4.9 Conclusion 103 such as object retrieval in this challenge. 4.9 Conclusion Wepresent DLSBD, a metric to quantify Linear Symmetry-Based Disentanglement (LSBD) as defined by Higgins et al. (2018). From this metric formulation we motivate LSBD-VAE, a semi-supervised method to learn LSBD representations given some expert knowledge on the underlying group symmetries that are to be disentangled. We use DLSBD to evaluate various disentanglement methods—both traditional methods and recent methods that specifically focus on LSBD—and show that LSBD-VAE can learn LSBD representations where traditional methods fail to do so. We also compare DLSBD to traditional disentanglement metrics, showing that LSBD captures many of the same desirable properties that are expressed by existing disentanglement methods. Conversely, we also show that traditional disentanglement methods and metrics do not usually achieve or measure LSBD, indicating that LSBD provides a useful new perspective on disentanglement. To show the applicability of LSBD-VAE in a real-world setting, we deploy it in the SHREC 2021 3D Object Retrieval Challenge (Sipiran et al., 2021), where we augment the LSBD-VAE with a triplet loss to accommodate retrieval. Results show that although the method has potential, there are still some practical challenges to overcome for the method to achieve state-of-the-art performance on such a retrieval task. Challenges that remain for future work are expanding and testing LSBD-VAE andDLSBDon different group structures and towards more practical applications, as well as focusing on the utility of LSBD representations for downstream tasks.
RkJQdWJsaXNoZXIy MjY0ODMw