Thin cuvette speckle sample movies

Sample movies in RAW6 format of the dynamic speckle patterns from spherical nanoparticles suspensions in a thin (0.11 mm) cuvette. The complete set (as of Nov 2, 2020) consists of 639 files, which gives ~ 320000 scattering images. The images were used to train, validate and test neural networks (classifiers) in order to automatically recognise suspensions. The movies are hardly recognisable to the human experimenter, while the convolution network performed very well. Observation at 26.9 deg (from the forward). Illumination with polarised 532 nm light, 1.7 mm collimated beam. Camera Pike F-032C, AVT. In order to make use of the 14-bit raw images, a dedicated codec is required. Filters converting RAW6 to 24-bit RGB are provided by the camera manufacturer. Experimental details in the paper to be published. The files are labelled with substance name, nanoparticles size and suspension dilution. Every movie is ~500 frames long and was recorded at 80 fps. Since scattered light intensity differed by several orders of magnitude between the samples, the exposition (and if necessary – the camera gain) had to be adjusted. Only several discrete values of these were used. In case of doubt, whether the dynamic range was fully utilised, the measurement was repeated for a neighbouring exposition/gain value. Changing the camera parameters between the experimental runs – in particular, the exposition (image integration time) and gain (also amplifying the electronic noise) – may introduce non-meritorical information into the pictures and mislead/hinder the learning. Some difference of brightness between movies (series of images) seems unavoidable and was coped with at the data preparation stage. _v... indicates the number of the location in the cuvette本数据集包含装于厚度0.11 mm比色皿中的球形纳米颗粒悬浮液的动态散斑图案RAW6格式影片样本。截至2020年11月2日的完整数据集共包含639个文件，对应约320000幅散射图像。 该图像集被用于训练、验证及测试神经网络（分类器），以实现悬浮液的自动识别。人类实验者几乎无法直接辨识这些影片中的内容，但卷积神经网络却表现极佳。 实验采用前向26.9°的观测角度，使用偏振532 nm激光照明，准直光束直径为1.7 mm。成像设备为AVT公司的Pike F-032C相机。由于使用14位原始图像，需搭配专用编解码器方可读取；相机厂商提供了将RAW6格式转换为24位RGB格式的转换工具。实验详细细节将见于待发表的论文中。 所有文件均以物质名称、纳米颗粒尺寸及悬浮液稀释比例进行标注。 每段影片时长约500帧，采集帧率为80 fps。由于不同样本的散射光强度相差可达数个数量级，需对曝光参数（必要时还需调整相机增益）进行调节，且仅使用若干离散的参数组合。若无法确认是否充分利用了相机动态范围，则需更换相邻的曝光/增益参数重新采集。在不同实验批次间更改相机参数——尤其是曝光（即图像积分时间）与增益（增益同时会放大电子噪声）——可能会引入无效信息，干扰或阻碍模型学习。不同影片（即图像序列）间存在一定亮度差异在所难免，该问题已在数据预处理阶段得到处理。 文件名中的_v后缀代表比色皿内的采样位置编号。