CHiMP Detector Datasets: Images of Sitting Drop Protein Crystallisation Experiments with Associated Image Masks of Drops and Crystals

The CHiMP Detector Datasets consist of images of protein crystallisation experiments along with corresponding zipped NumPy archive files (.npz). All images have had their histograms adjusted using the Contrast Limited Adaptive Histogram Equalization ((CLAHE) algorithm using the OpenCV library with grid size of 12 and are in JPEG format. The .npz files contain class labels and instance segmentation masks for both the experimental droplets and any crystals that an expert annotator has deemed to be interesting/mountable. To class labels and masks can be loaded in the following way:

import numpy as np
# load in the mask and class label list from .npz file located at mask_path
mask_file = np.load(mask_path)
masks = list(mask_file["masks"].astype(int))
class_labels = list(mask_file["class_labels"])

There are two datasets within this archive:

The VMXi CHiMP Detector Dataset. This consists of 237 images of resolution 1688 × 1352 pixels with corresponding masks. These images were collected on a Rock Imager 1000 (Formulatrix, USA) automated microplate imager at the VMXi experimental facility at Diamond Light Source, UK. These images and masks were used to train the VMXi CHiMP (Crystal Hits in My Plate) Detector network that performs object detection and instance segmentation of crystals in experimental micrographs using a Mask-R-CNN architecture. The files "vmxi_detector_training.csv" and "vmxi_detector_validation.csv" provide the filenames of the members of the training and validation sets respectively.

The XChem CHiMP Detector Dataset. This consists of 350 images of resolution 1024 × 1224 pixels with corresponding masks. These images were collected on a Rock Imager 1000 (Formulatrix, USA) automated microplate imager at the Crystallisation Facility@Harwell, located in the Research Complex at Harwell (RCaH). In addition to the images in the VMXi CHiMP Detector, these images were used to train the XChem CHiMP (Crystal Hits in My Plate) Detector network that performs object detection and instance segmentation of masks and crystals in experimental micrographs using a Mask-R-CNN architecture. The files "xchem_detector_training.csv" and "xchem_detector_validation.csv" provide the filenames of the members of the training and validation sets respectively.

CHiMP检测器数据集包含蛋白质结晶实验的图像以及对应的压缩NumPy存档文件（.npz）。所有图像均通过OpenCV库的对比度受限自适应直方图均衡化（Contrast Limited Adaptive Histogram Equalization，CLAHE）算法进行了直方图调整，网格尺寸设置为12，图像格式均为JPEG。.npz文件中包含类别标签与实例分割掩码，对应实验液滴以及专家标注人员认定为有研究价值/可挂载的所有晶体。类别标签与掩码可通过以下方式加载：
import numpy as np
# 从路径为mask_path的.npz文件中加载掩码与类别标签列表
mask_file = np.load(mask_path)
masks = list(mask_file["masks"].astype(int))
class_labels = list(mask_file["class_labels"])

本存档包含两个数据集：

一、VMXi CHiMP检测器数据集
该数据集包含237张分辨率为1688×1352像素的图像及对应掩码。这些图像由美国Formulatrix公司的Rock Imager 1000自动化微孔板成像仪，在英国钻石光源（Diamond Light Source）的VMXi实验站采集所得。本数据集及其对应掩码曾用于训练VMXi CHiMP（Crystal Hits in My Plate，即“我的板中晶体发现”）检测器网络，该网络基于Mask-R-CNN架构实现实验显微照片中晶体的目标检测与实例分割。文件"vmxi_detector_training.csv"与"vmxi_detector_validation.csv"分别对应训练集与验证集的成员文件名。

二、XChem CHiMP检测器数据集
该数据集包含350张分辨率为1024×1224像素的图像及对应掩码。这些图像由美国Formulatrix公司的Rock Imager 1000自动化微孔板成像仪，在位于哈威尔研究中心（RCaH）的哈威尔结晶设施（Crystallisation Facility@Harwell）采集所得。除VMXi CHiMP检测器数据集包含的图像外，本数据集还曾用于训练XChem CHiMP（Crystal Hits in My Plate，即“我的板中晶体发现”）检测器网络，该网络基于Mask-R-CNN架构实现实验显微照片中掩码与晶体的目标检测及实例分割。文件"xchem_detector_training.csv"与"xchem_detector_validation.csv"分别对应训练集与验证集的成员文件名。