Recent advances in text-to-video generation have demonstrated the substantial superiority of diffusion models. Nevertheless, generating high-resolution videos based on text description still faces a great challenge due to the enormous computation overhead for video diffusion model training. In this paper, we present a tuning-free video diffusion approach with Spatial-Temporal LAtent Grouping (ST-LAG), for highresolution video generation. ST-LAG exploits the prior knowledge of a pre-trained low-resolution video diffusion model for regionwise video latent denoising, and then combines all the denoised regions of video latent as a whole one to achieve global-wise spatial-temporal coherence. Specifically, ST-LAG denoises the whole video latents via two deliberately designed modules, e.g., Spatial Latent Grouping (SLG) and Temporal Latent Grouping (TLG), at spatial and temporal level, respectively. SLG spatially slices the latent of each frame into different local patches, and then feeds them into the low-resolution video diffusion model for local-region latent denoising. A text re-weighting scheme is devised in SLG to strength the cross-attention between features of text tokens and spatial regions to facilitate spatial-level finegrained details generation. TLG capitalizes on the segmentlevel latent grouping to match the length of each denoised local segment with the frame number in the training stage. The well aligned temporal receptive field facilitates better preservation of motion patterns. In each denoising step, all groups of video latent at spatial and temporal levels are fused together for highresolution video generation. Extensive experiments conducted on the ECTV-Prompt dataset demonstrate the effectiveness of our approach quantitatively and qualitatively.

The first-in-first-out (FIFO) video diffusion, built on a pre-trained text-to-video model, has recently emerged as an effective approach for tuning-free long video generation. This technique maintains a queue of video frames with progressively increasing noise, continuously producing clean frames at the queue's head while Gaussian noise is enqueued at the tail. However, FIFO-Diffusion often struggles to keep long-range temporal consistency in the generated videos due to the lack of correspondence modeling across frames. In this paper, we propose Ouroboros-Diffusion, a novel video denoising framework designed to enhance structural and content (subject) consistency, enabling the generation of consistent videos of arbitrary length. Specifically, we introduce a new latent sampling technique at the queue tail to improve structural consistency, ensuring perceptually smooth transitions among frames. To enhance subject consistency, we devise a Subject-Aware Cross-Frame Attention (SACFA) mechanism, which aligns subjects across frames within short segments to achieve better visual coherence. Furthermore, we introduce self-recurrent guidance. This technique leverages information from all previous cleaner frames at the front of the queue to guide the denoising of noisier frames at the end, fostering rich and contextual global information interaction. Extensive experiments of long video generation on the VBench benchmark demonstrate the superiority of our Ouroboros-Diffusion, particularly in terms of subject consistency, motion smoothness, and temporal consistency.

Animating images with interactive motion control has garnered popularity for image-to-video (I2V) generation. Modern approaches typically rely on large Gaussian kernels to extend motion trajectories as condition without explicitly defining movement region, leading to coarse motion control and failing to disentangle object and camera moving. To alleviate these, we present MotionPro, a precise motion controller that novelly leverages region-wise trajectory and motion mask to regulate fine-grained motion synthesis and identify target motion category (i.e., object or camera moving), respectively. Technically, MotionPro first estimates the flow maps on each training video via a tracking model, and then samples the region-wise trajectories to simulate inference scenario. Instead of extending flow through large Gaussian kernels, our region-wise trajectory approach enables more precise control by directly utilizing trajectories within local regions, thereby effectively characterizing fine-grained movements. A motion mask is simultaneously derived from the predicted flow maps to capture the holistic motion dynamics of the movement regions. To pursue natural motion control, MotionPro further strengthens video denoising by incorporating both region-wise trajectories and motion mask through feature modulation. More remarkably, we meticulously construct a benchmark, i.e., MC-Bench, with 1.1K user-annotated image-trajectory pairs, for the evaluation of both fine-grained and object-level I2V motion control. Extensive experiments conducted on WebVid-10M and MC-Bench demonstrate the effectiveness of MotionPro. Please refer to our project page for more results: https://zhw-zhang.github.io/MotionPro-page/.

Recent advances in diffusion-based image editing models have demonstrated remarkable success. However, these models primarily rely on high-quality textual prompts to guide image manipulation, creating a significant barrier for non-expert users. In this demonstration, we present an exemplar-based image editing framework named Edit-by-Example, which eliminates the reliance on textual prompts, requires only a single pair of before-and-after images to encapsulate the desired editing effect that can readily be applied on the user-provided query image without any model fine-tuning. Technically, our framework comprises two components: an Adaptive Editing Policy Module (AEPM) and a Generation Module (GM). The AEPM jointly analyzes cross-image relationships in exemplar pairs and query image content to derive optimal editing directions, while GM executes these policies through an off-the-shelf image editor with optional semantic alignment verification. We introduce EEdBench, a comprehensive benchmark for exemplar-based image editing containing 1,500 test cases across 15 categories. Experiments demonstrate that our framework outperforms existing prompt-free methods in editing direction accuracy (S-Visual) and fidelity (FID).

This paper demonstrates a pioneering unified multimodal agent that transforms complex visual content creation into an intuitive, conversational experience, allowing users to talk, imagine, and evolve their ideas. Overcoming the limitations of fragmented multimodal technique tools, our system seamlessly integrates text-to-image generation, instruction-based image editing, text/image-to-video generation, and interactive understanding within a single AI interface. Users of all skill levels can perform sophisticated visual tasks using natural language and visual inputs. The system's architecture features a central Coordinator module processing multimodal inputs and directing tasks to Generation or Chat pathways. For Generation, a Planner utilizes our state-of-the-art specialized models in image/video generation and image editing, while the Chat function facilitates clarification and collaboration. The interactive demonstration will showcase intuitive multimodal input, seamless real-time content creation/editing, dynamic interactive understanding, and a unified workflow. This agent pioneers a new way for accessible, interactive visual storytelling and collaborative content creation in multimodal generative AI.

