What Is Considered Complete for Visual Recognition?

This is an opinion paper. We hope to deliver a key message that current visual recognition systems are far from complete, i.e., recognizing everything that human can recognize, yet it is very unlikely that the gap can be bridged by continuously increasing human annotations. Based on the observation, we advocate for a new type of pre-training task named learning-by-compression. The computational models (e.g., a deep network) are optimized to represent the visual data using compact features, and the features preserve the ability to recover the original data. Semantic annotations, when available, play the role of weak supervision. An important yet challenging issue is the evaluation of image recovery, where we suggest some design principles and future research directions. We hope our proposal can inspire the community to pursue the compression-recovery tradeoff rather than the accuracy-complexity tradeoff.

Semi-supervised Contrastive Learning with Similarity Co-calibration

Semi-supervised learning acts as an effective way to leverage massive unlabeled data. In this paper, we propose a novel training strategy, termed as Semi-supervised Contrastive Learning (SsCL), which combines the well-known contrastive loss in self-supervised learning with the cross entropy loss in semi-supervised learning, and jointly optimizes the two objectives in an end-to-end way. The highlight is that different from self-training based semi-supervised learning that conducts prediction and retraining over the same model weights, SsCL interchanges the predictions over the unlabeled data between the two branches, and thus formulates a co-calibration procedure, which we find is beneficial for better prediction and avoid being trapped in local minimum. Towards this goal, the contrastive loss branch models pairwise similarities among samples, using the nearest neighborhood generated from the cross entropy branch, and in turn calibrates the prediction distribution of the cross entropy branch with the contrastive similarity. We show that SsCL produces more discriminative representation and is beneficial to few shot learning. Notably, on ImageNet with ResNet50 as the backbone, SsCL achieves 60.2% and 72.1% top-1 accuracy with 1% and 10% labeled samples, respectively, which significantly outperforms the baseline, and is better than previous semi-supervised and self-supervised methods.

Swin-Unet: Unet-like Pure Transformer for Medical Image Segmentation

In the past few years, convolutional neural networks (CNNs) have achieved milestones in medical image analysis. Especially, the deep neural networks based on U-shaped architecture and skip-connections have been widely applied in a variety of medical image tasks. However, although CNN has achieved excellent performance, it cannot learn global and long-range semantic information interaction well due to the locality of the convolution operation. In this paper, we propose Swin-Unet, which is an Unet-like pure Transformer for medical image segmentation. The tokenized image patches are fed into the Transformer-based U-shaped Encoder-Decoder architecture with skip-connections for local-global semantic feature learning. Specifically, we use hierarchical Swin Transformer with shifted windows as the encoder to extract context features. And a symmetric Swin Transformer-based decoder with patch expanding layer is designed to perform the up-sampling operation to restore the spatial resolution of the feature maps. Under the direct down-sampling and up-sampling of the inputs and outputs by 4x, experiments on multi-organ and cardiac segmentation tasks demonstrate that the pure Transformer-based U-shaped Encoder-Decoder network outperforms those methods with full-convolution or the combination of transformer and convolution. The codes and trained models will be publicly available at https://github.com/HuCaoFighting/Swin-Unet.

Deep Deformation Detail Synthesis for Thin Shell Models

In physics-based cloth animation, rich folds and detailed wrinkles are achieved at the cost of expensive computational resources and huge labor tuning. Data-driven techniques make efforts to reduce the computation significantly by a database. One type of methods relies on human poses to synthesize fitted garments which cannot be applied to general cloth. Another type of methods adds details to the coarse meshes without such restrictions. However, existing works usually utilize coordinate-based representations which cannot cope with large-scale deformation, and requires dense vertex correspondences between coarse and fine meshes. Moreover, as such methods only add details, they require coarse meshes to be close to fine meshes, which can be either impossible, or require unrealistic constraints when generating fine meshes. To address these challenges, we develop a temporally and spatially as-consistent-as-possible deformation representation (named TS-ACAP) and a DeformTransformer network to learn the mapping from low-resolution meshes to detailed ones. This TS-ACAP representation is designed to ensure both spatial and temporal consistency for sequential large-scale deformations from cloth animations. With this representation, our DeformTransformer network first utilizes two mesh-based encoders to extract the coarse and fine features, respectively. To transduct the coarse features to the fine ones, we leverage the Transformer network that consists of frame-level attention mechanisms to ensure temporal coherence of the prediction. Experimental results show that our method is able to produce reliable and realistic animations in various datasets at high frame rates: 10 ~ 35 times faster than physics-based simulation, with superior detail synthesis abilities than existing methods.

Rethinking Rotated Object Detection with Gaussian Wasserstein Distance Loss

Boundary discontinuity and its inconsistency to the final detection metric have been the bottleneck for rotating detection regression loss design. In this paper, we propose a novel regression loss based on Gaussian Wasserstein distance as a fundamental approach to solve the problem. Specifically, the rotated bounding box is converted to a 2-D Gaussian distribution, which enables to approximate the indifferentiable rotational IoU induced loss by the Gaussian Wasserstein distance (GWD) which can be learned efficiently by gradient back-propagation. GWD can still be informative for learning even there is no overlapping between two rotating bounding boxes which is often the case for small object detection. Thanks to its three unique properties, GWD can also elegantly solve the boundary discontinuity and square-like problem regardless how the bounding box is defined. Experiments on five datasets using different detectors show the effectiveness of our approach. Codes are available at https://github.com/yangxue0827/RotationDetection.

