Linear sequence modeling methods, such as linear attention, state space modeling, and linear RNNs, have recently been recognized as potential alternatives to softmax attention thanks to their linear complexity and competitive performance. However, although their linear-memory advantage during training enables dealing with long sequences, it is still hard to handle extremely long sequences with very limited computational resources. In this paper, we propose Sequence Accumulation (SA) which leverages the common recurrence feature of linear sequence modeling methods to manage infinite context length even on a single GPU. Specifically, SA divides long input sequences into fixed-length sub-sequences and accumulates intermediate states sequentially, which reaches only constant-memory consumption. Additionally, we further propose Sequence Accumulation with Pipeline Parallelism (SAPP), to train large models with infinite context length, without incurring any additional synchronization costs in the sequence dimension. Extensive experiments with a wide range of context lengths are conducted to validate the effectiveness of SA and SAPP on both single and multiple GPUs. Results show that SA and SAPP enable the training of infinite context length on even very limited resources, and are well compatible with the out-of-the-box distributed training techniques.

Hybrid dynamical systems are prevalent in science and engineering to express complex systems with continuous and discrete states. To learn laws of systems, all previous methods for equation discovery in hybrid systems follow a two-stage paradigm, i.e. they first group time series into small cluster fragments and then discover equations in each fragment separately through methods in non-hybrid systems. Although effective, performance is then limited because these methods ignore the commonalities in the shared dynamics of fragments that are driven by the same equations. Besides, the two-stage paradigm breaks the interdependence between categorizing and representing dynamics that jointly form hybrid systems. In this paper, we reformulate the problem and propose an end-to-end learning framework, i.e. Amortized Equation Discovery (AMORE), to jointly categorize modes and discover equations characterizing motion dynamics of each mode by all segments of the mode. Experiments on four hybrid and six non-hybrid systems demonstrate the superior performance of our method against previous methods on equation discovery, segmentation, and forecasting.

Crowd image is arguably one of the most laborious data to annotate. In this paper, we aim to reduce the massive demand for densely labeled crowd data, and propose a novel weakly-supervised setting, in which we leverage the binary ranking of two images with high-contrast crowd counts as training guidance. To enable training under this new setting, we convert the crowd count regression problem to a ranking potential prediction problem. In particular, we tailor a Siamese Ranking Network that predicts the potential scores of two images indicating the ordering of the counts. Hence, the ultimate goal is to assign appropriate potentials for all the crowd images to ensure their orderings obey the ranking labels. On the other hand, potentials reveal the relative crowd sizes but cannot yield an exact crowd count. We resolve this problem by introducing "anchors" during the inference stage. Concretely, anchors are a few images with count labels used for referencing the corresponding counts from potential scores by a simple linear mapping function. We conduct extensive experiments to study various combinations of supervision, and we show that our method outperforms existing weakly-supervised methods by a large margin without additional labeling effort. The code is available at https://github.com/pandaszzzzz/CCRanking.

Due to domain shift, a large performance drop is usually observed when a trained crowd counting model is deployed in the wild. While existing domain-adaptive crowd counting methods achieve promising results, they typically regard each crowd image as a whole and reduce domain discrepancies in a holistic manner, thus limiting further improvement of domain adaptation performance. To this end, we propose to untangle domain-invariant crowd and domain-specific background from crowd images and design a fine-grained domain adaption method for crowd counting. Specifically, to disentangle crowd from background, we propose to learn crowd segmentation from point-level crowd counting annotations in a weakly-supervised manner. Based on the derived segmentation, we design a crowd-aware domain adaptation mechanism consisting of two crowd-aware adaptation modules, i.e., Crowd Region Transfer (CRT) and Crowd Density Alignment (CDA). The CRT module is designed to guide crowd features transfer across domains beyond background distractions. The CDA module dedicates to regularising target-domain crowd density generation by its own crowd density distribution. Our method outperforms previous approaches consistently in the widely-used adaptation scenarios.