Recent advancements in multimodal AIGC have enabled impressive text-to-video synthesis, but a critical challenge remains: maintaining consistent identity of key subjects across generated frames. To address this limitation, we introduce the Identity-Preserving Video Generation (IPVG) grand challenge. This challenge aims to propel the field toward more controllable generative models by focusing community efforts on preserving identity during the video generation process. To support these efforts, we publicly release the Identity-Preserving Video Benchmark (VIP-200K), a novel dataset comprising approximately 500,000 video-prompt pairs with 200,000 unique identities, each coupled with a reference identity image. Through this grand challenge and dataset, we provide a fertile ground for developing solutions that lead to more user-steerable video synthesis systems. The challenge homepage is https://hidream-ai.github.io/ipvg-challenge.github.io/.

Diffusion models are just at a tipping point for image super-resolution task. Nevertheless, it is not trivial to capitalize on diffusion models for video super-resolution which necessitates not only the preservation of visual appearance from low-resolution to high-resolution videos, but also the temporal consistency across video frames. In this paper, we propose a novel approach, pursuing Spatial Adaptation and Temporal Coherence (SATeCo), for video super-resolution. SATeCo pivots on learning spatial-temporal guidance from low-resolution videos to calibrate both latent-space high-resolution video denoising and pixel-space video reconstruction. Technically, SATeCo freezes all the parameters of the pre-trained UNet and VAE, and only optimizes two deliberately-designed spatial feature adaptation (SFA) and temporal feature alignment (TFA) modules, in the decoder of UNet and VAE. SFA modulates frame features via adaptively estimating affine parameters for each pixel, guaran-teeing pixel-wise guidance for high-resolution frame syn-thesis. TFA delves into feature interaction within a 3D local window (tube let) through self-attention, and executes cross-attention between tubelet and its low-resolution counterpart to guide temporal feature alignment. Extensive experiments conducted on the REDS4 and Vid4 datasets demonstrate the effectiveness of our approach.

Recent advances in text-to- video generation have demonstrated the utility of powerful diffusion models. Nev-ertheless, the problem is not trivial when shaping diffusion models to animate static image (i.e., image-to-video generation). The difficulty originates from the aspect that the diffusion process of subsequent animated frames should not only preserve the faithful alignment with the given image but also pursue temporal coherence among adjacent frames. To alleviate this, we present TRIP, a new recipe of image-to-video diffusion paradigm that pivots on image noise prior derived from static image to jointly trigger inter-frame relational reasoning and ease the coherent temporal modeling via temporal residual learning. Technically, the image noise prior is first attained through one-step backward dif-fusion process based on both static image and noised video latent codes. Next, TRIP executes a residual-like dual-path scheme for noise prediction: 1) a shortcut path that directly takes image noise prior as the reference noise of each frame to amplify the alignment between the first frame and sub-sequent frames; 2) a residual path that employs 3D-UNet over noised video and static image latent codes to enable inter-frame relational reasoning, thereby easing the learning of the residual noise for each frame. Furthermore, both reference and residual noise of each frame are dynamically merged via attention mechanism for final video generation. Extensive experiments on WebVid-10M, DTDB and MSR-VTT datasets demonstrate the effectiveness of our TRIP for image-to-video generation. Please see our project page at https://trip-i2v.github.io/TRIP/.

The emergence of text-to-image generation models has led to the recognition that image enhancement, performed as post-processing, would significantly improve the visual quality of the generated images. Exploring diffusion models to enhance the generated images nevertheless is not trivial and necessitates to delicately enrich plentiful details while preserving the visual appearance of key content in the original image. In this paper, we propose a novel framework, namely FreeEnhance, for content-consistent image enhancement using the off-the-shelf image diffusion models. Technically, FreeEnhance is a two-stage process that firstly adds random noise to the input image and then capitalizes on a pre-trained image diffusion model (i.e., Latent Diffusion Models) to denoise and enhance the image details. In the noising stage, FreeEnhance is devised to add lighter noise to the region with higher frequency to preserve the high-frequent patterns (e.g., edge, corner) in the original image. In the denoising stage, we present three target properties as constraints to regularize the predicted noise, enhancing images with high acutance and high visual quality. Extensive experiments conducted on the HPDv2 dataset demonstrate that our FreeEnhance outperforms the state-of-the-art image enhancement models in terms of quantitative metrics and human preference. More remarkably, FreeEnhance also shows higher human preference compared to the commercial image enhancement solution of Magnific AI.

How to model the effect of reflection is crucial for single image reflection removal (SIRR) task. Modern SIRR methods usually simplify the reflection formulation with the assumption of linear combination of a transmission layer and a reflection layer. However, the large variations in image content and the real-world picture-taking conditions often result in far more complex reflection. In this paper, we introduce a new screen-blur combination based on two important factors, namely the intensity and the blurriness of reflection, to better characterize the reflection formulation in SIRR. Specifically, we present Screen-blur Reflection Networks (SRNet), which executes the screen-blur formulation in its network design and adapts to the complex reflection on real scenes. Technically, SRNet consists of three components: a blended image generator, a reflection estimator and a reflection removal module. The image generator exploits the screen-blur combination to synthesize the training blended images. The reflection estimator learns the reflection layer and a blur degree that measures the level of blurriness for reflection. The reflection removal module further uses the blended image, blur degree and reflection layer to filter out the transmission layer in a cascaded manner. Superior results on three different SIRR methods are reported when generating the training data on the principle of the screen-blur combination. Moreover, extensive experiments on six datasets quantitatively and qualitatively demonstrate the efficacy of SRNet over the state-of-the-art methods.

By formulating the data generation as a sequence procedure of denoising autoencoding, diffusion models have achieved superior in-painting performance on image data and beyond. Nevertheless, it is not trivial when capitalizing on diffusion models to generate missing 3D points. The difficulty originates from the intrinsic structure where 3D point cloud is a set of unordered and irregular coordinates. That motivates us to delve into the 3D structural information for designing point cloud encoder-decoder and shape latent generator, to precisely formulate the latent distribution of the complete point cloud and partial observation. In this paper, we propose Point cloud completion with Latent Diffusion Models (PointLDM), a new approach that leverages the conditional denoising diffusion probabilistic modeling (DDPM) in the 3D latent space for shape reconstruction. The architecture of PointLDM consists of a transformer-based variational auto-encoder (VAE) to model the complete shape latent, and a diffusion network for shape latent prediction. The encoder of VAE exploits both of global shape latent and local point features in shape distribution learning. With the learnt shape latent, the decoder first decodes the shape latent into coarse points, and then recovers the fine-grained details around each coarse point by deforming a 2D grid. To reconstruct the shape latent from partial observation, the diffusion network treats the partial observation as the conditional input and generates the shape latent via DDPM. Extensive experiments conducted on MVP, Completion3D, and KITTI quantitatively and qualitatively demonstrate the efficacy of PointLDM over the state-of-the-art shape completion approaches.

