Contrastive learning has shown great promise over annotation scarcity problems in the context of medical image segmentation. Existing approaches typically assume a balanced class distribution for both labeled and unlabeled medical images. However, medical image data in reality is commonly imbalanced (i.e., multi-class label imbalance), which naturally yields blurry contours and usually incorrectly labels rare objects. Moreover, it remains unclear whether all negative samples are equally negative. In this work, we present ACTION, an Anatomical-aware ConTrastive dIstillatiON framework, for semi-supervised medical image segmentation. Specifically, we first develop an iterative contrastive distillation algorithm by softly labeling the negatives rather than binary supervision between positive and negative pairs. We also capture more semantically similar features from the randomly chosen negative set compared to the positives to enforce the diversity of the sampled data. Second, we raise a more important question: Can we really handle imbalanced samples to yield better performance? Hence, the key innovation in ACTION is to learn global semantic relationship across the entire dataset and local anatomical features among the neighbouring pixels with minimal additional memory footprint. During the training, we introduce anatomical contrast by actively sampling a sparse set of hard negative pixels, which can generate smoother segmentation boundaries and more accurate predictions. Extensive experiments across two benchmark datasets and different unlabeled settings show that ACTION significantly outperforms the current state-of-the-art semi-supervised methods.
Almost all multi-agent reinforcement learning algorithms without communication follow the principle of centralized training with decentralized execution. During centralized training, agents can be guided by the same signals, such as the global state. During decentralized execution, however, agents lack the shared signal. Inspired by viewpoint invariance and contrastive learning, we propose consensus learning for cooperative multi-agent reinforcement learning in this paper. Although based on local observations, different agents can infer the same consensus in discrete space. During decentralized execution, we feed the inferred consensus as an explicit input to the network of agents, thereby developing their spirit of cooperation. Our proposed method can be extended to various multi-agent reinforcement learning algorithms with small model changes. Moreover, we carry out them on some fully cooperative tasks and get convincing results.
Data augmentation is a crucial component in unsupervised contrastive learning (CL). It determines how positive samples are defined and, ultimately, the quality of the learned representation. In this work, we open the door to new perspectives for CL by integrating prior knowledge, given either by generative models -- viewed as prior representations -- or weak attributes in the positive and negative sampling. To this end, we use kernel theory to propose a novel loss, called decoupled uniformity, that i) allows the integration of prior knowledge and ii) removes the negative-positive coupling in the original InfoNCE loss. We draw a connection between contrastive learning and conditional mean embedding theory to derive tight bounds on the downstream classification loss. In an unsupervised setting, we empirically demonstrate that CL benefits from generative models to improve its representation both on natural and medical images. In a weakly supervised scenario, our framework outperforms other unconditional and conditional CL approaches.
Self-supervised pretraining on unlabeled data followed by supervised fine-tuning on labeled data is a popular paradigm for learning from limited labeled examples. We extend this paradigm to the classical positive unlabeled (PU) setting, where the task is to learn a binary classifier given only a few labeled positive samples, and (often) a large amount of unlabeled samples (which could be positive or negative). We first propose a simple extension of standard infoNCE family of contrastive losses, to the PU setting; and show that this learns superior representations, as compared to existing unsupervised and supervised approaches. We then develop a simple methodology to pseudo-label the unlabeled samples using a new PU-specific clustering scheme; these pseudo-labels can then be used to train the final (positive vs. negative) classifier. Our method handily outperforms state-of-the-art PU methods over several standard PU benchmark datasets, while not requiring a-priori knowledge of any class prior (which is a common assumption in other PU methods). We also provide a simple theoretical analysis that motivates our methods.