Dynamical systems with complex behaviours, e.g. immune system cells interacting with a pathogen, are commonly modelled by splitting the behaviour in different regimes, or modes, each with simpler dynamics, and then learn the switching behaviour from one mode to another. To achieve this, Switching Dynamical Systems (SDS) are a powerful tool that automatically discovers these modes and mode-switching behaviour from time series data. While effective, these methods focus on independent objects, where the modes of one object are independent of the modes of the other objects. In this paper, we focus on the more general interacting object setting for switching dynamical systems, where the per-object dynamics also depend on an unknown and dynamically changing subset of other objects and their modes. To this end, we propose a novel graph-based approach for switching dynamical systems, GRAph Switching dynamical Systems (GRASS), in which we use a dynamic graph to characterize interactions between objects and learn both intra-object and inter-object mode-switching behaviour. For benchmarking, we create two new datasets, a synthesized ODE-driven particles dataset and a real-world Salsa-couple dancing dataset. Experiments show that GRASS can consistently outperforms previous state-of-the-art methods. We will release code and data after acceptance.

Labeling is onerous for crowd counting as it should annotate each individual in crowd images. Recently, several methods have been proposed for semi-supervised crowd counting to reduce the labeling efforts. Given a limited labeling budget, they typically select a few crowd images and densely label all individuals in each of them. Despite the promising results, we argue the None-or-All labeling strategy is suboptimal as the densely labeled individuals in each crowd image usually appear similar while the massive unlabeled crowd images may contain entirely diverse individuals. To this end, we propose to break the labeling chain of previous methods and make the first attempt to reduce spatial labeling redundancy for semi-supervised crowd counting. First, instead of annotating all the regions in each crowd image, we propose to annotate the representative ones only. We analyze the region representativeness from both vertical and horizontal directions of initially estimated density maps, and formulate them as cluster centers of Gaussian Mixture Models. Additionally, to leverage the rich unlabeled regions, we exploit the similarities among individuals in each crowd image to directly supervise the unlabeled regions via feature propagation instead of the error-prone label propagation employed in the previous methods. In this way, we can transfer the original spatial labeling redundancy caused by individual similarities to effective supervision signals on the unlabeled regions. Extensive experiments on the widely-used benchmarks demonstrate that our method can outperform previous best approaches by a large margin.

Anime is an abstract art form that is substantially different from the human portrait, leading to a challenging misaligned image translation problem that is beyond the capability of existing methods. This can be boiled down to a highly ambiguous unconstrained translation between two domains. To this end, we design a new anime translation framework by deriving the prior knowledge of a pre-trained StyleGAN model. We introduce disentangled encoders to separately embed structure and appearance information into the same latent code, governed by four tailored losses. Moreover, we develop a FaceBank aggregation method that leverages the generated data of the StyleGAN, anchoring the prediction to produce in-domain animes. To empower our model and promote the research of anime translation, we propose the first anime portrait parsing dataset, Danbooru-Parsing , containing 4,921 densely labeled images across 17 classes. This dataset connects the face semantics with appearances, enabling our new constrained translation setting. We further show the editability of our results, and extend our method to manga images, by generating the first manga parsing pseudo data. Extensive experiments demonstrate the values of our new dataset and method, resulting in the first feasible solution on anime translation.

Person re-identification aims to retrieve persons in highly varying settings across different cameras and scenarios, in which robust and discriminative representation learning is crucial. Most research considers learning representations from single images, ignoring any potential interactions between them. However, due to the high intraidentity variations, ignoring such interactions typically leads to outlier features. To tackle this issue, we propose a Neighbor Transformer Network, or NFormer, which explicitly models interactions across all input images, thus suppressing outlier features and leading to more robust representations overall. As modelling interactions between enormous amount of images is a massive task with lots of distractors, NFormer introduces two novel modules, the Landmark Agent Attention, and the Reciprocal Neighbor Softmax. Specifically, the Landmark Agent Attention efficiently models the relation map between images by a low-rank factorization with a few landmarks in feature space. Moreover, the Reciprocal Neighbor Softmax achieves sparse attention to relevant -rather than all- neighbors only, which alleviates interference of irrelevant representations and further relieves the computational burden. In experiments on four large-scale datasets, NFormer achieves a new state-of-the-art. The code is released at https://github.com/haochenheheda/NFormer.