Temporal dynamics represent the evolving of video content over time, which are critical for action recognition. In this paper, we ask the question: can the off-the-shelf image transformer architecture learn temporal dynamics in videos? To this end, we propose Multidimensional Stacked Image (MSImage) as a new arrangement of video data, which can be fed to image transformers. Technically, MSImage is a high-resolution image that is composed of several evenly-sampled video clips stacked along the channel and space dimensions. The frames in each clip are concatenated along the channel dimension for the transformers to infer short-term dynamics. Meanwhile, the clips are arranged on different spatial positions for learning long-term dynamics. On this basis we propose MSImageFormer-a new variant of image transformer that takes MSImage as the input and is jointly optimized by video classification loss and a new dynamics enhancement loss. The network optimization attends to the high-frequency component of MSImage, avoiding overfitting to static visual patterns. We empirically demonstrate the merits of the MSImageFormer on six action recognition benchmarks. With only 2D image transformer as the classifier, our MSImageFormer achieves 85.3% and 69.7% top-1 accuracy on Kinetics-400 and Something-Something V2 datasets, respectively. Despite with fewer computations, the results are comparable to the SOTA 3D CNNs and video transformers.

The recent advances in deep learning predominantly construct models in their internal representations, and it is opaque to explain the rationale behind and decisions to human users. Such explainability is especially essential for domain adaptation, whose challenges require developing more adaptive models across different domains. In this article, we ask the question: How much does each sample in the source domain contribute to the network’s prediction on the samples from the target domain? To address this, we devise a novel Interpretable Deep Classifier (IDC) that learns the nearest source samples of a target sample as evidence upon which the classifier makes the decision. Technically, IDC maintains a differentiable memory bank for each category, and the memory slot derives a form of key–value pair. The key records the features of discriminative source samples, and the value stores the corresponding properties, e.g., representative scores of the features for describing the category. IDC computes the loss between the output of IDC and the labels of source samples to back-propagate to adjust the representative scores and update the memory banks. Extensive experiments on Office-Home and VisDA-2017 datasets demonstrate that our IDC leads to a more explainable model with almost no accuracy degradation and effectively calibrates classification for optimum reject options. More remarkably, when taking IDC as a prior interpreter, capitalizing on 0.1% source training data selected by IDC still yields superior results than that uses full training set on VisDA-2017 for unsupervised domain adaptation.

Point cloud completion aims to recover the completed 3D shape of an object from its partial observation. A common strategy is to encode the observed points to a global feature vector and then predict the complete points through a generative process on this vector. Nevertheless, the results may suffer from the high-quality shape generation problem due to the fact that a global feature vector cannot sufficiently characterize diverse patterns in one object. In this paper, we present a new shape completion architecture, namely AnchorFormer, that innovatively leverages pattern-aware discriminative nodes, i.e., anchors, to dynamically capture regional information of objects. Technically, AnchorFormer models the regional discrimination by learning a set of anchors based on the point features of the input partial observation. Such anchors are scattered to both observed and unobserved locations through estimating particular offsets, and form sparse points together with the down-sampled points of the input observation. To reconstruct the finegrained object patterns, AnchorFormer further employs a modulation scheme to morph a canonical 2D grid at individual locations of the sparse points into a detailed 3D structure. Extensive experiments on the PCN, ShapeNet-55/34 and KITTI datasets quantitatively and qualitatively demonstrate the efficacy of AnchorFormer over the state-of-the-art point cloud completion approaches. Source code is available at https://github.com/chenzhik/AnchorFormer.

Generalized few-shot semantic segmentation (GFSS) distinguishes pixels of base and novel classes from the background simultaneously, conditioning on sufficient data of base classes and a few examples from novel class. A typical GFSS approach has two training phases: base class learning and novel class updating. Nevertheless, such a stand-alone updating process often compromises the well-learnt features and results in performance drop on base classes. In this paper, we propose a new idea of leveraging Projection onto Orthogonal Prototypes (POP), which updates features to identify novel classes without compromising base classes. POP builds a set of orthogonal prototypes, each of which represents a semantic class, and makes the prediction for each class separately based on the features projected onto its prototype. Technically, POP first learns prototypes on base data, and then extends the prototype set to novel classes. The orthogonal constraint of POP encourages the orthogonality between the learnt prototypes and thus mitigates the influence on base class features when generalizing to novel prototypes. Moreover, we capitalize on the residual of feature projection as the background representation to dynamically fit semantic shifting (i.e., background no longer includes the pixels of novel classes in updating phase). Extensive experiments on two benchmarks demonstrate that our POP achieves superior performances on novel classes without sacrificing much accuracy on base classes. Notably, POP outperforms the state-of-the-art fine-tuning by 3.93% overall mIoU on PASCAL-5iin 5-shot scenario.

Feature invariance under different data transformations, i.e., transformation invariance, can be regarded as a type of self-supervision for representation learning. In this paper, we present PointClustering, a new unsupervised representation learning scheme that leverages transformation invariance for point cloud pre-training. PointClustering formulates the pretext task as deep clustering and employs transformation invariance as an inductive bias, following the philosophy that common point cloud transformation will not change the geometric properties and semantics. Technically, PointClustering iteratively optimizes the feature clusters and backbone, and delves into the transformation invariance as learning regularization from two perspectives: point level and instance level. Point-level invariance learning maintains local geometric properties through gathering point features of one instance across transformations, while instance-level invariance learning further measures clusters over the entire dataset to explore semantics of instances. Our PointClustering is architecture-agnostic and readily applicable to MLP-based, CNN-based and Transformer-based backbones. We empirically demonstrate that the models pre-learnt on the ScanNet dataset by PointClustering provide superior performances on six benchmark-s, across downstream tasks of classification and segmentation. More remarkably, PoinrClustering achieves an accuracy of 94.5% on ModelNet40 with Transformer backbone. Source code is available at https://github.com/FuchenUSTC/PointClustering.

