Convallaria Photon Counting Dataset (Conv-PC)

Data was recorded from a Convallaria majalis rhizome section sample slide at a Leica TCS SP8 TPE DIVE with FALCON and the HC PL IRAPO 25x dipping objective. We used 850 nm excitation at 1% laser power, HyD-RLD detector, emission range of 600 – 650 nm, pixel size 0.6 x 0.6 µm, an 8 MHz resonant scanner and 4x averaging.We imaged 5 fields of view (FOVs), each containing 512x512x2048 voxels (xyt). Four FOVs were used as training data for supervised denoising, and the fifth FOV was used as test data.We used a time binning of 4, resulting in datasets of 512x512x512 voxels, and named these low-SNR raw data 'trainingData.tif' and 'testData.tif' respectively.Furthermore, we summed these 512 frames to produce the high SNR ('ground truth') version and named it 'trainingDataGT.tif' and 'testDataGT.tif'.Each frame of the raw data contains a significant amount of image noise. All voxel values correspond to photon counts.

本数据集的数据由搭载FALCON模块与HC PL IRAPO 25x浸液物镜的徕卡TCS SP8 TPE DIVE显微镜，对铃兰（Convallaria majalis）根状茎切片样本玻片成像采集得到。实验采用850 nm激光作为激发光源，激光功率设置为1%，搭配HyD-RLD探测器，采集的发射光波段范围为600–650 nm，像素尺寸为0.6 × 0.6 µm，使用8 MHz共振扫描仪并执行4次帧平均操作。共获取5个视场（Field of View, FOV），每个视场包含512×512×2048个体素（xyt维度）。其中4个视场作为监督去噪任务的训练数据集，剩余1个视场作为测试数据集。本次实验采用4倍时间分箱处理，最终得到体素尺寸为512×512×512的数据集，并将此类低信噪比（low-SNR）原始数据分别命名为`trainingData.tif`与`testData.tif`。此外，对单组512帧的原始数据进行求和操作，以生成高信噪比的"真值（ground truth）"版本，并分别将其命名为`trainingDataGT.tif`与`testDataGT.tif`。所有原始数据的单帧图像均含有显著的图像噪声，所有体素值均对应光子计数结果。