Video-Language Pretraining (VLP), which aims to learn transferable representation to advance a wide range of video-text downstream tasks, has recently received increasing attention. Best performing works rely on large-scale, 3rd-person video-text datasets, such as HowTo100M. In this work, we exploit the recently released Ego4D dataset to pioneer Egocentric VLP along three directions. (i) We create EgoClip, a 1st-person video-text pretraining dataset comprising 3.8M clip-text pairs well-chosen from Ego4D, covering a large variety of human daily activities. (ii) We propose a novel pretraining objective, dubbed EgoNCE, which adapts video-text contrastive learning to the egocentric domain by mining egocentric-aware positive and negative samples. (iii) We introduce EgoMCQ, a development benchmark that is close to EgoClip and hence can support effective validation and fast exploration of our design decisions in EgoClip and EgoNCE. Furthermore, we demonstrate strong performance on five egocentric downstream tasks across three datasets: video-text retrieval on EPIC-KITCHENS-100; action recognition on Charades-Ego; natural language query, moment query, and object state change classification on Ego4D challenge benchmarks. The dataset and code are available at https://github.com/showlab/EgoVLP.
Graph contrastive learning (GCL) alleviates the heavy reliance on label information for graph representation learning (GRL) via self-supervised learning schemes. The core idea is to learn by maximising mutual information for similar instances, which requires similarity computation between two node instances. However, GCL is inefficient in both time and memory consumption. In addition, GCL normally requires a large number of training epochs to be well-trained on large-scale datasets. Inspired by an observation of a technical defect (i.e., inappropriate usage of Sigmoid function) commonly used in two representative GCL works, DGI and MVGRL, we revisit GCL and introduce a new learning paradigm for self-supervised graph representation learning, namely, Group Discrimination (GD), and propose a novel GD-based method called Graph Group Discrimination (GGD). Instead of similarity computation, GGD directly discriminates two groups of node samples with a very simple binary cross-entropy loss. In addition, GGD requires much fewer training epochs to obtain competitive performance compared with GCL methods on large-scale datasets. These two advantages endow GGD with very efficient property. Extensive experiments show that GGD outperforms state-of-the-art self-supervised methods on eight datasets. In particular, GGD can be trained in 0.18 seconds (6.44 seconds including data preprocessing) on ogbn-arxiv, which is orders of magnitude (10,000+) faster than GCL baselines while consuming much less memory. Trained with 9 hours on ogbn-papers100M with billion edges, GGD outperforms its GCL counterparts in both accuracy and efficiency.
While the empirical success of self-supervised learning (SSL) heavily relies on the usage of deep nonlinear models, existing theoretical works on SSL understanding still focus on linear ones. In this paper, we study the role of nonlinearity in the training dynamics of contrastive learning (CL) on one and two-layer nonlinear networks with homogeneous activation $h(x) = h'(x)x$. We have two major theoretical discoveries. First, the presence of nonlinearity can lead to many local optima even in 1-layer setting, each corresponding to certain patterns from the data distribution, while with linear activation, only one major pattern can be learned. This suggests that models with lots of parameters can be regarded as a \emph{brute-force} way to find these local optima induced by nonlinearity. Second, in the 2-layer case, linear activation is proven not capable of learning specialized weights into diverse patterns, demonstrating the importance of nonlinearity. In addition, for 2-layer setting, we also discover \emph{global modulation}: those local patterns discriminative from the perspective of global-level patterns are prioritized to learn, further characterizing the learning process. Simulation verifies our theoretical findings.
Image-classification datasets have been used to pretrain image recognition models. Recently, web-scale image-caption datasets have emerged as a source of powerful pretraining alternative. Image-caption datasets are more ``open-domain'', containing a wider variety of scene types and vocabulary words than traditional classification datasets, and models trained on these datasets have demonstrated strong performance on few- and zero-shot recognition tasks. When naively unifying image-classification and -caption dataset, we show that such dataset biases negatively affect pre-training by reducing the generalizability of learned representations and thus jeopardizing zero-shot performance since the unification can tailor the model for the classification dataset, making it vulnerable to the distribution shift from the dataset. In this work, we address the problem by disentangling the dataset bias using prefix tokens that inform a language encoder of the type of the input dataset (e.g., image-classification or caption) at training time. This approach allows the language encoder to share the knowledge from two datasets as well as switch the mode of feature extraction, i.e., image-classification dataset or image-caption dataset tailored mode, where we use image-caption mode in the zero-shot evaluation. Our method is generic and can be easily integrated into existing VL pre-training objectives such as CLIP or UniCL. In experiments, we show that this simple technique improves the performance in zero-shot image recognition accuracy and robustness to the image-level distribution shift.