Recent transformer-based solutions have shown great success in 3D human pose estimation. Nevertheless, to calculate the joint-to-joint affinity matrix, the computational cost has a quadratic growth with the increasing number of joints. Such drawback becomes even worse especially for pose estimation in a video sequence, which necessitates spatio-temporal correlation spanning over the entire video. In this paper, we facilitate the issue by decomposing correlation learning into space and time, and present a novel Spatio-Temporal Criss-cross attention (STC) block. Technically, STC first slices its input feature into two partitions evenly along the channel dimension, followed by performing spatial and temporal attention respectively on each partition. STC then models the interactions between joints in an identical frame and joints in an identical trajectory simultaneously by concatenating the outputs from attention layers. On this basis, we devise STCFormer by stacking multiple STC blocks and further integrate a new Structure-enhanced Positional Embedding (SPE) into STCFormer to take the structure of human body into consideration. The embedding function consists of two components: spatio-temporal convolution around neighboring joints to capture local structure, and part-aware embedding to indicate which part each joint belongs to. Extensive experiments are conducted on Human3.6M and MPI-INF-3DHP benchmarks, and superior results are reported when comparing to the state-of-the-art approaches. More remarkably, STCFormer achieves to-date the best published performance: 40.5mm P1 error on the challenging Human3.6M dataset.

Recent progress on multi-view 3D object reconstruction has featured neural implicit surfaces via learning high-fidelity radiance fields. However, most approaches hinge on the visual hull derived from cost-expensive silhouette masks to obtain object surfaces. In this paper, we propose a novel Object-aware Radiance Fields (ORF) to automatically learn an object-aware geometry reconstruction. The geometric correspondences between multi-view 2D object regions and 3D implicit/explicit object surfaces are additionally exploited to boost the learning of object surfaces. Technically, a critical transparency discriminator is designed to distinguish the object-intersected and object-bypassed rays based on the estimated 2D object regions, leading to 3D implicit object surfaces. Such implicit surfaces can be directly converted into explicit object surfaces (e.g., meshes) via marching cubes. Then, we build the geometric correspondence between 2D planes and 3D meshes by rasterization, and project the estimated object regions into 3D explicit object surfaces by aggregating the object information across multiple views. The aggregated object information in 3D explicit object surfaces is further reprojected back to 2D planes, aiming to update 2D object regions and enforce them to be multi-view consistent. Extensive experiments on DTU and BlendedMVS verify the capability of ORF to produce comparable surfaces against the state-of-the-art models that demand silhouette masks.

Referring image segmentation aims to segment the target object described by a natural-language utterance. Recent approaches typically distinguish pixels by aligning pixel-wise visual features with linguistic features extracted from the referring description. Nevertheless, such a free-form description only specifies certain discriminative attributes of the target object or its relations to a limited number of objects, which fails to represent the rich visual context adequately. The stand-alone linguistic features are therefore unable to align with all visual concepts, resulting in inaccurate segmentation. In this paper, we propose to address this issue by incorporating rich visual context into linguistic features for sufficient vision-language alignment. Specifically, we present Context-Aware Referring Image Segmentation (CARIS), a novel architecture that enhances the contextual awareness of linguistic features via sequential vision-language attention and learnable prompts. Technically, CARIS develops a context-aware mask decoder with sequential bidirectional cross-modal attention to integrate the linguistic features with visual context, which are then aligned with pixel-wise visual features. Furthermore, two groups of learnable prompts are employed to delve into additional contextual information from the input image and facilitate the alignment with non-target pixels, respectively. Extensive experiments demonstrate that CARIS achieves new state-of-the-art performances on three public benchmarks. Code is available at https://github.com/lsa1997/CARIS.

Semi-supervised video object segmentation is the task of segmenting the target in sequential frames given the ground truth mask in the first frame. The modern approaches usually utilize such a mask as pixel-level supervision and typically exploit pixel-to-pixel matching between the reference frame and current frame. However, the matching at pixel level, which overlooks the high-level information beyond local areas, often suffers from confusion caused by similar local appearances. In this paper, we present Prototypical Matching Networks (PMNet) - a novel architecture that integrates prototypes into matching-based video objection segmentation frameworks as high-level supervision. Specifically, PMNet first divides the foreground and background areas into several parts according to the similarity to the global prototypes. The part-level prototypes and instance-level prototypes are generated by encapsulating the semantic information of identical parts and identical instances, respectively. To model the correlation between prototypes, the prototype representations are propagated to each other by reasoning on a graph structure. Then, PMNet stores both the pixel-level features and prototypes in the memory bank as the target cues. Three affinities, i.e., pixel-to-pixel affinity, prototype-to-pixel affinity, and prototype-to-prototype affinity, are derived to measure the similarity between the query frame and the features in the memory bank. The features aggregated from the memory bank using these affinities provide powerful discrimination from both the pixel-level and prototype-level perspectives. Extensive experiments conducted on four benchmarks demonstrate superior results than the state-of-the-art video object segmentation techniques.

Motion, as the uniqueness of a video, has been critical to the development of video understanding models. Modern deep learning models leverage motion by either executing spatio-temporal 3D convolutions, factorizing 3D convolutions into spatial and temporal convolutions separately, or computing self-attention along temporal dimension. The implicit assumption behind such successes is that the feature maps across consecutive frames can be nicely aggregated. Nevertheless, the assumption may not always hold especially for the regions with large deformation. In this paper, we present a new recipe of inter-frame attention block, namely Stand-alone Inter-Frame Attention (SIFA), that novelly delves into the deformation across frames to estimate local self-attention on each spatial location. Technically, SIFA remoulds the deformable design via re-scaling the offset predictions by the difference between two frames. Taking each spatial location in the current frame as the query, the locally deformable neighbors in the next frame are regarded as the keys/values. Then, SIFA measures the similarity between query and keys as stand-alone attention to weighted average the values for temporal aggregation. We further plug SIFA block into ConvNets and Vision Transformer, respectively, to devise SIFA-Net and SIFA-Transformer. Extensive experiments conducted on four video datasets demonstrate the superiority of SIFA-Net and SIFA-Transformer as stronger backbones. More remarkably, SIFA-Transformer achieves an accuracy of 83.1% on Kinetics-400 dataset. Source code is available at https://github.com/FuchenUSTC/SIFA.

