DT4PEIS: Detection Transformers for Parasitic Egg Instance Segmentation
Description
Supplementary Materials for "DT4PEIS: Detection Transformers for Parasitic Egg Instance Segmentation". This dataset is an extension of the publicly available dataset "Parasitic Egg Detection and Classification in Microscopic Images" (Chula-ParasiteEgg-11, with DOI: 10.21227/vyh8-4h71), now including binary segmentation masks for the images. The segmentation was performed automatically using the Segment Anything Model (SAM).
Files
Steps to reproduce
The dataset can be described the following way. 1. Main Dataset Folder (Segmented-Chula-ParasiteEgg-11) The main folder. It contains two subfolders: “Train_Masks” and “Test_Masks”. There are also two JSON files included: “train_labels_SPoly.json” and “test_labels_SPoly.json”. 2. Masks Subfolders (Train_Masks and Test_Masks): The masks stored in these folders have been generated by carrying out a process of segmentation of the original (with “original” we refer to the original dataset "Parasitic Egg Detection and Classification in Microscopic Images") images using the Segment Anything Model (SAM). The names of the masks are identical to the originals but including “_mask” before the .jpg extension. 3. JSON Files (train_labels_SPoly.json and test_labels_SPoly.json) These files constitute a modified version of the ones included in the original dataset. Ours, as well as the annotation information for each image, contain segmentation information. Two fields have been added: “iscrowd” and “segmentation”, following COCO standard format. The first one has always a 0 value, as the segmentation is in polygon format (object instance); the second one includes the encoded segmentation for each mask. This method encodes masks into multiple lists of different lengths. Each list stores X and Y coordinates alternately, capturing all the white pixels in a given row until the complete definition of the image matrix. Mathematically, the encoding of a mask would be as follows: [X1,1, Y1,1, X1,2, Y1,2, ..., X1,N, Y1,N], ... [XM,1, YM,1, XM,2, YM,2, ..., XM,N, YM,N]. Note also that when downloaded de .zip, apart from the main dataset folder, there is another one named "Figures" at the same level. This folder includes two .svg figures for the better understanding of the dataset. The figures are: 1. "dataset_structure.svg" It clarifies the explained structure of the dataset. 2. "data_collection.svg" This figure graphically explains the process followed to obtain the Segmented-Chula-ParasiteEgg-11 dataset from the Chula-ParasiteEgg-11 dataset.
Institutions
Categories
Funding
Ministerio de Ciencia, Innovación y Universidades
TED2021-132147B-100
Ministerio de Ciencia, Innovación y Universidades
PID2021-127567NB-I00
University of Castilla-La Mancha
2022-GRIN-34352