Continual learning (CL) desires that the neural network sequentially perform learning tasks from a dynamic data stream without forgetting learned knowledge. To overcome forgetting, a line of work relies on gradient projection to minimize the influence between gradients during optimization. This article focuses on a challenging problem concerning CL: When, how, and where to implement gradient projection to promote CL. Tackling this problem can be divided into two perspectives, namely the gradient direction ( when and how ) and the area of gradient conflict ( where ). First, we propose a plug-and-play method UniGrad to tackle the inconsistency of conflicting and nonconflicting gradients during optimization in CL. Second, we explore the interaction mechanism of gradient projection and loss landscape in CL, and further propose a pluggable method UniGrad-FS to improve the CL performance. In short, this work expects to overcome forgetting through an efficient gradient projection at the area where the gradient conflicts are less intense. In essence, the proposed method is a general and pluggable method that can be used in any gradient-based optimizer. For evaluation, we plug UniGrad and UniGrad-FS into two top-performing baselines (WA and MEMO). Our method shows clear improvements, i.e., boosting WA and MEMO by +2.09% and 1.72% in the 20-step of the CIFAR100 benchmark. In addition, we observe performance enhancement on all settings of CIFAR100 and Tiny-ImageNet datasets. Extensive experiments demonstrate the simplicity and effectiveness of the proposed method.

Building extraction from satellite images has been a hot research topic in the field of remote sensing image analysis. Most of the related studies are focusing on the urban areas with dense populations, while solutions for underpopulated mountainous areas are still missing. Given the scarcity of the research materials, it is still an important topic for applications like mountain hazard damage management. To fill this gap, we present a new dataset for scattered mountainous area building segmentation, consisting of the manual labels, the coordinates of latitude and longitude for 2125 satellite images of 303 diverse human settlements in the mountainous areas of southwest China. Compared with the public remote sensing image datasets, our dataset is very challenging with relatively low resolution (2.2m/pixel) and small-object, blurry-boundary and high-imbalance features. In this paper, we propose a novel Copy-Fusion data augmentation strategy, and designed a VGG-16+U-Net with Dice Focal loss to address the difficulties in this building extraction task. Experiment results have shown that our approach has achieved 81.79% mean IoU and outperformed the baseline building segmentation methods like HRNet, DeeplabV3+, and U-net+ResNet by 6.29%. Our dataset and model will be available at https://github.com/AngCV/SMAB_DATASET.