Hierarchical Semantic Aggregation for Contrastive Representation Learning

Self-supervised learning based on instance discrimination has shown remarkable progress. In particular, contrastive learning, which regards each image as well as its augmentations as a separate class, and pushes all other images away, has been proved effective for pretraining. However, contrasting two images that are de facto similar in semantic space is hard for optimization and not applicable for general representations. In this paper, we tackle the representation inefficiency of contrastive learning and propose a hierarchical training strategy to explicitly model the invariance to semantic similar images in a bottom-up way. This is achieved by extending the contrastive loss to allow for multiple positives per anchor, and explicitly pulling semantically similar images/patches together at the earlier layers as well as the last embedding space. In this way, we are able to learn feature representation that is more discriminative throughout different layers, which we find is beneficial for fast convergence. The hierarchical semantic aggregation strategy produces more discriminative representation on several unsupervised benchmarks. Notably, on ImageNet with ResNet-50 as backbone, we reach $76.4\%$ top-1 accuracy with linear evaluation, and $75.1\%$ top-1 accuracy with only $10\%$ labels.

Heterogeneous Contrastive Learning: Encoding Spatial Information for Compact Visual Representations

Contrastive learning has achieved great success in self-supervised visual representation learning, but existing approaches mostly ignored spatial information which is often crucial for visual representation. This paper presents heterogeneous contrastive learning (HCL), an effective approach that adds spatial information to the encoding stage to alleviate the learning inconsistency between the contrastive objective and strong data augmentation operations. We demonstrate the effectiveness of HCL by showing that (i) it achieves higher accuracy in instance discrimination and (ii) it surpasses existing pre-training methods in a series of downstream tasks while shrinking the pre-training costs by half. More importantly, we show that our approach achieves higher efficiency in visual representations, and thus delivers a key message to inspire the future research of self-supervised visual representation learning.

Scene text removal via cascaded text stroke detection and erasing

Recent learning-based approaches show promising performance improvement for scene text removal task. However, these methods usually leave some remnants of text and obtain visually unpleasant results. In this work, we propose a novel "end-to-end" framework based on accurate text stroke detection. Specifically, we decouple the text removal problem into text stroke detection and stroke removal. We design a text stroke detection network and a text removal generation network to solve these two sub-problems separately. Then, we combine these two networks as a processing unit, and cascade this unit to obtain the final model for text removal. Experimental results demonstrate that the proposed method significantly outperforms the state-of-the-art approaches for locating and erasing scene text. Since current publicly available datasets are all synthetic and cannot properly measure the performance of different methods, we therefore construct a new real-world dataset, which will be released to facilitate the relevant research.

Can Semantic Labels Assist Self-Supervised Visual Representation Learning?

Recently, contrastive learning has largely advanced the progress of unsupervised visual representation learning. Pre-trained on ImageNet, some self-supervised algorithms reported higher transfer learning performance compared to fully-supervised methods, seeming to deliver the message that human labels hardly contribute to learning transferrable visual features. In this paper, we defend the usefulness of semantic labels but point out that fully-supervised and self-supervised methods are pursuing different kinds of features. To alleviate this issue, we present a new algorithm named Supervised Contrastive Adjustment in Neighborhood (SCAN) that maximally prevents the semantic guidance from damaging the appearance feature embedding. In a series of downstream tasks, SCAN achieves superior performance compared to previous fully-supervised and self-supervised methods, and sometimes the gain is significant. More importantly, our study reveals that semantic labels are useful in assisting self-supervised methods, opening a new direction for the community.

Center-wise Local Image Mixture For Contrastive Representation Learning

Recent advances in unsupervised representation learning have experienced remarkable progress, especially with the achievements of contrastive learning, which regards each image as well its augmentations as a separate class, while does not consider the semantic similarity among images. This paper proposes a new kind of data augmentation, named Center-wise Local Image Mixture, to expand the neighborhood space of an image. CLIM encourages both local similarity and global aggregation while pulling similar images. This is achieved by searching local similar samples of an image, and only selecting images that are closer to the corresponding cluster center, which we denote as center-wise local selection. As a result, similar representations are progressively approaching the clusters, while do not break the local similarity. Furthermore, image mixture is used as a smoothing regularization to avoid overconfidence on the selected samples. Besides, we introduce multi-resolution augmentation, which enables the representation to be scale invariant. Integrating the two augmentations produces better feature representation on several unsupervised benchmarks. Notably, we reach 75.5% top-1 accuracy with linear evaluation over ResNet-50, and 59.3% top-1 accuracy when fine-tuned with only 1% labels, as well as consistently outperforming supervised pretraining on several downstream transfer tasks.