Most of the current scene flow methods choose to model scene flow as a per point translation vector without differentiating between static and dynamic components of 3D motion. In this work we present an alternative method for end-to-end scene flow learning by joint estimation of non-rigid residual flow and ego-motion flow for dynamic 3D scenes. We propose to learn the relative rigid transformation from a pair of point clouds followed by an iterative refinement. We then learn the non-rigid flow from transformed inputs with the deducted rigid part of the flow. Furthermore, we extend the supervised framework with self-supervisory signals based on the temporal consistency property of a point cloud sequence. Our solution allows both training in a supervised mode complemented by self-supervisory loss terms as well as training in a fully self-supervised mode. We demonstrate that decomposition of scene flow into non-rigid flow and ego-motion flow along with an introduction of the self-supervisory signals allowed us to outperform the current state-of-the-art supervised methods.
Conditional sequence generation aims to instruct the generation procedure by conditioning the model with additional context information, which is a self-supervised learning issue (a form of unsupervised learning with supervision information from data itself). Unfortunately, the current state-of-the-art generative models have limitations in sequence generation with multiple attributes. In this paper, we propose a novel conditional hybrid GAN (C-Hybrid-GAN) to solve this issue. Discrete sequence with triplet attributes are separately generated when conditioned on the same context. Most importantly, relational reasoning technique is exploited to model not only the dependency inside each sequence of the attribute during the training of the generator but also the consistency among the sequences of attributes during the training of the discriminator. To avoid the non-differentiability problem in GANs encountered during discrete data generation, we exploit the Gumbel-Softmax technique to approximate the distribution of discrete-valued sequences.Through evaluating the task of generating melody (associated with note, duration, and rest) from lyrics, we demonstrate that the proposed C-Hybrid-GAN outperforms the existing methods in context-conditioned discrete-valued sequence generation.
We considered a novel practical problem of online learning with episodically revealed rewards, motivated by several real-world applications, where the contexts are nonstationary over different episodes and the reward feedbacks are not always available to the decision making agents. For this online semi-supervised learning setting, we introduced Background Episodic Reward LinUCB (BerlinUCB), a solution that easily incorporates clustering as a self-supervision module to provide useful side information when rewards are not observed. Our experiments on a variety of datasets, both in stationary and nonstationary environments of six different scenarios, demonstrated clear advantages of the proposed approach over the standard contextual bandit. Lastly, we introduced a relevant real-life example where this problem setting is especially useful.
Video representation learning has recently attracted attention in computer vision due to its applications for activity and scene forecasting or vision-based planning and control. Video prediction models often learn a latent representation of video which is encoded from input frames and decoded back into images. Even when conditioned on actions, purely deep learning based architectures typically lack a physically interpretable latent space. In this study, we use a differentiable physics engine within an action-conditional video representation network to learn a physical latent representation. We propose supervised and self-supervised learning methods to train our network and identify physical properties. The latter uses spatial transformers to decode physical states back into images. The simulation scenarios in our experiments comprise pushing, sliding and colliding objects, for which we also analyze the observability of the physical properties. In experiments we demonstrate that our network can learn to encode images and identify physical properties like mass and friction from videos and action sequences in the simulated scenarios. We evaluate the accuracy of our supervised and self-supervised methods and compare it with a system identification baseline which directly learns from state trajectories. We also demonstrate the ability of our method to predict future video frames from input images and actions.
We propose a webly-supervised representation learning method that does not suffer from the annotation unscalability of supervised learning, nor the computation unscalability of self-supervised learning. Most existing works on webly-supervised representation learning adopt a vanilla supervised learning method without accounting for the prevalent noise in the training data, whereas most prior methods in learning with label noise are less effective for real-world large-scale noisy data. We propose momentum prototypes (MoPro), a simple contrastive learning method that achieves online label noise correction, out-of-distribution sample removal, and representation learning. MoPro achieves state-of-the-art performance on WebVision, a weakly-labeled noisy dataset. MoPro also shows superior performance when the pretrained model is transferred to down-stream image classification and detection tasks. It outperforms the ImageNet supervised pretrained model by +10.5 on 1-shot classification on VOC, and outperforms the best self-supervised pretrained model by +17.3 when finetuned on 1\% of ImageNet labeled samples. Furthermore, MoPro is more robust to distribution shifts. Code and pretrained models are available at https://github.com/salesforce/MoPro.