Few-shot instance segmentation aims to train an instance segmentation model that can fast adapt to novel classes with only a few reference images. Existing methods are usually derived from standard detection models and tackle few-shot instance segmentation indirectly by conducting classification, box regression, and mask prediction on a large set of redundant proposals followed by indispensable post-processing, e.g., Non-Maximum Suppression. Such complicated hand-crafted procedures and hyperparameters lead to degraded optimization and insufficient generalization ability. In this work, we propose an end-to-end Dynamic Transformer Network, DTN for short, to directly segment all target object instances from arbitrary categories given by reference images, relieving the requirements of dense proposal generation and post-processing. Specifically, a small set of Dynamic Queries, conditioned on reference images, are exclusively assigned to target object instances and generate all the instance segmentation masks of reference categories simultaneously. Moreover, a Semantic-induced Transformer Decoder is introduced to constrain the cross-attention between dynamic queries and target images within the pixels of the reference category, which suppresses the noisy interaction with the background and irrelevant categories. Extensive experiments are conducted on the COCO-20 dataset. The experiment results demonstrate that our proposed Dynamic Transformer Network significantly outperforms the state-of-the-arts.

In this paper, we introduce a novel yet challenging research problem, interactive crowd video generation, committed to producing diverse and continuous crowd video, and relieving the difficulty of insufficient annotated real-world datasets in crowd analysis. Our goal is to recursively generate realistic future crowd video frames given few context frames, under the user-specified guidance, namely individual positions of the crowd. To this end, we propose a deep network architecture specifically designed for crowd video generation that is composed of two complementary modules, each of which combats the problems of crowd dynamic synthesis and appearance preservation respectively. Particularly, a spatio-temporal transfer module is proposed to infer the crowd position and structure from guidance and temporal information, and a point-aware flow prediction module is presented to preserve appearance consistency by flow-based warping. Then, the outputs of the two modules are integrated by a self-selective fusion unit to produce an identity-preserved and continuous video. Unlike previous works, we generate continuous crowd behaviors beyond identity annotations or matching. Extensive experiments show that our method is effective for crowd video generation. More importantly, we demonstrate the generated video can produce diverse crowd behaviors and be used for augmenting different crowd analysis tasks, i.e., crowd counting, anomaly detection, crowd video prediction. Code is available at https://github.com/Icep2020/CrowdGAN.

Pixel art is a unique art style with the appearance of low resolution images. In this paper, we propose a data-driven pixelization method that can produce sharp and crisp cell effects with controllable cell sizes. Our approach overcomes the limitation of existing learning-based methods in cell size control by introducing a reference pixel art to explicitly regularize the cell structure. In particular, the cell structure features of the reference pixel art are used as an auxiliary input for the pixelization process, and for measuring the style similarity between the generated result and the reference pixel art. Furthermore, we disentangle the pixelization process into specific cell-aware and aliasing-aware stages, mitigating the ambiguities in joint learning of cell size, aliasing effect, and color assignment. To train our model, we construct a dedicated pixel art dataset and augment it with different cell sizes and different degrees of anti-aliasing effects. Extensive experiments demonstrate its superior performance over state-of-the-arts in terms of cell sharpness and perceptual expressiveness. We also show promising results of video game pixelization for the first time. Code and dataset are available at https://github.com/WuZongWei6/Pixelization.