Image deblurring aims to restore the detailed texture information or structures from blurry images, which has become an indispensable step in many computer vision tasks. Although various methods have been proposed to deal with the image deblurring problem, most of them treated the blurry image as a whole and neglected the characteristics of different image frequencies. In this paper, we present a new method called multi-scale frequency separation network (MSFS-Net) for image deblurring. MSFS-Net introduces the frequency separation module (FSM) into an encoder-decoder network architecture to capture the low- and high-frequency information of image at multiple scales. Then, a cycle-consistency strategy and a contrastive learning module (CLM) are respectively designed to retain the low-frequency information and recover the high-frequency information during deblurring. At last, the features of different scales are fused by a cross-scale feature fusion module (CSFFM). Extensive experiments on benchmark datasets show that the proposed network achieves state-of-the-art performance.
One of the challenges in contrastive learning is the selection of appropriate \textit{hard negative} examples, in the absence of label information. Random sampling or importance sampling methods based on feature similarity often lead to sub-optimal performance. In this work, we introduce UnReMix, a hard negative sampling strategy that takes into account anchor similarity, model uncertainty and representativeness. Experimental results on several benchmarks show that UnReMix improves negative sample selection, and subsequently downstream performance when compared to state-of-the-art contrastive learning methods.
Recent advances in contrastive learning have enlightened diverse applications across various semi-supervised fields. Jointly training supervised learning and unsupervised learning with a shared feature encoder becomes a common scheme. Though it benefits from taking advantage of both feature-dependent information from self-supervised learning and label-dependent information from supervised learning, this scheme remains suffering from bias of the classifier. In this work, we systematically explore the relationship between self-supervised learning and supervised learning, and study how self-supervised learning helps robust data-efficient deep learning. We propose hyperspherical consistency regularization (HCR), a simple yet effective plug-and-play method, to regularize the classifier using feature-dependent information and thus avoid bias from labels. Specifically, HCR first projects logits from the classifier and feature projections from the projection head on the respective hypersphere, then it enforces data points on hyperspheres to have similar structures by minimizing binary cross entropy of pairwise distances' similarity metrics. Extensive experiments on semi-supervised and weakly-supervised learning demonstrate the effectiveness of our method, by showing superior performance with HCR.
Bundle recommendation aims to recommend a bundle of related items to users, which can satisfy the users' various needs with one-stop convenience. Recent methods usually take advantage of both user-bundle and user-item interactions information to obtain informative representations for users and bundles, corresponding to bundle view and item view, respectively. However, they either use a unified view without differentiation or loosely combine the predictions of two separate views, while the crucial cooperative association between the two views' representations is overlooked. In this work, we propose to model the cooperative association between the two different views through cross-view contrastive learning. By encouraging the alignment of the two separately learned views, each view can distill complementary information from the other view, achieving mutual enhancement. Moreover, by enlarging the dispersion of different users/bundles, the self-discrimination of representations is enhanced. Extensive experiments on three public datasets demonstrate that our method outperforms SOTA baselines by a large margin. Meanwhile, our method requires minimal parameters of three set of embeddings (user, bundle, and item) and the computational costs are largely reduced due to more concise graph structure and graph learning module. In addition, various ablation and model studies demystify the working mechanism and justify our hypothesis. Codes and datasets are available at https://github.com/mysbupt/CrossCBR.
