Single instrument frames subset from PhaKIR

To access the dataset, the following steps have to be performed: Register on the Zenodo platform and login: Only then the Access Request Form below will become visible. Fill out and submit the Request Access Form appearing below. The access request will be verified and you will be notified by email as soon as access is granted. This dataset is a subset of the original PhaKIR dataset and is used by the OpenMIBOOD framework as In-Distribution dataset.When using this dataset, it is mandatory to cite the corresponding publication (OpenMIBOOD) and to follow the acknowledgement and citation requirements of the original dataset The original PhaKIR dataset (Challenge Paper, Dataset Paper) consists of 8 cholecystectomy surgery videos with 485,875 frames recorded at 25 fps, stemming from 3 hospitals. It includes phase annotations for every frame and instance segementation annotations for 19 instrument classes for every 25th frame. Additionally, keypoint annotations are available for instrument tracking. Multiple instruments may be present in each frame. For this dataset subset, frames were extracted based on instance segmentation annotations, selecting only those containing Clip-Applicator, Grasper, PE-Forceps, Needle-Probe, Palpation-Probe, Suction-Rod, or No-Instrument, while ensuring that at most a single instrument appears per frame. Frames were exluded if the instrument only covered a very small percentage (0.5%) of the image area. Using smoke annotations from a previous publication of the PhaKIR challenge organizers (Smoke Annotations), frames were categorized into No-Smoke, Medium-Smoke, and Heavy-Smoke. This resulting subset has 2769 frames. All resulting works using this dataset in part or in whole must cite the following publications: @article{rueckert2025comparative, author = {Tobias Rueckert and David Rauber and Raphaela Maerkl and Leonard Klausmann and Suemeyye R. Yildiran and Max Gutbrod and Danilo Weber Nunes and Alvaro Fernandez Moreno and Imanol Luengo and Danail Stoyanov and Nicolas Toussaint and Enki Cho and Hyeon Bae Kim and Oh Sung Choo and Ka Young Kim and Seong Tae Kim and Gon{\c{c}}alo Arantes and Kehan Song and Jianjun Zhu and Junchen Xiong and Tingyi Lin and Shunsuke Kikuchi and Hiroki Matsuzaki and Atsushi Kouno and Jo{\~{a}}o Renato Ribeiro Manesco and Jo{\~{a}}o Paulo Papa and Tae{-}Min Choi and Tae Kyeong Jeong and Juyoun Park and Oluwatosin Alabi and Meng Wei and Tom Vercauteren and Runzhi Wu and Mengya Xu and An Wang and Long Bai and Hongliang Ren and Amine Yamlahi and Jakob Hennighausen and Lena Maier{-}Hein and Satoshi Kondo and Satoshi Kasai and Kousuke Hirasawa and Shu Yang and Yihui Wang and Hao Chen and Santiago Rodr{\'{\i}}guez and Nicol{\'{a}}s Aparicio and Leonardo Manrique and Juan Camilo Lyons and Olivia Hosie and Nicol{\'{a}}s Ayobi and Pablo Arbel{\'{a}}ez and Yiping Li and Yasmina Al Khalil and Sahar Nasirihaghighi and Stefanie Speidel and Daniel Rueckert and Hubertus Feussner and Dirk Wilhelm and Christoph Palm}, title = {{Comparative} validation of surgical phase recognition, instrument keypoint estimation, and instrument instance segmentation in endoscopy: {Results} of the {PhaKIR} 2024 challenge}, journal = {Medical Image Analysis}, publisher = {Elsevier}, volume = {109}, pages = {103945}, year = {2026}, url = {https://doi.org/10.1016/j.media.2026.103945}, doi = {10.1016/j.media.2026.103945}} @article{rueckert2025video, author = {Tobias Rueckert and Raphaela Maerkl and David Rauber and Leonard Klausmann and Max Gutbrod and Daniel Rueckert and Hubertus Feussner and Dirk Wilhelm and Christoph Palm}, title = {{Video} {Dataset} for {Surgical} {Phase}, {Keypoint}, and {Instrument} {Recognition} in {Laparoscopic} {Surgery} {(PhaKIR)}}, journal = {CoRR}, volume = {abs/2511.06549}, year = {2025}, url = {https://doi.org/10.48550/arXiv.2511.06549}, doi = {10.48550/arXiv.2511.06549} }