Object recognition and instance segmentation are fundamental skills in any robotic or autonomous system. Existing state-of-the-art methods are often unable to capture meaningful uncertainty in challenging or ambiguous scenes, and as such can cause critical errors in high-performance applications. In this paper, we explore a class of distributional instance segmentation models using latent codes that can model uncertainty over plausible hypotheses of object masks. For robotic picking applications, we propose a confidence mask method to achieve the high precision necessary in industrial use cases. We show that our method can significantly reduce critical errors in robotic systems, including our newly released dataset of ambiguous scenes in a robotic application. On a real-world apparel-picking robot, our method significantly reduces double pick errors while maintaining high performance.
We introduce a distributional instance segmentation model using latent codes, Latent-MaskRCNN, which can predict multiple hypotheses of object masks
We propose new methods for using the output of a distributional instance segmentation model. For robotic applications, we propose high-precision predictions with Confidence Masks, and we achieve high recall with Union-NMS.
We are releasing a dataset of over 5000 annotated images from a real-world robotics application that highlights the ambiguity in instance segmentation.
The dataset has about 5000 images from an industrial robot picking application.
We provide train & val splits in the detectron2 format.
The dataset can be found here: https://www.dropbox.com/s/ujrqxfw2zu1xbyg/Apparel5kDataset.zip?dl=0
The code for our implementation is here: https://github.com/wyndwarrior/latent-maskrcnn