Convolutional Neural Networks (CNNs) have been re-garded as the go-to models for visual recognition. More re-cently, convolution-free networks, based on multi-head self-attention (MSA) or multi-layer perceptrons (MLPs), become more and more popular. Nevertheless, it is not trivial when utilizing these newly-minted networks for video recognition due to the large variations and complexities in video data. In this paper, we present MLP-3D networks, a novel MLP-like 3D architecture for video recognition. Specifically, the architecture consists of MLP-3D blocks, where each block contains one MLP applied across tokens (i.e., token-mixing MLP) and one MLP applied independently to each token (i.e., channel MLP). By deriving the novel grouped time mixing (GTM) operations, we equip the basic token-mixing MLP with the ability of temporal modeling. GTM divides the input tokens into several temporal groups and linearly maps the tokens in each group with the shared projection matrix. Furthermore, we devise several variants of GTM with different grouping strategies, and compose each vari-ant in different blocks of MLP-3D network by greedy ar-chitecture search. Without the dependence on convolutions or attention mechanisms, our MLP-3D networks achieves 68.5%/81.4% top-1 accuracy on Something-Something V2 and Kinetics-400 datasets, respectively. Despite with fewer computations, the results are comparable to state-of-the-art widely-used 3D CNNs and video transformers.

A steady momentum of innovations and breakthroughs has convincingly pushed the limits of unsupervised image representation learning. Compared to static 2D images, video has one more dimension (time). The inherent supervision existing in such sequential structure offers a fertile ground for building unsupervised learning models. In this paper, we compose a trilogy of exploring the basic and generic supervision in the sequence from spatial, spatiotemporal and sequential perspectives. We materialize the supervisory signals through determining whether a pair of samples is from one frame or from one video, and whether a triplet of samples is in the correct temporal order. We uniquely regard the signals as the foundation in contrastive learning and derive a particular form named Sequence Contrastive Learning (SeCo). SeCo shows superior results under the linear protocol on action recognition (Kinetics), untrimmed activity recognition (ActivityNet) and object tracking (OTB-100). More remarkably, SeCo demonstrates considerable improvements over recent unsupervised pre-training techniques, and leads the accuracy by 2.96% and 6.47% against fully-supervised ImageNet pre-training in action recognition task on UCF101 and HMDB51, respectively. Source code is available at https://github.com/YihengZhang-CV/SeCo-Sequence-Contrastive-Learning.

Human actions are typically of combinatorial structures or patterns, i.e., subjects, objects, plus spatio-temporal interactions in between. Discovering such structures is therefore a rewarding way to reason about the dynamics of interactions and recognize the actions. In this paper, we introduce a new design of sub-graphs to represent and encode the discriminative patterns of each action in the videos. Specifically, we present MUlti-scale Sub-graph LEarning (MUSLE) framework that novelly builds space-time graphs and clusters the graphs into compact sub-graphs on each scale with respect to the number of nodes. Technically, MUSLE produces 3D bounding boxes, i.e., tubelets, in each video clip, as graph nodes and takes dense connectivity as graph edges between tubelets. For each action category, we execute online clustering to decompose the graph into sub-graphs on each scale through learning Gaussian Mixture Layer and select the discriminative sub-graphs as action prototypes for recognition. Extensive experiments are conducted on both Something-Something V1 & V2 and Kinetics-400 datasets, and superior results are reported when comparing to state-of-the-art methods. More remarkably, our MUSLE achieves to-date the best reported accuracy of 65.0% on Something-Something V2 validation set.

Video content is multifaceted, consisting of objects, scenes, interactions or actions. The existing datasets mostly label only one of the facets for model training, resulting in the video representation that biases to only one facet depending on the training dataset. There is no study yet on how to learn a video representation from multifaceted labels, and whether multifaceted information is helpful for video representation learning. In this paper, we propose a new learning framework, MUlti-Faceted Integration (MUFI), to aggregate facets from different datasets for learning a representation that could reflect the full spectrum of video content. Technically, MUFI formulates the problem as visual-semantic embedding learning, which explicitly maps video representation into a rich semantic embedding space, and jointly optimizes video representation from two perspectives. One is to capitalize on the intra-facet supervision between each video and its own label descriptions, and the second predicts the "semantic representation" of each video from the facets of other datasets as the inter-facet supervision. Extensive experiments demonstrate that learning 3D CNN via our MUFI framework on a union of four large-scale video datasets plus two image datasets leads to superior capability of video representation. The prelearnt 3D CNN with MUFI also shows clear improvements over other approaches on several downstream video applications. More remarkably, MUFI achieves 98.1%/80.9% on UCF101/HMDB51 for action recognition and 101.5% in terms of CIDEr-D score on MSVD for video captioning.

Motion, as the most distinct phenomenon in a video to involve the changes over time, has been unique and critical to the development of video representation learning. In this paper, we ask the question: how important is the motion particularly for self-supervised video representation learning. To this end, we compose a duet of exploiting the motion for data augmentation and feature learning in the regime of contrastive learning. Specifically, we present a Motion-focused Contrastive Learning (MCL) method that regards such duet as the foundation. On one hand, MCL capitalizes on optical flow of each frame in a video to temporally and spatially sample the tubelets (i.e., sequences of associated frame patches across time) as data augmentations. On the other hand, MCL further aligns gradient maps of the convolutional layers to optical flow maps from spatial, temporal and spatio-temporal perspectives, in order to ground motion information in feature learning. Extensive experiments conducted on R(2+1)D backbone demonstrate the effectiveness of our MCL. On UCF101, the linear classifier trained on the representations learnt by MCL achieves 81.91% top-1 accuracy, outperforming ImageNet supervised pre-training by 6.78%. On Kinetics-400, MCL achieves 66.62% top-1 accuracy under the linear protocol.

