MVTec D2S: Densely Segmented Supermarket Dataset
This provides a new benchmark for instance segmentation in industrial domains, but it is incremental as it focuses on dataset creation rather than novel methods.
The authors introduced the MVTec D2S dataset, a benchmark for instance-aware semantic segmentation in industrial settings like automated checkout, containing 21,000 high-resolution images with pixel-wise labels across 60 categories, and found that state-of-the-art methods show significant room for improvement on it.
We introduce the Densely Segmented Supermarket (D2S) dataset, a novel benchmark for instance-aware semantic segmentation in an industrial domain. It contains 21,000 high-resolution images with pixel-wise labels of all object instances. The objects comprise groceries and everyday products from 60 categories. The benchmark is designed such that it resembles the real-world setting of an automatic checkout, inventory, or warehouse system. The training images only contain objects of a single class on a homogeneous background, while the validation and test sets are much more complex and diverse. To further benchmark the robustness of instance segmentation methods, the scenes are acquired with different lightings, rotations, and backgrounds. We ensure that there are no ambiguities in the labels and that every instance is labeled comprehensively. The annotations are pixel-precise and allow using crops of single instances for articial data augmentation. The dataset covers several challenges highly relevant in the field, such as a limited amount of training data and a high diversity in the test and validation sets. The evaluation of state-of-the-art object detection and instance segmentation methods on D2S reveals significant room for improvement.