Masked autoencoder (MAE) is a promising self-supervised pre-training technique that can improve the representation learning of a neural network without human intervention. However, applying MAE directly to volumetric medical images poses two challenges: (i) a lack of global information that is crucial for understanding the clinical context of the holistic data, (ii) no guarantee of stabilizing the representations learned from randomly masked inputs. To address these limitations, we propose the \textbf{G}lobal-\textbf{L}ocal \textbf{M}asked \textbf{A}uto\textbf{E}ncoder (GL-MAE), a simple yet effective self-supervised pre-training strategy. In addition to reconstructing masked local views, as in previous methods, GL-MAE incorporates global context learning by reconstructing masked global views. Furthermore, a complete global view is integrated as an anchor to guide the reconstruction and stabilize the learning process through global-to-global consistency learning and global-to-local consistency learning. Finetuning results on multiple datasets demonstrate the superiority of our method over other state-of-the-art self-supervised algorithms, highlighting its effectiveness on versatile volumetric medical image segmentation tasks, even when annotations are scarce. Our codes and models will be released upon acceptance.

This is a technical report for CVPR 2021 AliProducts Challenge. AliProducts Challenge is a competition proposed for studying the large-scale and fine-grained commodity image recognition problem encountered by worldleading ecommerce companies. The large-scale product recognition simultaneously meets the challenge of noisy annotations, imbalanced (long-tailed) data distribution and fine-grained classification. In our solution, we adopt stateof-the-art model architectures of both CNNs and Transformer, including ResNeSt, EfficientNetV2, and DeiT. We found that iterative data cleaning, classifier weight normalization, high-resolution finetuning, and test time augmentation are key components to improve the performance of training with the noisy and imbalanced dataset. Finally, we obtain 6.4365% mean class error rate in the leaderboard with our ensemble model.

While self-supervised representation learning (SSL) has received widespread attention from the community, recent research argue that its performance will suffer a cliff fall when the model size decreases. The current method mainly relies on contrastive learning to train the network and in this work, we propose a simple yet effective Distilled Contrastive Learning (DisCo) to ease the issue by a large margin. Specifically, we find the final embedding obtained by the mainstream SSL methods contains the most fruitful information, and propose to distill the final embedding to maximally transmit a teacher's knowledge to a lightweight model by constraining the last embedding of the student to be consistent with that of the teacher. In addition, in the experiment, we find that there exists a phenomenon termed Distilling BottleNeck and present to enlarge the embedding dimension to alleviate this problem. Our method does not introduce any extra parameter to lightweight models during deployment. Experimental results demonstrate that our method achieves the state-of-the-art on all lightweight models. Particularly, when ResNet-101/ResNet-50 is used as teacher to teach EfficientNet-B0, the linear result of EfficientNet-B0 on ImageNet is very close to ResNet-101/ResNet-50, but the number of parameters of EfficientNet-B0 is only 9.4%/16.3% of ResNet-101/ResNet-50.