Video is complex due to large variations in motion and rich content in fine-grained visual details. Abstracting useful information from such information-intensive media requires exhaustive computing resources. This paper studies a two-step alternative that first condenses the video sequence to an informative "frame" and then exploits off-the-shelf image recognition system on the synthetic frame. A valid question is how to define "useful information" and then distill it from a video sequence down to one synthetic frame. This paper presents a novel Informative Frame Synthesis (IFS) architecture that incorporates three objective tasks, i.e., appearance reconstruction, video categorization, motion estimation, and two regularizers, i.e., adversarial learning, color consistency. Each task equips the synthetic frame with one ability, while each regularizer enhances its visual quality. With these, by jointly learning the frame synthesis in an end-to-end manner, the generated frame is expected to encapsulate the required spatio-temporal information useful for video analysis. Extensive experiments are conducted on the large-scale Kinetics dataset. When comparing to baseline methods that map video sequence to a single image, IFS shows superior performance. More remarkably, IFS consistently demonstrates evident improvements on image-based 2D networks and clip-based 3D networks, and achieves comparable performance with the state-of-the-art methods with less computational cost.

It is not trivial to optimally learn a 3D Convolutional Neural Networks (3D ConvNets) due to high complexity and various options of the training scheme. The most common hand-tuning process starts from learning 3D ConvNets using short video clips and then is followed by learning long-term temporal dependency using lengthy clips, while gradually decaying the learning rate from high to low as training progresses. The fact that such process comes along with several heuristic settings motivates the study to seek an optimal "path" to automate the entire training. In this paper, we decompose the path into a series of training "states" and specify the hyper-parameters, e.g., learning rate and the length of input clips, in each state. The estimation of the knee point on the performance-epoch curve triggers the transition from one state to another. We perform dynamic programming over all the candidate states to plan the optimal permutation of states, i.e., optimization path. Furthermore, we devise a new 3D ConvNets with a unique design of dual-head classifier to improve spatial and temporal discrimination. Extensive experiments on seven public video recognition benchmarks demonstrate the advantages of our proposal. With the optimization planning, our 3D ConvNets achieves superior results when comparing to the state-of-the-art recognition methods. More remarkably, we obtain the top-1 accuracy of 80.5% and 82.7% on Kinetics-400 and Kinetics-600 datasets, respectively.

Semantic segmentation often requires a large set of images with pixel-level annotations. In the view of extremely expensive expert labeling, recent research has shown that the models trained on photo-realistic synthetic data (e.g., computer games) with computer-generated annotations can be adapted to real images. Despite this progress, without constraining the prediction on real images, the models will easily overfit on synthetic data due to severe domain mismatch. In this paper, we novelly exploit the intrinsic properties of semantic segmentation to alleviate such problem for model transfer. Specifically, we present a Regularizer of Prediction Transfer (RPT) that imposes the intrinsic properties as constraints to regularize model transfer in an unsupervised fashion. These constraints include patch-level, cluster-level and context-level semantic prediction consistencies at different levels of image formation. As the transfer is label-free and data-driven, the robustness of prediction is addressed by selectively involving a subset of image regions for model regularization. Extensive experiments are conducted to verify the proposal of RPT on the transfer of models trained on GTA5 and SYNTHIA (synthetic data) to Cityscapes dataset (urban street scenes). RPT shows consistent improvements when injecting the constraints on several neural networks for semantic segmentation. More remarkably, when integrating RPT into the adversarial-based segmentation framework, we report to-date the best results: mIoU of 53.2%/51.7% when transferring from GTA5/SYNTHIA to Cityscapes, respectively.

Localizing temporal action proposals from long videos is a fundamental challenge in video analysis (e.g., action detection and recognition or dense video captioning). Most existing approaches often overlook the hierarchical granularities of actions and thus fail to discriminate fine-grained action proposals (e.g., hand washing laundry or changing a tire in vehicle repair). In this paper, we propose a novel coarse-to-fine temporal proposal (CFTP) approach to localize temporal action proposals by exploring different action granularities. Our proposed CFTP consists of three stages: a coarse proposal network (CPN) to generate long action proposals, a temporal convolutional anchor network (CAN) to localize finer proposals, and a proposal reranking network (PRN) to further identify proposals from previous stages. Specifically, CPN explores three complementary actionness curves (namely pointwise, pairwise, and recurrent curves) that represent actions at different levels for generating coarse proposals, while CAN refines these proposals by a multiscale cascaded 1D-convolutional anchor network. In contrast to existing works, our coarse-to-fine approach can progressively localize fine-grained action proposals. We conduct extensive experiments on two action benchmarks (THUMOS14 and ActivityNet v1.3) and demonstrate the superior performance of our approach when compared to the state-of-the-art techniques on various video understanding tasks.

Temporally localizing actions in a video is a fundamental challenge in video understanding. Most existing approaches have often drawn inspiration from image object detection and extended the advances, e.g., SSD and Faster R-CNN, to produce temporal locations of an action in a 1D sequence. Nevertheless, the results can suffer from robustness problem due to the design of predetermined temporal scales, which overlooks the temporal structure of an action and limits the utility on detecting actions with complex variations. In this paper, we propose to address the problem by introducing Gaussian kernels to dynamically optimize temporal scale of each action proposal. Specifically, we present Gaussian Temporal Awareness Networks (GTAN) - a new architecture that novelly integrates the exploitation of temporal structure into an one-stage action localization framework. Technically, GTAN models the temporal structure through learning a set of Gaussian kernels, each for a cell in the feature maps. Each Gaussian kernel corresponds to a particular interval of an action proposal and a mixture of Gaussian kernels could further characterize action proposals with various length. Moreover, the values in each Gaussian curve reflect the contextual contributions to the localization of an action proposal. Extensive experiments are conducted on both THUMOS14 and ActivityNet v1.3 datasets, and superior results are reported when comparing to state-of-the-art approaches. More remarkably, GTAN achieves 1.9% and 1.1% improvements in mAP on testing set of the two datasets.