Self-supervised representation learning is an emerging research topic for its powerful capacity in learning with unlabeled data. As a mainstream self-supervised learning method, augmentation-based contrastive learning has achieved great success in various computer vision tasks that lack manual annotations. Despite current progress, the existing methods are often limited by extra cost on memory or storage, and their performance still has large room for improvement. Here we present a self-supervised representation learning method, namely AAG, which is featured by an auxiliary augmentation strategy and GNT-Xent loss. The auxiliary augmentation is able to promote the performance of contrastive learning by increasing the diversity of images. The proposed GNT-Xent loss enables a steady and fast training process and yields competitive accuracy. Experiment results demonstrate the superiority of AAG to previous state-of-the-art methods on CIFAR10, CIFAR100, and SVHN. Especially, AAG achieves 94.5% top-1 accuracy on CIFAR10 with batch size 64, which is 0.5% higher than the best result of SimCLR with batch size 1024.
Event cameras are novel vision sensors that sample, in an asynchronous fashion, brightness increments with low latency and high temporal resolution. The resulting streams of events are of high value by themselves, especially for high speed motion estimation. However, a growing body of work has also focused on the reconstruction of intensity frames from the events, as this allows bridging the gap with the existing literature on appearance- and frame-based computer vision. Recent work has mostly approached this problem using neural networks trained with synthetic, ground-truth data. In this work we approach, for the first time, the intensity reconstruction problem from a self-supervised learning perspective. Our method, which leverages the knowledge of the inner workings of event cameras, combines estimated optical flow and the event-based photometric constancy to train neural networks without the need for any ground-truth or synthetic data. Results across multiple datasets show that the performance of the proposed self-supervised approach is in line with the state-of-the-art. Additionally, we propose a novel, lightweight neural network for optical flow estimation that achieves high speed inference with only a minor drop in performance.
A common practice in unsupervised representation learning is to use labeled data to evaluate the quality of the learned representations. This supervised evaluation is then used to guide critical aspects of the training process such as selecting the data augmentation policy. However, guiding an unsupervised training process through supervised evaluations is not possible for real-world data that does not actually contain labels (which may be the case, for example, in privacy sensitive fields such as medical imaging). Therefore, in this work we show that evaluating the learned representations with a self-supervised image rotation task is highly correlated with a standard set of supervised evaluations (rank correlation $> 0.94$). We establish this correlation across hundreds of augmentation policies, training settings, and network architectures and provide an algorithm (SelfAugment) to automatically and efficiently select augmentation policies without using supervised evaluations. Despite not using any labeled data, the learned augmentation policies perform comparably with augmentation policies that were determined using exhaustive supervised evaluations.
Various hand-crafted features representations of bio-signals rely primarily on the amplitude or power of the signal in specific frequency bands. The phase component is often discarded as it is more sample specific, and thus more sensitive to noise, than the amplitude. However, in general, the phase component also carries information relevant to the underlying biological processes. In fact, in this paper we show the benefits of learning the coupling of both phase and amplitude components of a bio-signal. We do so by introducing a novel self-supervised learning task, which we call Phase-Swap, that detects if bio-signals have been obtained by merging the amplitude and phase from different sources. We show in our evaluation that neural networks trained on this task generalize better across subjects and recording sessions than their fully supervised counterpart.
Graph classification is a widely studied problem and has broad applications. In many real-world problems, the number of labeled graphs available for training classification models is limited, which renders these models prone to overfitting. To address this problem, we propose two approaches based on contrastive self-supervised learning (CSSL) to alleviate overfitting. In the first approach, we use CSSL to pretrain graph encoders on widely-available unlabeled graphs without relying on human-provided labels, then finetune the pretrained encoders on labeled graphs. In the second approach, we develop a regularizer based on CSSL, and solve the supervised classification task and the unsupervised CSSL task simultaneously. To perform CSSL on graphs, given a collection of original graphs, we perform data augmentation to create augmented graphs out of the original graphs. An augmented graph is created by consecutively applying a sequence of graph alteration operations. A contrastive loss is defined to learn graph encoders by judging whether two augmented graphs are from the same original graph. Experiments on various graph classification datasets demonstrate the effectiveness of our proposed methods.
It is a classical compute vision problem to obtain real scene depth maps by using a monocular camera, which has been widely concerned in recent years. However, training this model usually requires a large number of artificially labeled samples. To solve this problem, some researchers use a self-supervised learning model to overcome this problem and reduce the dependence on manually labeled data. Nevertheless, the accuracy and reliability of these methods have not reached the expected standard. In this paper, we propose a new self-supervised learning method based on cascade networks. Compared with the previous self-supervised methods, our method has improved accuracy and reliability, and we have proved this by experiments. We show a cascaded neural network that divides the target scene into parts of different sight distances and trains them separately to generate a better depth map. Our approach is divided into the following four steps. In the first step, we use the self-supervised model to estimate the depth of the scene roughly. In the second step, the depth of the scene generated in the first step is used as a label to divide the scene into different depth parts. The third step is to use models with different parameters to generate depth maps of different depth parts in the target scene, and the fourth step is to fuse the depth map. Through the ablation study, we demonstrated the effectiveness of each component individually and showed high-quality, state-of-the-art results in the KITTI benchmark.
