Emerging Semantic Segmentation from Positive and Negative Coarse Label Learning

Le Zhang, Fuping Wu, Kevin Bronik, Thomas E. Nichols, Bartlomiej Papiez

Primary Area: unsupervised, self-supervised, semi-supervised, and supervised representation learning

Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.

Keywords: Image Segmentation; Weakly-supervised learning; Coarse Annotation;

Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2024/AuthorGuide.

Abstract: Large annotated dataset is of crucial importance for developing machine learning models for segmentation. However, the process of producing labels at the pixel level is time-consuming, error-prone, and even requires expert-level annotators for medical imaging, which is rare to have in practice. We note that it is simpler and less expensive to draw merely rough and approximate annotations, e.g., coarse annotations, which reduce the effort for expert and non-expert level annotators. In this paper, we propose to use coarse drawings from both positive (e.g., objects to be segmented) and negative (objects not to be segmented) classes in the image, even with noisy pixels, to train a convolutional neural network (CNN) for semantic segmentation. We present a method for learning the true segmentation label distributions from purely noisy coarse annotations using two coupled CNNs. The separation of the two CNNs is achieved by high fidelity with the characters of the noisy training annotations. We propose to add a complementary label learning that encourages estimating negative label distribution. To illustrate the properties of our method, we first use a toy segmentation dataset based on MNIST. We then present the quantitative results on publicly available datasets: Cityscapes dataset for multi-class segmentation, and retinal images for medical applications. In all experiments, our method outperforms the state-of-the-art methods, particularly in the cases where the ratio of coarse annotations is small compared to the given dense annotations.

Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors' identity.

No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.

Submission Number: 458