This paper studies "unsupervised finetuning", the symmetrical problem of the well-known "supervised finetuning". Given a pretrained model and small-scale unlabeled target data, unsupervised finetuning is to adapt the representation pretrained from the source domain to the target domain so that better transfer performance can be obtained. This problem is more challenging than the supervised counterpart, as the low data density in the small-scale target data is not friendly for unsupervised learning, leading to the damage of the pretrained representation and poor representation in the target domain. In this paper, we find the source data is crucial when shifting the finetuning paradigm from supervise to unsupervise, and propose two simple and effective strategies to combine source and target data into unsupervised finetuning: "sparse source data replaying", and "data mixing". The motivation of the former strategy is to add a small portion of source data back to occupy their pretrained representation space and help push the target data to reside in a smaller compact space; and the motivation of the latter strategy is to increase the data density and help learn more compact representation. To demonstrate the effectiveness of our proposed ``unsupervised finetuning'' strategy, we conduct extensive experiments on multiple different target datasets, which show better transfer performance than the naive strategy.
Automatic modulation classification (AMC) is of crucial importance for realizing wireless intelligence communications. Many deep learning based models especially convolution neural networks (CNNs) have been proposed for AMC. However, the computation cost is very high, which makes them inappropriate for beyond the fifth generation wireless communication networks that have stringent requirements on the classification accuracy and computing time. In order to tackle those challenges, a novel involution enabled AMC scheme is proposed by using the bottleneck structure of the residual networks. Involution is utilized instead of convolution to enhance the discrimination capability and expressiveness of the model by incorporating a self-attention mechanism. Simulation results demonstrate that our proposed scheme achieves superior classification performance and faster convergence speed comparing with other benchmark schemes.
We present Mobile-Former, a parallel design of MobileNet and Transformer with a two-way bridge in between. This structure leverages the advantage of MobileNet at local processing and transformer at global interaction. And the bridge enables bidirectional fusion of local and global features. Different with recent works on vision transformer, the transformer in Mobile-Former contains very few tokens (e.g. less than 6 tokens) that are randomly initialized, resulting in low computational cost. Combining with the proposed light-weight cross attention to model the bridge, Mobile-Former is not only computationally efficient, but also has more representation power, outperforming MobileNetV3 at low FLOP regime from 25M to 500M FLOPs on ImageNet classification. For instance, it achieves 77.9\% top-1 accuracy at 294M FLOPs, gaining 1.3\% over MobileNetV3 but saving 17\% of computations. When transferring to object detection, Mobile-Former outperforms MobileNetV3 by 8.6 AP.
This paper aims at addressing the problem of substantial performance degradation at extremely low computational cost (e.g. 5M FLOPs on ImageNet classification). We found that two factors, sparse connectivity and dynamic activation function, are effective to improve the accuracy. The former avoids the significant reduction of network width, while the latter mitigates the detriment of reduction in network depth. Technically, we propose micro-factorized convolution, which factorizes a convolution matrix into low rank matrices, to integrate sparse connectivity into convolution. We also present a new dynamic activation function, named Dynamic Shift Max, to improve the non-linearity via maxing out multiple dynamic fusions between an input feature map and its circular channel shift. Building upon these two new operators, we arrive at a family of networks, named MicroNet, that achieves significant performance gains over the state of the art in the low FLOP regime. For instance, under the constraint of 12M FLOPs, MicroNet achieves 59.4\% top-1 accuracy on ImageNet classification, outperforming MobileNetV3 by 9.6\%. Source code is at \href{https://github.com/liyunsheng13/micronet}{https://github.com/liyunsheng13/micronet}.
Unsupervised pretraining has achieved great success and many recent works have shown unsupervised pretraining can achieve comparable or even slightly better transfer performance than supervised pretraining on downstream target datasets. But in this paper, we find this conclusion may not hold when the target dataset has very few labeled samples for finetuning, \ie, few-label transfer. We analyze the possible reason from the clustering perspective: 1) The clustering quality of target samples is of great importance to few-label transfer; 2) Though contrastive learning is essential to learn how to cluster, its clustering quality is still inferior to supervised pretraining due to lack of label supervision. Based on the analysis, we interestingly discover that only involving some unlabeled target domain into the unsupervised pretraining can improve the clustering quality, subsequently reducing the transfer performance gap with supervised pretraining. This finding also motivates us to propose a new progressive few-label transfer algorithm for real applications, which aims to maximize the transfer performance under a limited annotation budget. To support our analysis and proposed method, we conduct extensive experiments on nine different target datasets. Experimental results show our proposed method can significantly boost the few-label transfer performance of unsupervised pretraining.
