the use of such interfaces for controlling a supernumerary robotic limb (SRLs) in coordination with the natural limbs, is not well understood. In this paper, a teleoperation task imitating physically coupled hands in a virtual reality scene was conducted with 14 subjects to evaluate human performance during tri-manipulation. The participants were required to move three limbs together in a coordinated way mimicking three arms holding a shared physical object. It was found that after a short practice session, the three-hand tri-manipulation using a single subject's hands and foot was still slower than dyad operation, however, they displayed similar performance in success rate and higher motion efficiency than two person's cooperation.