Creating a dataset for a classification or segmentation task. If an annotation file is present, the annotations are also prepared. The data-set is created based on an image-set.
Image-sets are collected images to build a data-set from, stored in the imagesets folder.
The imagesets folder contains the following folder structure:
- imagesets/[image_set_type]/[image_set_name]
Inside the [image_set_name] folder are the following files / folders
test/: test images (benchmark)trainval/: training and validation images for cross validationcategories.txt: all categories (classes) the image-set contains
Data-sets are stored in the datasets base folder.
The datasets folder contains the following folder structure:
- datasets/[data_set_type]/[data_set_name]
where
[data_set_type]is the same as the corresponding[image_set_type]and[data_set_name]is the same as the corresponding[image_set_name].
Inside the [data_set_name] folder are the following files / folders
test/: test set (benchmark)train/: training setval/: validation setcategories.txt: all categories (classes) the data-set contains
Object detection data-set can be created from a segmentation or object-detection image-set. All images are validated against the annotations, if they contain at least one annotation and that the annotation category belongs to one of the given categories. The annotations have to be in VIA v1 json format. Polygon annotations are converted into rectangle annotations for unique bounding-box generation.