Existing research addresses scene graph generation (SGG) -- a critical technology for scene understanding in images -- from a detection perspective, i.e., objects are detected using bounding boxes followed by prediction of their pairwise relationships. We argue that such a paradigm causes several problems that impede the progress of the field. For instance, bounding box-based labels in current datasets usually contain redundant classes like hairs, and leave out background information that is crucial to the understanding of context. In this work, we introduce panoptic scene graph generation (PSG), a new problem task that requires the model to generate a more comprehensive scene graph representation based on panoptic segmentations rather than rigid bounding boxes. A high-quality PSG dataset, which contains 49k well-annotated overlapping images from COCO and Visual Genome, is created for the community to keep track of its progress. For benchmarking, we build four two-stage baselines, which are modified from classic methods in SGG, and two one-stage baselines called PSGTR and PSGFormer, which are based on the efficient Transformer-based detector, i.e., DETR. While PSGTR uses a set of queries to directly learn triplets, PSGFormer separately models the objects and relations in the form of queries from two Transformer decoders, followed by a prompting-like relation-object matching mechanism. In the end, we share insights on open challenges and future directions.
We propose united implicit functions (UNIF), a part-based method for clothed human reconstruction and animation with raw scans and skeletons as the input. Previous part-based methods for human reconstruction rely on ground-truth part labels from SMPL and thus are limited to minimal-clothed humans. In contrast, our method learns to separate parts from body motions instead of part supervision, thus can be extended to clothed humans and other articulated objects. Our Partition-from-Motion is achieved by a bone-centered initialization, a bone limit loss, and a section normal loss that ensure stable part division even when the training poses are limited. We also present a minimal perimeter loss for SDF to suppress extra surfaces and part overlapping. Another core of our method is an adjacent part seaming algorithm that produces non-rigid deformations to maintain the connection between parts which significantly relieves the part-based artifacts. Under this algorithm, we further propose "Competing Parts", a method that defines blending weights by the relative position of a point to bones instead of the absolute position, avoiding the generalization problem of neural implicit functions with inverse LBS (linear blend skinning). We demonstrate the effectiveness of our method by clothed human body reconstruction and animation on the CAPE and the ClothSeq datasets.
Though impressive performance has been achieved in specific visual realms (e.g. faces, dogs, and places), an omni-vision representation generalizing to many natural visual domains is highly desirable. But, existing benchmarks are biased and inefficient to evaluate the omni-vision representation -- these benchmarks either only include several specific realms, or cover most realms at the expense of subsuming numerous datasets that have extensive realm overlapping. In this paper, we propose Omni-Realm Benchmark (OmniBenchmark). It includes 21 realm-wise datasets with 7,372 concepts and 1,074,346 images. Without semantic overlapping, these datasets cover most visual realms comprehensively and meanwhile efficiently. In addition, we propose a new supervised contrastive learning framework, namely Relational Contrastive learning (ReCo), for a better omni-vision representation. Beyond pulling two instances from the same concept closer -- the typical supervised contrastive learning framework -- ReCo also pulls two instances from the same semantic realm closer, encoding the semantic relation between concepts, and facilitating omni-vision representation learning. We benchmark ReCo and other advances in omni-vision representation studies that are different in architectures (from CNNs to transformers) and in learning paradigms (from supervised learning to self-supervised learning) on OmniBenchmark. We illustrate the superior of ReCo to other supervised contrastive learning methods and reveal multiple practical observations to facilitate future research.
Human relighting is a highly desirable yet challenging task. Existing works either require expensive one-light-at-a-time (OLAT) captured data using light stage or cannot freely change the viewpoints of the rendered body. In this work, we propose a principled framework, Relighting4D, that enables free-viewpoints relighting from only human videos under unknown illuminations. Our key insight is that the space-time varying geometry and reflectance of the human body can be decomposed as a set of neural fields of normal, occlusion, diffuse, and specular maps. These neural fields are further integrated into reflectance-aware physically based rendering, where each vertex in the neural field absorbs and reflects the light from the environment. The whole framework can be learned from videos in a self-supervised manner, with physically informed priors designed for regularization. Extensive experiments on both real and synthetic datasets demonstrate that our framework is capable of relighting dynamic human actors with free-viewpoints.
Video-to-Video synthesis (Vid2Vid) has achieved remarkable results in generating a photo-realistic video from a sequence of semantic maps. However, this pipeline suffers from high computational cost and long inference latency, which largely depends on two essential factors: 1) network architecture parameters, 2) sequential data stream. Recently, the parameters of image-based generative models have been significantly compressed via more efficient network architectures. Nevertheless, existing methods mainly focus on slimming network architectures and ignore the size of the sequential data stream. Moreover, due to the lack of temporal coherence, image-based compression is not sufficient for the compression of the video task. In this paper, we present a spatial-temporal compression framework, \textbf{Fast-Vid2Vid}, which focuses on data aspects of generative models. It makes the first attempt at time dimension to reduce computational resources and accelerate inference. Specifically, we compress the input data stream spatially and reduce the temporal redundancy. After the proposed spatial-temporal knowledge distillation, our model can synthesize key-frames using the low-resolution data stream. Finally, Fast-Vid2Vid interpolates intermediate frames by motion compensation with slight latency. On standard benchmarks, Fast-Vid2Vid achieves around real-time performance as 20 FPS and saves around 8x computational cost on a single V100 GPU.
