If it is the author's pre-published version, changes introduced as a result of publishing processes such as copy-editing and formatting may not be reflected in this document. For a definitive version of this work, please refer to the published version.

Deep learning has achieved tremendous success with independent and identically distributed (i.i.d.) data. However, the performance of neural networks often degenerates drastically when encountering out-of-distribution (OoD) data, i.e., training and test data are sampled from different distributions. While a plethora of algorithms has been proposed to deal with OoD generalization, our understanding of the data used to train and evaluate these algorithms remains stagnant. In this work, we position existing datasets and algorithms from various research areas (e.g., domain generalization, stable learning, invariant risk minimization) seemingly unconnected into the same coherent picture. First, we identify and measure two distinct kinds of distribution shifts that are ubiquitous in various datasets. Next, we compare various OoD generalization algorithms with a new benchmark dominated by the two distribution shifts. Through extensive experiments, we show that existing OoD algorithms that outperform empirical risk minimization on one distribution shift usually have limitations on the other distribution shift. The new benchmark may serve as a strong foothold that can be resorted to by future OoD generalization research. * These authors contributed equally. 2 In the rest of this paper, we make no distinction between OoD generalization and DG (or between environments and domains), but the former is preferred for a greater audience.Preprint. Under review.

Due to the poor generalization performance of traditional empirical risk minimization (ERM) in the case of distributional shift, Out-of-Distribution (OoD) generalization algorithms receive increasing attention. However, OoD generalization algorithms overlook the great variance in the quality of training data, which significantly compromises the accuracy of these methods. In this paper, we theoretically reveal the relationship between training data quality and algorithm performance, and analyze the optimal regularization scheme for Lipschitz regularized invariant risk minimization. A novel algorithm is proposed based on the theoretical results to alleviate the influence of low quality data at both the sample level and the domain level. The experiments on both the regression and classification benchmarks validate the effectiveness of our method with statistical significance.

Vertical retargeting for stereoscopic images using seam manipulation-based approaches has remained an open challenge over the years. Even though horizontal retargeting had attracted a huge amount of interest, its seam coupling strategies were not capable to construct valid seam pairs for vertical retargeting. In this article, we propose two seam coupling strategies for vertical retargeting, namely, real mapping and virtual mapping. Our proposed mapping strategies were implemented to address the problems of multiple assignments and missing assignments, which are able to occur in the straightforward generalization from horizontal retargeting to vertical retargeting. On the basis of our proposed method, stereo seams were allowed to lay across occluded regions and occluding regions in stereo images. We maintained the geometric consistency by removing occluded pixels and corresponding occluding pixels in both stereo images. As a result, our method guarantees valid and geometrically consistent stereo seam pairs to be found in the horizontal direction. We generate vertically retargeted stereo images by removing or adding horizontal seam pairs iteratively. We conducted experiments on a number of indoor and outdoor scenes. Experimental results demonstrated that our method overcomes the limitations of vertical retargeting and is effective in preserving the geometric consistency.

No abstract