Convolutional Neural Networks (CNN) have been regarded as a powerful class of models for visual recognition problems. Nevertheless, the convolutional filters in these networks are local operations while ignoring the large-range dependency. Such drawback becomes even worse particularly for video recognition, since video is an information-intensive media with complex temporal variations. In this paper, we present a novel framework to boost the spatio-temporal representation learning by Local and Global Diffusion (LGD). Specifically, we construct a novel neural network architecture that learns the local and global representations in parallel. The architecture is composed of LGD blocks, where each block updates local and global features by modeling the diffusions between these two representations. Diffusions effectively interact two aspects of information, i.e., localized and holistic, for more powerful way of representation learning. Furthermore, a kernelized classifier is introduced to combine the representations from two aspects for video recognition. Our LGD networks achieve clear improvements on the large-scale Kinetics-400 and Kinetics-600 video classification datasets against the best competitors by 3.5% and 0.7%. We further examine the generalization of both the global and local representations produced by our pre-trained LGD networks on four different benchmarks for video action recognition and spatio-temporal action detection tasks. Superior performances over several state-of-the-art techniques on these benchmarks are reported.

In this paper, we propose a Customizable Architecture Search (CAS) approach to automatically generate a network architecture for semantic image segmentation. The generated network consists of a sequence of stacked computation cells. A computation cell is represented as a directed acyclic graph, in which each node is a hidden representation (i.e., feature map) and each edge is associated with an operation (e.g., convolution and pooling), which transforms data to a new layer. During the training, the CAS algorithm explores the search space for an optimized computation cell to build a network. The cells of the same type share one architecture but with different weights. In real applications, however, an optimization may need to be conducted under some constraints such as GPU time and model size. To this end, a cost corresponding to the constraint will be assigned to each operation. When an operation is selected during the search, its associated cost will be added to the objective. As a result, our CAS is able to search an optimized architecture with customized constraints. The approach has been thoroughly evaluated on Cityscapes and CamVid datasets, and demonstrates superior performance over several state-of-the-art techniques. More remarkably, our CAS achieves 72.3% mIoU on the Cityscapes dataset with speed of 108 FPS on an Nvidia TitanXp GPU.

It has been well recognized that modeling human-object or object-object relations would be helpful for detection task. Nevertheless, the problem is not trivial especially when exploring the interactions between human actor, object and scene (collectively as human-context) to boost video action detectors. The difficulty originates from the aspect that reliable relations in a video should depend on not only short-term human-context relation in the present clip but also the temporal dynamics distilled over a long-range span of the video. This motivates us to capture both short-term and long-term relations in a video. In this paper, we present a new Long Short-Term Relation Networks, dubbed as LSTR, that novelly aggregates and propagates relation to augment features for video action detection. Technically, Region Proposal Networks (RPN) is remoulded to first produce 3D bounding boxes, i.e., tubelets, in each video clip. LSTR then models short-term human-context interactions within each clip through spatio-temporal attention mechanism and reasons long-term temporal dynamics across video clips via Graph Convolutional Networks (GCN) in a cascaded manner. Extensive experiments are conducted on four benchmark datasets, and superior results are reported when comparing to state-of-the-art methods.

The recent advances in deep neural networks have convincingly demonstrated high capability in learning vision models on large datasets. Nevertheless, collecting expert labeled datasets especially with pixel-level annotations is an extremely expensive process. An appealing alternative is to render synthetic data (e.g., computer games) and generate ground truth automatically. However, simply applying the models learnt on synthetic images may lead to high generalization error on real images due to domain shift. In this paper, we facilitate this issue from the perspectives of both visual appearance-level and representation-level domain adaptation. The former adapts source-domain images to appear as if drawn from the "style" in the target domain and the latter attempts to learn domain-invariant representations. Specifically, we present Fully Convolutional Adaptation Networks (FCAN), a novel deep architecture for semantic segmentation which combines Appearance Adaptation Networks (AAN) and Representation Adaptation Networks (RAN). AAN learns a transformation from one domain to the other in the pixel space and RAN is optimized in an adversarial learning manner to maximally fool the domain discriminator with the learnt source and target representations. Extensive experiments are conducted on the transfer from GTA5 (game videos) to Cityscapes (urban street scenes) on semantic segmentation and our proposal achieves superior results when comparing to state-of-the-art unsupervised adaptation techniques. More remarkably, we obtain a new record: mIoU of 47.5% on BDDS (drive-cam videos) in an unsupervised setting.

Semantically labeling every pixel in a video is a very challenging task as video is an information-intensive media with complex spatio-temporal dependence. We present in this paper a novel deep convolutional network architecture, called deep spatio-temporal fully convolutional networks (DST-FCN), which leverages both spatial and temporal dependencies among pixels and voxels by training them in an end-to-end manner. Specifically, we introduce a two-stream network by learning the deep spatio-temporal dependence, in which a 2D FCN followed by the convolutional long short-term memory (ConvLSTM) is employed on the pixel level and a 3-D FCN is exploited on the voxel level. Our model differs from conventional FCN in that it not only extends FCN by adding ConvLSTM on the pixel level for exploring long-term dependence, but also proposes 3-D FCN to enable voxel level prediction. On two benchmarks of A2D and CamVid, our DST-FCN achieves superior results to state-of-the-art techniques. More remarkably, we obtain to-date the best reported results: 45.0% per-label accuracy on A2D and 68.8% mean IoU on CamVid.

Convolutional Neural Networks (CNN) have been regarded as a powerful class of models for image recognition problems. Nevertheless, it is not trivial when utilizing a CNN for learning spatio-temporal video representation. A few studies have shown that performing 3D convolutions is a rewarding approach to capture both spatial and temporal dimensions in videos. However, the development of a very deep 3D CNN from scratch results in expensive computational cost and memory demand. A valid question is why not recycle off-the-shelf 2D networks for a 3D CNN. In this paper, we devise multiple variants of bottleneck building blocks in a residual learning framework by simulating 3 x 3 x 3 convolutions with 1 × 3 × 3 convolutional filters on spatial domain (equivalent to 2D CNN) plus 3 × 1 × 1 convolutions to construct temporal connections on adjacent feature maps in time. Furthermore, we propose a new architecture, named Pseudo-3D Residual Net (P3D ResNet), that exploits all the variants of blocks but composes each in different placement of ResNet, following the philosophy that enhancing structural diversity with going deep could improve the power of neural networks. Our P3D ResNet achieves clear improvements on Sports-1M video classification dataset against 3D CNN and frame-based 2D CNN by 5.3% and 1.8%, respectively. We further examine the generalization performance of video representation produced by our pre-trained P3D ResNet on five different benchmarks and three different tasks, demonstrating superior performances over several state-of-the-art techniques.