With the recent promising results of contrastive learning in the self-supervised learning paradigm, supervised contrastive learning has successfully extended these contrastive approaches to supervised contexts, outperforming cross-entropy on various datasets. However, supervised contrastive learning inherently employs label information in a binary form--either positive or negative--using a one-hot target vector. This structure struggles to adapt to methods that exploit label information as a probability distribution, such as CutMix and knowledge distillation. In this paper, we introduce a generalized supervised contrastive loss, which measures cross-entropy between label similarity and latent similarity. This concept enhances the capabilities of supervised contrastive loss by fully utilizing the label distribution and enabling the adaptation of various existing techniques for training modern neural networks. Leveraging this generalized supervised contrastive loss, we construct a tailored framework: the Generalized Supervised Contrastive Learning (GenSCL). Compared to existing contrastive learning frameworks, GenSCL incorporates additional enhancements, including advanced image-based regularization techniques and an arbitrary teacher classifier. When applied to ResNet50 with the Momentum Contrast technique, GenSCL achieves a top-1 accuracy of 77.3% on ImageNet, a 4.1% relative improvement over traditional supervised contrastive learning. Moreover, our method establishes new state-of-the-art accuracies of 98.2% and 87.0% on CIFAR10 and CIFAR100 respectively when applied to ResNet50, marking the highest reported figures for this architecture.
Deep clustering has attracted increasing attention in recent years due to its capability of joint representation learning and clustering via deep neural networks. In its latest developments, the contrastive learning has emerged as an effective technique to substantially enhance the deep clustering performance. However, the existing contrastive learning based deep clustering algorithms mostly focus on some carefully-designed augmentations (often with limited transformations to preserve the structure), referred to as weak augmentations, but cannot go beyond the weak augmentations to explore the more opportunities in stronger augmentations (with more aggressive transformations or even severe distortions). In this paper, we present an end-to-end deep clustering approach termed Strongly Augmented Contrastive Clustering (SACC), which extends the conventional two-augmentation-view paradigm to multiple views and jointly leverages strong and weak augmentations for strengthened deep clustering. Particularly, we utilize a backbone network with triply-shared weights, where a strongly augmented view and two weakly augmented views are incorporated. Based on the representations produced by the backbone, the weak-weak view pair and the strong-weak view pairs are simultaneously exploited for the instance-level contrastive learning (via an instance projector) and the cluster-level contrastive learning (via a cluster projector), which, together with the backbone, can be jointly optimized in a purely unsupervised manner. Experimental results on five challenging image datasets have shown the superiority of our SACC approach over the state-of-the-art. The code is available at https://github.com/dengxiaozhi/SACC.
Deep clustering has recently emerged as a promising technique for complex data clustering. Despite the considerable progress, previous deep clustering works mostly build or learn the final clustering by only utilizing a single layer of representation, e.g., by performing the K-means clustering on the last fully-connected layer or by associating some clustering loss to a specific layer, which neglect the possibilities of jointly leveraging multi-layer representations for enhancing the deep clustering performance. In view of this, this paper presents a Deep Clustering via Ensembles (DeepCluE) approach, which bridges the gap between deep clustering and ensemble clustering by harnessing the power of multiple layers in deep neural networks. In particular, we utilize a weight-sharing convolutional neural network as the backbone, which is trained with both the instance-level contrastive learning (via an instance projector) and the cluster-level contrastive learning (via a cluster projector) in an unsupervised manner. Thereafter, multiple layers of feature representations are extracted from the trained network, upon which the ensemble clustering process is further conducted. Specifically, a set of diversified base clusterings are generated from the multi-layer representations via a highly efficient clusterer. Then the reliability of clusters in multiple base clusterings is automatically estimated by exploiting an entropy-based criterion, based on which the set of base clusterings are re-formulated into a weighted-cluster bipartite graph. By partitioning this bipartite graph via transfer cut, the final consensus clustering can be obtained. Experimental results on six image datasets confirm the advantages of DeepCluE over the state-of-the-art deep clustering approaches.