Due to the challenges of densely compacted body parts, nonrigid clothing items, and severe overlap in crowd scenes, human parsing needs to focus more on multilevel feature representations compared to general scene parsing tasks. Based on this observation, we propose to introduce the auxiliary task of human mask and edge detection to facilitate human parsing. Different from human parsing, which exploits the discriminative features of each category, human mask and edge detection emphasizes the boundaries of semantic parsing regions and the difference between foreground humans and background clutter, which benefits the parsing predictions of crowd scenes and small human parts. Specifically, we extract human mask and edge labels from the human parsing annotations and train a shared encoder with three independent decoders for the three mutually beneficial tasks. Furthermore, the decoder feature maps of the human mask prediction branch are further exploited as attention maps, indicating human regions to facilitate the decoding process of human parsing and human edge detection. In addition to these auxiliary tasks, we further alleviate the problem of deformed clothing items under various human poses by tracking the deformation patterns with the deformable convolution. Extensive experiments show that the proposed method can achieve superior performance against state-of-the-art methods on both single and multiple human parsing datasets. Codes and trained models are available https://github.com/ViktorLiang/MGDAN .

Present studies have discovered that state-of-the-art deep learning models can be attacked by small but well-designed perturbations. Existing attack algorithms for the image captioning task is time-consuming, and their generated adversarial examples cannot transfer well to other models. To generate adversarial examples faster and stronger, we propose to learn the perturbations by a generative model that is governed by three novel loss functions. Image feature distortion loss is designed to maximize the encoded image feature distance between original images and the corresponding adversarial examples at the image domain, and local-global mismatching loss is introduced to separate the mapping encoding representation of the adversarial images and the ground true captions from a local and global perspective in the common semantic space as far as possible cross image and caption domain. Language diversity loss is to make the image captions generated by the adversarial examples as different as possible from the correct image caption at the language domain. Extensive experiments show that our proposed generative model can efficiently generate adversarial examples that successfully generalize to attack image captioning models trained on unseen large-scale datasets or with different architectures, or even the image captioning commercial service.

Unsupervised video object segmentation (UVOS) aims at segmenting the primary objects in videos without any human intervention. Due to the lack of prior knowledge about the primary objects, identifying them from videos is the major challenge of UVOS. Previous methods often regard the moving objects as primary ones and rely on optical flow to capture the motion cues in videos, but the flow information alone is insufficient to distinguish the primary objects from the background objects that move together. This is because, when the noisy motion features are combined with the appearance features, the localization of the primary objects is misguided. To address this problem, we propose a novel reciprocal transformation network to discover primary objects by correlating three key factors: the intra-frame contrast, the motion cues, and temporal coherence of recurring objects. Each corresponds to a representative type of primary object, and our reciprocal mechanism enables an organic coordination of them to effectively remove ambiguous distractions from videos. Additionally, to exclude the information of the moving background objects from motion features, our transformation module enables to reciprocally transform the appearance features to enhance the motion features, so as to focus on the moving objects with salient appearance while removing the co-moving outliers. Experiments on the public benchmarks demonstrate that our model significantly outperforms the state-of-the-art methods. Code is available at https://github.com/OliverRensu/RTNet.

Crowd counting is challenging due to unconstrained imaging factors, e.g., background clutters, non-uniform distribution of people, large scale and perspective variations. Dealing with these problems using deep neural networks requires rich prior knowledge and multi-scale contextual representations. In this paper, we propose a Cross-stage Refinement Network (CRNet) that can refine predicted density maps progressively based on hierarchical multi-level density priors. In particular, CRNet is composed of several fully convolutional networks. They are stacked together recursively with the previous output as the next input, and each of them serves to utilize previous density output to gradually correct prediction errors of crowd areas and refine the predicted density maps at different stages. Cross-stage multi-level density priors are further exploited in our recurrent framework by the cross-stage skip layers based on ConvLSTM. To cope with different challenges of unconstrained crowd scenes, we explore different crowd-specific data augmentation methods to mimic real-world scenarios and enrich crowd feature representations from different aspects. Extensive experiments show the proposed method achieves superior performances against state-of-the-art methods on four widely-used challenging benchmarks in terms of counting accuracy and density map quality. Code and models are available at this https://github.com/lytgftyf/Crowd-Counting-via-Cross-stage-Refinement-Networks.