Masked autoencoders have become popular training paradigms for self-supervised visual representation learning. These models randomly mask a portion of the input and reconstruct the masked portion according to the target representations. In this paper, we first show that a careful choice of the target representation is unnecessary for learning good representations, since different targets tend to derive similarly behaved models. Driven by this observation, we propose a multi-stage masked distillation pipeline and use a randomly initialized model as the teacher, enabling us to effectively train high-capacity models without any efforts to carefully design target representations. Interestingly, we further explore using teachers of larger capacity, obtaining distilled students with remarkable transferring ability. On different tasks of classification, transfer learning, object detection, and semantic segmentation, the proposed method to perform masked knowledge distillation with bootstrapped teachers (dBOT) outperforms previous self-supervised methods by nontrivial margins. We hope our findings, as well as the proposed method, could motivate people to rethink the roles of target representations in pre-training masked autoencoders.
相關內容
This paper studies a conceptually simple extension of Masked Autoencoders (MAE) to spatiotemporal representation learning from videos. We randomly mask out spacetime patches in videos and learn an autoencoder to reconstruct them in pixels. Interestingly, we show that our MAE method can learn strong representations with almost no inductive bias on spacetime (only except for patch and positional embeddings), and spacetime-agnostic random masking performs the best. We observe that the optimal masking ratio is as high as 90% (vs. 75% on images), supporting the hypothesis that this ratio is related to information redundancy of the data. A high masking ratio leads to a large speedup, e.g., > 4x in wall-clock time or even more. We report competitive results on several challenging video datasets using vanilla Vision Transformers. We observe that MAE can outperform supervised pre-training by large margins. We further report encouraging results of training on real-world, uncurated Instagram data. Our study suggests that the general framework of masked autoencoding (BERT, MAE, etc.) can be a unified methodology for representation learning with minimal domain knowledge.
Building scalable models to learn from diverse, multimodal data remains an open challenge. For vision-language data, the dominant approaches are based on contrastive learning objectives that train a separate encoder for each modality. While effective, contrastive learning approaches introduce sampling bias depending on the data augmentations used, which can degrade performance on downstream tasks. Moreover, these methods are limited to paired image-text data, and cannot leverage widely-available unpaired data. In this paper, we investigate whether a large multimodal model trained purely via masked token prediction, without using modality-specific encoders or contrastive learning, can learn transferable representations for downstream tasks. We propose a simple and scalable network architecture, the Multimodal Masked Autoencoder (M3AE), which learns a unified encoder for both vision and language data via masked token prediction. We provide an empirical study of M3AE trained on a large-scale image-text dataset, and find that M3AE is able to learn generalizable representations that transfer well to downstream tasks. Surprisingly, we find that M3AE benefits from a higher text mask ratio (50-90%), in contrast to BERT whose standard masking ratio is 15%, due to the joint training of two data modalities. We also provide qualitative analysis showing that the learned representation incorporates meaningful information from both image and language. Lastly, we demonstrate the scalability of M3AE with larger model size and training time, and its flexibility to train on both paired image-text data as well as unpaired data.
Masked image modeling (MIM) has been recognized as a strong and popular self-supervised pre-training approach in the vision domain. However, the interpretability of the mechanism and properties of the learned representations by such a scheme are so far not well-explored. In this work, through comprehensive experiments and empirical studies on Masked Autoencoders (MAE), we address two critical questions to explore the behaviors of the learned representations: (i) Are the latent representations in Masked Autoencoders linearly separable if the input is a mixture of two images instead of one? This can be concrete evidence used to explain why MAE-learned representations have superior performance on downstream tasks, as proven by many literature impressively. (ii) What is the degree of semantics encoded in the latent feature space by Masked Autoencoders? To explore these two problems, we propose a simple yet effective Interpretable MAE (i-MAE) framework with a two-way image reconstruction and a latent feature reconstruction with distillation loss to help us understand the behaviors inside MAE's structure. Extensive experiments are conducted on CIFAR-10/100, Tiny-ImageNet and ImageNet-1K datasets to verify the observations we discovered. Furthermore, in addition to qualitatively analyzing the characteristics of the latent representations, we examine the existence of linear separability and the degree of semantics in the latent space by proposing two novel metrics. The surprising and consistent results across the qualitative and quantitative experiments demonstrate that i-MAE is a superior framework design for interpretability research of MAE frameworks, as well as achieving better representational ability. Code is available at //github.com/vision-learning-acceleration-lab/i-mae.
Recent advances in self-supervised learning integrate Masked Modeling and Siamese Networks into a single framework to fully reap the advantages of both the two techniques. However, previous erasing-based masking scheme in masked image modeling is not originally designed for siamese networks. Existing approaches simply inherit the default loss design from previous siamese networks, and ignore the information loss and distance change after employing masking operation in the frameworks. In this paper, we propose a filling-based masking strategy called MixMask to prevent information loss due to the randomly erased areas of an image in vanilla masking method. We further introduce a dynamic loss function design with soft distance to adapt the integrated architecture and avoid mismatches between transformed input and objective in Masked Siamese ConvNets (MSCN). The dynamic loss distance is calculated according to the proposed mix-masking scheme. Extensive experiments are conducted on various datasets of CIFAR-100, Tiny-ImageNet and ImageNet-1K. The results demonstrate that the proposed framework can achieve better accuracy on linear probing, semi-supervised and {supervised finetuning}, which outperforms the state-of-the-art MSCN by a significant margin. We also show the superiority on downstream tasks of object detection and segmentation. Our source code is available at //github.com/LightnessOfBeing/MixMask.
