SCD: A Stacked Carton Dataset for Detection and Segmentation

Abstract

Carton detection is an important technique in the automatic logistics system and can be applied to many applications such as the stacking and unstacking of cartons, the unloading of cartons in the containers. However, there is no public large-scale carton dataset for the research community to train and evaluate the carton detection models up to now, which hinders the development of carton detection. In this paper, we present a large-scale carton dataset named Stacked Carton Dataset(SCD) with the goal of advancing the state-of-the-art in carton detection. Images are collected from the internet and several warehourses, and objects are labeled using per-instance segmentation for precise localization. There are totally 250,000 instance masks from 16,136 images. In addition, we design a carton detector based on RetinaNet by embedding Boundary Guided Supervision module(BGS) and Offset Prediction between Classification and Localization module(OPCL). OPCL alleviates the imbalance problem between classification and localization quality which boosts AP by 3.1% ∼ 4.7% on SCD while BGS guides the detector to pay more attention to boundary information of cartons and decouple repeated carton textures. To demonstrate the generalization of OPCL to other datasets, we conduct extensive experiments on MS COCO and PASCAL VOC. The improvements of AP on MS COCO and PASCAL VOC are 1.8% ∼ 2.2% and 3.4% ∼ 4.3% respectively.

Carton dataset synthesis method for domain shift based on foreground texture decoupling and replacement

“…SCD [10] mainly focus on the task of carton detection in the logistics industry. The images in SCD are collected from 3 scenarios of different locations.…”

“…For example, McLaughlin N et al [32] proposed to extract the foreground of the target dataset and then fuse with the street background to reduce the bias of the datasets. Georgakis G et al [33] used the depth information of the scene and combined with foreground examples Firstly, we have built a Stacked Carton Datasets(SCD) [10]; secondly, we label the images with our labeling rules; thirdly, we perform surfaces segmentation algorithm to get different surfaces; then we use the contour reconstruction algorithm to construct a complete quadrilateral contour in part 4; next we use Gaussian filtering algorithm for foreground texture replacement and image synthesis; finally, we use the generated data to train the detection model.…”

“…(1) Collect source images: We have built a Stacked Carton Dataset(SCD) [10] for carton detection task in the warehousing and logistics industry.…”

“…One of the most critical techniques for these intelligent robots is the detection of dense cartons. We have built a carton dataset SCD [10] for carton detection, but the goods are diverse in different spots. So the carton datasets from different scenes are inconsistent in carton logo, color, texture, etc as shown in Figure .2, which means detectors trained by SCD can not satisfy the assumption of independent and identical distribution.…”

Gaussian Guided IoU: A Better Metric for Balanced Learning on Object Detection

“…Dataset. All detection baselines are trained and evaluated on three object detection datasets including MS COCO [13], PASCAL VOC [6] and SCD [23]. MS COCO contains about 118K images for training(train-2017), 5k images for validation(val-2017) and 20k for testing(testdev).…”