Convolutional Neural Networks Rarely Learn Shape for Semantic Segmentation
This addresses the problem of understanding CNN behavior for researchers and practitioners in computer vision, particularly for segmentation tasks where shape is important, but it is incremental as it builds on existing research without introducing a new method.
The study systematically investigates whether convolutional neural networks (CNNs) learn shape information for semantic segmentation, finding that CNNs rarely learn shape in typical settings but can do so under specific conditions like when shape is the only available feature, with large receptive fields, and limited augmentations, which improves performance on out-of-distribution data.
Shape learning, or the ability to leverage shape information, could be a desirable property of convolutional neural networks (CNNs) when target objects have specific shapes. While some research on the topic is emerging, there is no systematic study to conclusively determine whether and under what circumstances CNNs learn shape. Here, we present such a study in the context of segmentation networks where shapes are particularly important. We define shape and propose a new behavioral metric to measure the extent to which a CNN utilizes shape information. We then execute a set of experiments with synthetic and real-world data to progressively uncover under which circumstances CNNs learn shape and what can be done to encourage such behavior. We conclude that (i) CNNs do not learn shape in typical settings but rather rely on other features available to identify the objects of interest, (ii) CNNs can learn shape, but only if the shape is the only feature available to identify the object, (iii) sufficiently large receptive field size relative to the size of target objects is necessary for shape learning; (iv) a limited set of augmentations can encourage shape learning; (v) learning shape is indeed useful in the presence of out-of-distribution data.