The design of man-made objects is dominated by computer aided design (CAD) tools. Assisting design with data-driven machine learning methods is hampered by lack of labeled data in CAD's native format; the parametric boundary representation (B-Rep). Several data sets of mechanical parts in B-Rep format have recently been released for machine learning research. However, large scale databases are largely unlabeled, and labeled datasets are small. Additionally, task specific label sets are rare, and costly to annotate. This work proposes to leverage unlabeled CAD geometry on supervised learning tasks. We learn a novel, hybrid implicit/explicit surface representation for B-Rep geometry, and show that this pre-training significantly improves few-shot learning performance and also achieves state-of-the-art performance on several existing B-Rep benchmarks.
Masked image modeling has demonstrated great potential to eliminate the label-hungry problem of training large-scale vision Transformers, achieving impressive performance on various downstream tasks. In this work, we propose a unified view of masked image modeling after revisiting existing methods. Under the unified view, we introduce a simple yet effective method, termed as MaskDistill, which reconstructs normalized semantic features from teacher models at the masked positions, conditioning on corrupted input images. Experimental results on image classification and semantic segmentation show that MaskDistill achieves comparable or superior performance than state-of-the-art methods. When using the huge vision Transformer and pretraining 300 epochs, MaskDistill obtains 88.3% fine-tuning top-1 accuracy on ImageNet-1k (224 size) and 58.8% semantic segmentation mIoU metric on ADE20k (512 size). The code and pretrained models will be available at //aka.ms/unimim.
Continual Learning (CL) methods mainly focus on avoiding catastrophic forgetting and learning representations that are transferable to new tasks. Recently, Wortsman et al. (2020) proposed a CL method, SupSup, which uses a randomly initialized, fixed base network (model) and finds a supermask for each new task that selectively keeps or removes each weight to produce a subnetwork. They prevent forgetting as the network weights are not being updated. Although there is no forgetting, the performance of the supermask is sub-optimal because fixed weights restrict its representational power. Furthermore, there is no accumulation or transfer of knowledge inside the model when new tasks are learned. Hence, we propose ExSSNeT (Exclusive Supermask SubNEtwork Training), which performs exclusive and non-overlapping subnetwork weight training. This avoids conflicting updates to the shared weights by subsequent tasks to improve performance while still preventing forgetting. Furthermore, we propose a novel KNN-based Knowledge Transfer (KKT) module that dynamically initializes a new task's mask based on previous tasks for improving knowledge transfer. We demonstrate that ExSSNeT outperforms SupSup and other strong previous methods on both text classification and vision tasks while preventing forgetting. Moreover, ExSSNeT is particularly advantageous for sparse masks that activate 2-10% of the model parameters, resulting in an average improvement of 8.3% over SupSup. Additionally, ExSSNeT scales to a large number of tasks (100), and our KKT module helps to learn new tasks faster while improving overall performance. Our code is available at //github.com/prateeky2806/exessnet
This paper shows that masked autoencoders (MAE) are scalable self-supervised learners for computer vision. Our MAE approach is simple: we mask random patches of the input image and reconstruct the missing pixels. It is based on two core designs. First, we develop an asymmetric encoder-decoder architecture, with an encoder that operates only on the visible subset of patches (without mask tokens), along with a lightweight decoder that reconstructs the original image from the latent representation and mask tokens. Second, we find that masking a high proportion of the input image, e.g., 75%, yields a nontrivial and meaningful self-supervisory task. Coupling these two designs enables us to train large models efficiently and effectively: we accelerate training (by 3x or more) and improve accuracy. Our scalable approach allows for learning high-capacity models that generalize well: e.g., a vanilla ViT-Huge model achieves the best accuracy (87.8%) among methods that use only ImageNet-1K data. Transfer performance in downstream tasks outperforms supervised pre-training and shows promising scaling behavior.
We present a new method to learn video representations from large-scale unlabeled video data. Ideally, this representation will be generic and transferable, directly usable for new tasks such as action recognition and zero or few-shot learning. We formulate unsupervised representation learning as a multi-modal, multi-task learning problem, where the representations are shared across different modalities via distillation. Further, we introduce the concept of loss function evolution by using an evolutionary search algorithm to automatically find optimal combination of loss functions capturing many (self-supervised) tasks and modalities. Thirdly, we propose an unsupervised representation evaluation metric using distribution matching to a large unlabeled dataset as a prior constraint, based on Zipf's law. This unsupervised constraint, which is not guided by any labeling, produces similar results to weakly-supervised, task-specific ones. The proposed unsupervised representation learning results in a single RGB network and outperforms previous methods. Notably, it is also more effective than several label-based methods (e.g., ImageNet), with the exception of large, fully labeled video datasets.
Many tasks in natural language processing can be viewed as multi-label classification problems. However, most of the existing models are trained with the standard cross-entropy loss function and use a fixed prediction policy (e.g., a threshold of 0.5) for all the labels, which completely ignores the complexity and dependencies among different labels. In this paper, we propose a meta-learning method to capture these complex label dependencies. More specifically, our method utilizes a meta-learner to jointly learn the training policies and prediction policies for different labels. The training policies are then used to train the classifier with the cross-entropy loss function, and the prediction policies are further implemented for prediction. Experimental results on fine-grained entity typing and text classification demonstrate that our proposed method can obtain more accurate multi-label classification results.
注冊地址： 北京市海淀區羊坊店路18號2幢3層301-191