Deep MCEM for Weakly-Supervised Learning to Jointly Segment and Recognize Objects Using Very Few Expert Segmentations
| Autor: | Akshay V. Gaikwad, Suyash P. Awate |
|---|---|
| Rok vydání: | 2021 |
| Předmět: |
Class (computer programming)
business.industry Computer science Posterior probability Supervised learning ComputingMethodologies_IMAGEPROCESSINGANDCOMPUTERVISION Probabilistic logic Pattern recognition Object (computer science) Image (mathematics) Region of interest Segmentation Artificial intelligence business |
| Zdroj: | Lecture Notes in Computer Science ISBN: 9783030781903 IPMI |
| DOI: | 10.1007/978-3-030-78191-0_48 |
| Popis: | Typical methods for semantic image segmentation rely on large training sets comprising pixel-level segmentations and pixel-level classifications. In medical applications, a large number of training images with per-pixel segmentations are difficult to obtain. In addition, many applications involve images or image tiles containing a single object/region of interest, where the image/tile-level information about object/region class is readily available. We propose a novel deep-neural-network (DNN) framework for joint segmentation and recognition of objects relying on weakly-supervised learning from training sets having very few expert segmentations, but with object-class labels available for all images/tiles. For weakly-supervised learning, we propose a variational-learning framework relying on Monte Carlo expectation maximization (MCEM), inferring a posterior distribution on the missing segmentations. We design an effective Metropolis-Hastings posterior sampler coupled with sample reparametrizations to enable end-to-end learning. Our DNN first produces probabilistic segmentations of objects, and then their probabilistic classifications. Results on two publicly available real-world datasets show the benefits of our strategies of (i) joint object segmentation and recognition as well as (ii) weakly-supervised MCEM-based learning. |
| Databáze: | OpenAIRE |
| Externí odkaz: |
|