Automatically describing an image with a natural language has been an emerging challenge in both fields of computer vision and natural language processing. In this paper, we present Long Short-Term Memory with Attributes (LSTM-A) - a novel architecture that integrates attributes into the successful Convolutional Neural Networks (CNNs) plus Recurrent Neural Networks (RNNs) image captioning framework, by training them in an end-to-end manner. Particularly, the learning of attributes is strengthened by integrating inter-attribute correlations into Multiple Instance Learning (MIL). To incorporate attributes into captioning, we construct variants of architectures by feeding image representations and attributes into RNNs in different ways to explore the mutual but also fuzzy relationship between them. Extensive experiments are conducted on COCO image captioning dataset and our framework shows clear improvements when compared to state-of-the-art deep models. More remarkably, we obtain METEOR/CIDEr-D of 25.5%/100.2% on testing data of widely used and publicly available splits in [10] when extracting image representations by GoogleNet and achieve superior performance on COCO captioning Leaderboard.

We are creating multimedia contents everyday and everywhere. While automatic content generation has played a fundamental challenge to multimedia community for decades, recent advances of deep learning have made this problem feasible. For example, the Generative Adversarial Networks (GANs) is a rewarding approach to synthesize images. Nevertheless, it is not trivial when capitalizing on GANs to generate videos. The difficulty originates from the intrinsic structure where a video is a sequence of visually coherent and semantically dependent frames. This motivates us to explore semantic and temporal coherence in designing GANs to generate videos. In this paper, we present a novel Temporal GANs conditioning on Captions, namely TGANs-C, in which the input to the generator network is a concatenation of a latent noise vector and caption embedding, and then is transformed into a frame sequence with 3D spatio-temporal convolutions. Unlike the naive discriminator which only judges pairs as fake or real, our discriminator additionally notes whether the video matches the correct caption. In particular, the discriminator network consists of three discriminators: video discriminator classifying realistic videos from generated ones and optimizes video-caption matching, frame discriminator discriminating between real and fake frames and aligning frames with the conditioning caption, and motion discriminator emphasizing the philosophy that the adjacent frames in the generated videos should be smoothly connected as in real ones. We qualitatively demonstrate the capability of our TGANs-C to generate plausible videos conditioning on the given captions on two synthetic datasets (SBMG and TBMG) and one real-world dataset (MSVD). Moreover, quantitative experiments on MSVD are performed to validate our proposal via Generative Adversarial Metric and human study.

We demonstrate a video captioning bot, named Seeing Bot, which can generate a natural language description about what it is seeing in near real time. Specifically, given a live streaming video, Seeing Bot runs two pre-learned and complementary captioning modules in parallel - one for generating image-level caption for each sampled frame, and the other for generating video-level caption for each sampled video clip. In particular, both the image and video captioning modules are boosted by incorporating semantic attributes which can enrich the generated descriptions, leading to human-level caption generation. A visual-semantic embedding model is then exploited to rank and select the final caption from the two parallel modules by considering the semantic relevance between video content and the generated captions. The Seeing Bot finally converts the generated description to speech and sends the speech to an end user via an earphone. Our demonstration is conducted on any videos in the wild and supports live video captioning.

Hashing has been a widely-adopted technique for nearest neighbor search in large-scale image retrieval tasks. Recent research has shown that leveraging supervised information can lead to high quality hashing. However, the cost of annotating data is often an obstacle when applying supervised hashing to a new domain. Moreover, the results can suffer from the robustness problem as the data at training and test stage may come from different distributions. This paper studies the exploration of generating synthetic data through semi-supervised generative adversarial networks (GANs), which leverages largely unlabeled and limited labeled training data to produce highly compelling data with intrinsic invariance and global coherence, for better understanding statistical structures of natural data. We demonstrate that the above two limitations can be well mitigated by applying the synthetic data for hashing. Specifically, a novel deep semantic hashing with GANs (DSH-GANs) is presented, which mainly consists of four components: a deep convolution neural networks (CNN) for learning image representations, an adversary stream to distinguish synthetic images from real ones, a hash stream for encoding image representations to hash codes and a classification stream. The whole architecture is trained end-to-end by jointly optimizing three losses, i.e., adversarial loss to correct label of synthetic or real for each sample, triplet ranking loss to preserve the relative similarity ordering in the input real-synthetic triplets and classification loss to classify each sample accurately. Extensive experiments conducted on both CIFAR-10 and NUS-WIDE image benchmarks validate the capability of exploiting synthetic images for hashing. Our framework also achieves superior results when compared to state-of-the-art deep hash models.

Recognizing actions in videos is a challenging task as video is an information-intensive media with complex variations. Most existing methods have treated video as a flat data sequence while ignoring the intrinsic hierarchical structure of the video content. In particular, an action may span different granularities in this hierarchy including, from small to large, a single frame, consecutive frames (motion), a short clip, and the entire video. In this paper, we present a novel framework to boost action recognition by learning a deep spatio-temporal video representation at hierarchical multi-granularity. Specifically, we model each granularity as a single stream by 2D (for frame and motion streams) or 3D (for clip and video streams) convolutional neural networks (CNNs). The framework therefore consists of multi-stream 2D or 3D CNNs to learn both the spatial and temporal representations. Furthermore, we employ the Long Short-Term Memory (LSTM) networks on the frame, motion, and clip streams to exploit long-term temporal dynamics. With a softmax layer on the top of each stream, the classification scores can be predicted from all the streams, followed by a novel fusion scheme based on the multi-granular score distribution. Our networks are learned in an end-to-end fashion. On two video action benchmarks of UCF101 and HMDB51, our framework achieves promising performance compared with the state-of-the-art.