SegSTRONG-C2024

训练集包括 14 个没有损坏的模拟内窥镜手术序列和相应的二进制分割掩码，其中“1”代表机器人工具，“0”代表组织背景。在测试集中，有三个序列具有非对抗性损坏（烟雾、过度出血和低亮度）和数字损坏（ImageNet-C）。此外，三个具有替代背景的序列作为验证。参与者面临的挑战是仅在未损坏的序列上训练他们的算法，同时在二进制机器人工具分割任务中在损坏的序列上实现高性能。成功实现此基准测试中的这种非对抗性稳健性对于将研究算法转化为实际应用至关重要。它确保这些算法能够在手术过程中遇到意外但合理的复杂性时导航并有效执行。The training set consists of 14 undamaged simulated endoscopic surgical sequences and their corresponding binary segmentation masks, where "1" denotes robotic instruments and "0" denotes tissue background. The test set contains three sequences with non-adversarial corruptions (smoke, excessive bleeding, and low brightness) and digital corruptions (ImageNet-C). Additionally, three sequences with replaced backgrounds serve as the validation set. The challenge for participants is to train their algorithms solely on the undamaged sequences while achieving high performance on the corrupted sequences for the binary robotic instrument segmentation task. Successfully achieving this non-adversarial robustness in this benchmark is critical for translating researched algorithms into real-world applications. This ensures that these algorithms can navigate and perform effectively when encountering unexpected but reasonable complexities during surgical procedures.

搜集汇总

背景与挑战

背景概述

SegSTRONG-C2024是一个用于机器人辅助手术工具分割的数据集，特别关注算法在非对抗性损坏条件下的鲁棒性。它包含大量训练、验证和测试样本，并采用Dice相似系数和归一化表面距离作为评估标准，旨在推动研究算法向实际手术应用的转化。