A data augmentation module is utilized in contrastive learning to transform the given data example into two views, which is considered essential and irreplaceable. However, the predetermined composition of multiple data augmentations brings two drawbacks. First, the artificial choice of augmentation types brings specific representational invariances to the model, which have different degrees of positive and negative effects on different downstream tasks. Treating each type of augmentation equally during training makes the model learn non-optimal representations for various downstream tasks and limits the flexibility to choose augmentation types beforehand. Second, the strong data augmentations used in classic contrastive learning methods may bring too much invariance in some cases, and fine-grained information that is essential to some downstream tasks may be lost. This paper proposes a general method to alleviate these two problems by considering where and what to contrast in a general contrastive learning framework. We first propose to learn different augmentation invariances at different depths of the model according to the importance of each data augmentation instead of learning representational invariances evenly in the backbone. We then propose to expand the contrast content with augmentation embeddings to reduce the misleading effects of strong data augmentations. Experiments based on several baseline methods demonstrate that we learn better representations for various benchmarks on classification, detection, and segmentation downstream tasks.
In this paper, we introduce Cross-View Language Modeling, a simple and effective pre-training framework that unifies cross-lingual and cross-modal pre-training with shared architectures and objectives. Our approach is motivated by a key observation that cross-lingual and cross-modal pre-training share the same goal of aligning two different views of the same object into a common semantic space. To this end, the cross-view language modeling framework considers both multi-modal data (i.e., image-caption pairs) and multi-lingual data (i.e., parallel sentence pairs) as two different views of the same object, and trains the model to align the two views by maximizing the mutual information between them with conditional masked language modeling and contrastive learning. We pre-train CCLM, a Cross-lingual Cross-modal Language Model, with the cross-view language modeling framework. Empirical results on IGLUE, a multi-lingual multi-modal benchmark, and two multi-lingual image-text retrieval datasets show that while conceptually simpler, CCLM significantly outperforms the prior state-of-the-art with an average absolute improvement of over 10%. Moreover, CCLM is the first multi-lingual multi-modal pre-trained model that surpasses the translate-test performance of representative English vision-language models by zero-shot cross-lingual transfer.
Recent studies in biometric-based identification tasks have shown that deep learning methods can achieve better performance. These methods generally extract the global features as descriptor to represent the original image. Nonetheless, it does not perform well for biometric identification under fine-grained tasks. The main reason is that the single image descriptor contains insufficient information to represent image. In this paper, we present a dual global descriptor model, which combines multiple global descriptors to exploit multi level image features. Moreover, we utilize a contrastive loss to enlarge the distance between image representations of confusing classes. The proposed framework achieves the top2 on the CVPR2022 Biometrics Workshop Pet Biometric Challenge. The source code and trained models are publicly available at: https://github.com/flyingsheepbin/pet-biometrics
Self-supervised learning aims to learn a embedding space where semantically similar samples are close. Contrastive learning methods pull views of samples together and push different samples away, which utilizes semantic invariance of augmentation but ignores the relationship between samples. To better exploit the power of augmentation, we observe that semantically similar samples are more likely to have similar augmented views. Therefore, we can take the augmented views as a special description of a sample. In this paper, we model such a description as the augmentation distribution and we call it augmentation feature. The similarity in augmentation feature reflects how much the views of two samples overlap and is related to their semantical similarity. Without computational burdens to explicitly estimate values of the augmentation feature, we propose Augmentation Component Analysis (ACA) with a contrastive-like loss to learn principal components and an on-the-fly projection loss to embed data. ACA equals an efficient dimension reduction by PCA and extracts low-dimensional embeddings, theoretically preserving the similarity of augmentation distribution between samples. Empirical results show our method can achieve competitive results against various traditional contrastive learning methods on different benchmarks.
In audio-visual navigation (AVN), an intelligent agent needs to navigate to a constantly sound-making object in complex 3D environments based on its audio and visual perceptions. While existing methods attempt to improve the navigation performance with preciously designed path planning or intricate task settings, none has improved the model generalisation on unheard sounds with task settings unchanged. We thus propose a contrastive learning-based method to tackle this challenge by regularising the audio encoder, where the sound-agnostic goal-driven latent representations can be learnt from various audio signals of different classes. In addition, we consider two data augmentation strategies to enrich the training sounds. We demonstrate that our designs can be easily equipped to existing AVN frameworks to obtain an immediate performance gain (13.4%$\uparrow$ in SPL on Replica and 12.2%$\uparrow$ in SPL on MP3D). Our project is available at https://AV-GeN.github.io/.