We present CSWin Transformer, an efficient and effective Transformer-based backbone for general-purpose vision tasks. A challenging issue in Transformer design is that global self-attention is very expensive to compute whereas local self-attention often limits the field of interactions of each token. To address this issue, we develop the Cross-Shaped Window self-attention mechanism for computing self-attention in the horizontal and vertical stripes in parallel that form a cross-shaped window, with each stripe obtained by splitting the input feature into stripes of equal width. We provide a detailed mathematical analysis of the effect of the stripe width and vary the stripe width for different layers of the Transformer network which achieves strong modeling capability while limiting the computation cost. We also introduce Locally-enhanced Positional Encoding (LePE), which handles the local positional information better than existing encoding schemes. LePE naturally supports arbitrary input resolutions, and is thus especially effective and friendly for downstream tasks. Incorporated with these designs and a hierarchical structure, CSWin Transformer demonstrates competitive performance on common vision tasks. Specifically, it achieves 85.4% Top-1 accuracy on ImageNet-1K without any extra training data or label, 53.9 box AP and 46.4 mask AP on the COCO detection task, and 51.7 mIOU on the ADE20K semantic segmentation task, surpassing previous state-of-the-art Swin Transformer backbone by +1.2, +2.0, +1.4, and +2.0 respectively under the similar FLOPs setting. By further pretraining on the larger dataset ImageNet-21K, we achieve 87.5% Top-1 accuracy on ImageNet-1K and state-of-the-art segmentation performance on ADE20K with 55.7 mIoU. The code and models will be available at https://github.com/microsoft/CSWin-Transformer.
Recently, Vision Transformer and its variants have shown great promise on various computer vision tasks. The ability of capturing short- and long-range visual dependencies through self-attention is arguably the main source for the success. But it also brings challenges due to quadratic computational overhead, especially for the high-resolution vision tasks (e.g., object detection). In this paper, we present focal self-attention, a new mechanism that incorporates both fine-grained local and coarse-grained global interactions. Using this new mechanism, each token attends the closest surrounding tokens at fine granularity but the tokens far away at coarse granularity, and thus can capture both short- and long-range visual dependencies efficiently and effectively. With focal self-attention, we propose a new variant of Vision Transformer models, called Focal Transformer, which achieves superior performance over the state-of-the-art vision Transformers on a range of public image classification and object detection benchmarks. In particular, our Focal Transformer models with a moderate size of 51.1M and a larger size of 89.8M achieve 83.5 and 83.8 Top-1 accuracy, respectively, on ImageNet classification at 224x224 resolution. Using Focal Transformers as the backbones, we obtain consistent and substantial improvements over the current state-of-the-art Swin Transformers for 6 different object detection methods trained with standard 1x and 3x schedules. Our largest Focal Transformer yields 58.7/58.9 box mAPs and 50.9/51.3 mask mAPs on COCO mini-val/test-dev, and 55.4 mIoU on ADE20K for semantic segmentation, creating new SoTA on three of the most challenging computer vision tasks.
This paper investigates two techniques for developing efficient self-supervised vision transformers (EsViT) for visual representation learning. First, we show through a comprehensive empirical study that multi-stage architectures with sparse self-attentions can significantly reduce modeling complexity but with a cost of losing the ability to capture fine-grained correspondences between image regions. Second, we propose a new pre-training task of region matching which allows the model to capture fine-grained region dependencies and as a result significantly improves the quality of the learned vision representations. Our results show that combining the two techniques, EsViT achieves 81.3% top-1 on the ImageNet linear probe evaluation, outperforming prior arts with around an order magnitude of higher throughput. When transferring to downstream linear classification tasks, EsViT outperforms its supervised counterpart on 17 out of 18 datasets. The code and models will be publicly available.
The complex nature of combining localization and classification in object detection has resulted in the flourished development of methods. Previous works tried to improve the performance in various object detection heads but failed to present a unified view. In this paper, we present a novel dynamic head framework to unify object detection heads with attentions. By coherently combining multiple self-attention mechanisms between feature levels for scale-awareness, among spatial locations for spatial-awareness, and within output channels for task-awareness, the proposed approach significantly improves the representation ability of object detection heads without any computational overhead. Further experiments demonstrate that the effectiveness and efficiency of the proposed dynamic head on the COCO benchmark. With a standard ResNeXt-101-DCN backbone, we largely improve the performance over popular object detectors and achieve a new state-of-the-art at 54.0 AP. Furthermore, with latest transformer backbone and extra data, we can push current best COCO result to a new record at 60.6 AP. The code will be released at https://github.com/microsoft/DynamicHead.