Self-supervised learning has shown great potentials in improving the video representation ability of deep neural networks by getting supervision from the data itself. However, some of the current methods tend to cheat from the background, i.e., the prediction is highly dependent on the video background instead of the motion, making the model vulnerable to background changes. To mitigate the model reliance towards the background, we propose to remove the background impact by adding the background. That is, given a video, we randomly select a static frame and add it to every other frames to construct a distracting video sample. Then we force the model to pull the feature of the distracting video and the feature of the original video closer, so that the model is explicitly restricted to resist the background influence, focusing more on the motion changes. We term our method as \emph{Background Erasing} (BE). It is worth noting that the implementation of our method is so simple and neat and can be added to most of the SOTA methods without much efforts. Specifically, BE brings 16.4% and 19.1% improvements with MoCo on the severely biased datasets UCF101 and HMDB51, and 14.5% improvement on the less biased dataset Diving48.
Pre-trained language models (PrLMs) have achieved great success on a wide range of natural language processing tasks by virtue of the universal language representation ability obtained by self-supervised learning on a large corpus. These models are pre-trained on standard plain texts with general language model (LM) training objectives, which would be insufficient to model dialogue-exclusive attributes like specificity and informativeness reflected in these tasks that are not explicitly captured by the pre-trained universal language representations. In this work, we propose dialogue-adaptive pre-training objectives (DAPO) derived from quality estimation to simulate dialogue-specific features, namely coherence, specificity, and informativeness. As the foundation for model pre-training, we synthesize a new dialogue corpus and build our training set with two unsupervised methods: 1) coherence-oriented context corruption, including utterance ordering, insertion, and replacement, to help the model capture the coherence inside the dialogue contexts; and 2) specificity-oriented automatic rescoring, which encourages the model to measure the quality of the synthesized data for dialogue-adaptive pre-training by considering specificity and informativeness. Experimental results on widely used open-domain response selection and quality estimation benchmarks show that DAPO significantly improves the baseline models and achieves state-of-the-art performance on the MuTual leaderboard, verifying the effectiveness of estimating quality evaluation factors into pre-training.
Person re-identification (Re-ID) via gait features within 3D skeleton sequences is a newly-emerging topic with several advantages. Existing solutions either rely on hand-crafted descriptors or supervised gait representation learning. This paper proposes a self-supervised gait encoding approach that can leverage unlabeled skeleton data to learn gait representations for person Re-ID. Specifically, we first create self-supervision by learning to reconstruct unlabeled skeleton sequences reversely, which involves richer high-level semantics to obtain better gait representations. Other pretext tasks are also explored to further improve self-supervised learning. Second, inspired by the fact that motion's continuity endows adjacent skeletons in one skeleton sequence and temporally consecutive skeleton sequences with higher correlations (referred as locality in 3D skeleton data), we propose a locality-aware attention mechanism and a locality-aware contrastive learning scheme, which aim to preserve locality-awareness on intra-sequence level and inter-sequence level respectively during self-supervised learning. Last, with context vectors learned by our locality-aware attention mechanism and contrastive learning scheme, a novel feature named Constrastive Attention-based Gait Encodings (CAGEs) is designed to represent gait effectively. Empirical evaluations show that our approach significantly outperforms skeleton-based counterparts by 15-40% Rank-1 accuracy, and it even achieves superior performance to numerous multi-modal methods with extra RGB or depth information. Our codes are available at https://github.com/Kali-Hac/Locality-Awareness-SGE.
We propose Corder, a self-supervised contrastive learning framework for source code model. Corder is designed to alleviate the need of labeled data for code retrieval and code summarization tasks. The pre-trained model of Corder can be used in two ways: (1) it can produce vector representation of code which can be applied to code retrieval tasks that do not have labeled data; (2) it can be used in a fine-tuning process for tasks that might still require label data such as code summarization. The key innovation is that we train the source code model by asking it to recognize similar and dissimilar code snippets through a contrastive learning objective. To do so, we use a set of semantic-preserving transformation operators to generate code snippets that are syntactically diverse but semantically equivalent. Through extensive experiments, we have shown that the code models pretrained by Corder substantially outperform the other baselines for code-to-code retrieval, text-to-code retrieval, and code-to-text summarization tasks.