Since photorealistic faces can be readily generated by facial manipulation technologies nowadays, potential malicious abuse of these technologies has drawn great concerns. Numerous deepfake detection methods are thus proposed. However, existing methods only focus on detecting one-step facial manipulation. As the emergence of easy-accessible facial editing applications, people can easily manipulate facial components using multi-step operations in a sequential manner. This new threat requires us to detect a sequence of facial manipulations, which is vital for both detecting deepfake media and recovering original faces afterwards. Motivated by this observation, we emphasize the need and propose a novel research problem called Detecting Sequential DeepFake Manipulation (Seq-DeepFake). Unlike the existing deepfake detection task only demanding a binary label prediction, detecting Seq-DeepFake manipulation requires correctly predicting a sequential vector of facial manipulation operations. To support a large-scale investigation, we construct the first Seq-DeepFake dataset, where face images are manipulated sequentially with corresponding annotations of sequential facial manipulation vectors. Based on this new dataset, we cast detecting Seq-DeepFake manipulation as a specific image-to-sequence (e.g. image captioning) task and propose a concise yet effective Seq-DeepFake Transformer (SeqFakeFormer). Moreover, we build a comprehensive benchmark and set up rigorous evaluation protocols and metrics for this new research problem. Extensive experiments demonstrate the effectiveness of SeqFakeFormer. Several valuable observations are also revealed to facilitate future research in broader deepfake detection problems.
Densely annotating LiDAR point clouds is costly, which restrains the scalability of fully-supervised learning methods. In this work, we study the underexplored semi-supervised learning (SSL) in LiDAR segmentation. Our core idea is to leverage the strong spatial cues of LiDAR point clouds to better exploit unlabeled data. We propose LaserMix to mix laser beams from different LiDAR scans, and then encourage the model to make consistent and confident predictions before and after mixing. Our framework has three appealing properties: 1) Generic: LaserMix is agnostic to LiDAR representations (e.g., range view and voxel), and hence our SSL framework can be universally applied. 2) Statistically grounded: We provide a detailed analysis to theoretically explain the applicability of the proposed framework. 3) Effective: Comprehensive experimental analysis on popular LiDAR segmentation datasets (nuScenes, SemanticKITTI, and ScribbleKITTI) demonstrates our effectiveness and superiority. Notably, we achieve competitive results over fully-supervised counterparts with 2x to 5x fewer labels and improve the supervised-only baseline significantly by 10.8% on average. We hope this concise yet high-performing framework could facilitate future research in semi-supervised LiDAR segmentation. Code will be publicly available.
We present Masked Frequency Modeling (MFM), a unified frequency-domain-based approach for self-supervised pre-training of visual models. Instead of randomly inserting mask tokens to the input embeddings in the spatial domain, in this paper, we shift the perspective to the frequency domain. Specifically, MFM first masks out a portion of frequency components of the input image and then predicts the missing frequencies on the frequency spectrum. Our key insight is that predicting masked components in the frequency domain is more ideal to reveal underlying image patterns rather than predicting masked patches in the spatial domain, due to the heavy spatial redundancy. Our findings suggest that with the right configuration of mask-and-predict strategy, both the structural information within high-frequency components and the low-level statistics among low-frequency counterparts are useful in learning good representations. For the first time, MFM demonstrates that, for both ViT and CNN, a simple non-Siamese framework can learn meaningful representations even using none of the following: (i) extra data, (ii) extra model, (iii) mask token. Experimental results on ImageNet and several robustness benchmarks show the competitive performance and advanced robustness of MFM compared with recent masked image modeling approaches. Furthermore, we also comprehensively investigate the effectiveness of classical image restoration tasks for representation learning from a unified frequency perspective and reveal their intriguing relations with our MFM approach. Project page: https://www.mmlab-ntu.com/project/mfm/index.html.
The size of vision models has grown exponentially over the last few years, especially after the emergence of Vision Transformer. This has motivated the development of parameter-efficient tuning methods, such as learning adapter layers or visual prompt tokens, which allow a tiny portion of model parameters to be trained whereas the vast majority obtained from pre-training are frozen. However, designing a proper tuning method is non-trivial: one might need to try out a lengthy list of design choices, not to mention that each downstream dataset often requires custom designs. In this paper, we view the existing parameter-efficient tuning methods as "prompt modules" and propose Neural prOmpt seArcH (NOAH), a novel approach that learns, for large vision models, the optimal design of prompt modules through a neural architecture search algorithm, specifically for each downstream dataset. By conducting extensive experiments on over 20 vision datasets, we demonstrate that NOAH (i) is superior to individual prompt modules, (ii) has a good few-shot learning ability, and (iii) is domain-generalizable. The code and models are available at https://github.com/Davidzhangyuanhan/NOAH.
Domain generalization (DG) aims at learning generalizable models under distribution shifts to avoid redundantly overfitting massive training data. Previous works with complex loss design and gradient constraint have not yet led to empirical success on large-scale benchmarks. In this work, we reveal the mixture-of-experts (MoE) model's generalizability on DG by leveraging to distributively handle multiple aspects of the predictive features across domains. To this end, we propose Sparse Fusion Mixture-of-Experts (SF-MoE), which incorporates sparsity and fusion mechanisms into the MoE framework to keep the model both sparse and predictive. SF-MoE has two dedicated modules: 1) sparse block and 2) fusion block, which disentangle and aggregate the diverse learned signals of an object, respectively. Extensive experiments demonstrate that SF-MoE is a domain-generalizable learner on large-scale benchmarks. It outperforms state-of-the-art counterparts by more than 2% across 5 large-scale DG datasets (e.g., DomainNet), with the same or even lower computational costs. We further reveal the internal mechanism of SF-MoE from distributed representation perspective (e.g., visual attributes). We hope this framework could facilitate future research to push generalizable object recognition to the real world. Code and models are released at https://github.com/Luodian/SF-MoE-DG.