Contrastive self-supervised learning (CSL) is an approach to learn useful representations by solving a pretext task that selects and compares anchor, negative and positive (APN) features from an unlabeled dataset. We present a conceptual framework that characterizes CSL approaches in five aspects (1) data augmentation pipeline, (2) encoder selection, (3) representation extraction, (4) similarity measure, and (5) loss function. We analyze three leading CSL approaches--AMDIM, CPC, and SimCLR--, and show that despite different motivations, they are special cases under this framework. We show the utility of our framework by designing Yet Another DIM (YADIM) which achieves competitive results on CIFAR-10, STL-10 and ImageNet, and is more robust to the choice of encoder and the representation extraction strategy. To support ongoing CSL research, we release the PyTorch implementation of this conceptual framework along with standardized implementations of AMDIM, CPC (V2), SimCLR, BYOL, Moco (V2) and YADIM.
Autoencoder, as an essential part of many anomaly detection methods, is lacking flexibility on normal data in complex datasets. U-Net is proved to be effective for this purpose but overfits on the training data if trained by just using reconstruction error similar to other AE-based frameworks. Puzzle-solving, as a pretext task of self-supervised learning (SSL) methods, has earlier proved its ability in learning semantically meaningful features. We show that training U-Nets based on this task is an effective remedy that prevents overfitting and facilitates learning beyond pixel-level features. Shortcut solutions, however, are a big challenge in SSL tasks, including jigsaw puzzles. We propose adversarial robust training as an effective automatic shortcut removal. We achieve competitive or superior results compared to the State of the Art (SOTA) anomaly detection methods on various toy and real-world datasets. Unlike many competitors, the proposed framework is stable, fast, data-efficient, and does not require unprincipled early stopping.
We investigate the task of learning blind image denoising networks from an unpaired set of clean and noisy images. Such problem setting generally is practical and valuable considering that it is feasible to collect unpaired noisy and clean images in most real-world applications. And we further assume that the noise can be signal dependent but is spatially uncorrelated. In order to facilitate unpaired learning of denoising network, this paper presents a two-stage scheme by incorporating self-supervised learning and knowledge distillation. For self-supervised learning, we suggest a dilated blind-spot network (D-BSN) to learn denoising solely from real noisy images. Due to the spatial independence of noise, we adopt a network by stacking 1x1 convolution layers to estimate the noise level map for each image. Both the D-BSN and image-specific noise model (CNN\_est) can be jointly trained via maximizing the constrained log-likelihood. Given the output of D-BSN and estimated noise level map, improved denoising performance can be further obtained based on the Bayes' rule. As for knowledge distillation, we first apply the learned noise models to clean images to synthesize a paired set of training images, and use the real noisy images and the corresponding denoising results in the first stage to form another paired set. Then, the ultimate denoising model can be distilled by training an existing denoising network using these two paired sets. Experiments show that our unpaired learning method performs favorably on both synthetic noisy images and real-world noisy photographs in terms of quantitative and qualitative evaluation.
Recently web applications have been widely used in enterprises to assist employees in providing effective and efficient business processes. Forecasting upcoming web events in enterprise web applications can be beneficial in many ways, such as efficient caching and recommendation. In this paper, we present a web event forecasting approach, DeepEvent, in enterprise web applications for better anomaly detection. DeepEvent includes three key features: web-specific neural networks to take into account the characteristics of sequential web events, self-supervised learning techniques to overcome the scarcity of labeled data, and sequence embedding techniques to integrate contextual events and capture dependencies among web events. We evaluate DeepEvent on web events collected from six real-world enterprise web applications. Our experimental results demonstrate that DeepEvent is effective in forecasting sequential web events and detecting web based anomalies. DeepEvent provides a context-based system for researchers and practitioners to better forecast web events with situational awareness.
Unsupervised learning has always been appealing to machine learning researchers and practitioners, allowing them to avoid an expensive and complicated process of labeling the data. However, unsupervised learning of complex data is challenging, and even the best approaches show much weaker performance than their supervised counterparts. Self-supervised deep learning has become a strong instrument for representation learning in computer vision. However, those methods have not been evaluated in a fully unsupervised setting. In this paper, we propose a simple scheme for unsupervised classification based on self-supervised representations. We evaluate the proposed approach with several recent self-supervised methods showing that it achieves competitive results for ImageNet classification (39% accuracy on ImageNet with 1000 clusters and 46% with overclustering). We suggest adding the unsupervised evaluation to a set of standard benchmarks for self-supervised learning. The code is available at https://github.com/Randl/kmeans_selfsuper