论文检索

Simone Rossetti, Damiano Zappia, Marta Sanzari, Marco Schaerf, Fiora Pirri

Weakly Supervised Semantic Segmentation (WSSS) research has explored many directions to improve the typical pipeline CNN plus class activation maps (CAM) plus refinements, given the image-class label as the only supervision. Though the gap with the fully supervised methods is reduced, further abating the spread seems unlikely within this framework. On the other hand, WSSS methods based on Vision Transformers (ViT) have not yet explored valid alternatives to CAM. ViT features have been shown to retain a scene layout, and object boundaries in self-supervised learning. To confirm these findings, we prove that the advantages of transformers in self-supervised methods are further strengthened by Global Max Pooling (GMP), which can leverage patch features to negotiate pixel-label probability with class probability. This work proposes a new WSSS method dubbed ViT-PCM (ViT Patch-Class Mapping), not based on CAM. The end-to-end presented network learns with a single optimization process, refined shape and proper localization for segmentation masks. Our model outperforms the state-of-the-art on baseline pseudo-masks (BPM), where we achieve 69.3\% mIoU on PascalVOC 2012 validation set. We show that our approach has the least set of parameters, though obtaining higher accuracy than all other approaches. In a sentence, quantitative and qualitative results of our method reveal that ViT-PCM is an excellent alternative to CNN-CAM based architectures.

Junsong Fan, Zhaoxiang Zhang, Tieniu Tan

In this paper, we propose a new approach to applying point-level annotations for weakly-supervised panoptic segmentation. Instead of the dense pixel-level labels used by fully supervised methods, point-level labels only provide a single point for each target as supervision, significantly reducing the annotation burden. We formulate the problem in an end-to-end framework by simultaneously generating panoptic pseudo-masks from point-level labels and learning from them. To tackle the core challenge, i.e., panoptic pseudo-mask generation, we propose a principled approach to parsing pixels by minimizing pixel-to-point traversing costs, which model semantic similarity, low-level texture cues, and high-level manifold knowledge to discriminate panoptic targets. We conduct experiments on the Pascal VOC and the MS COCO datasets to demonstrate the approach’s effectiveness and show state-of-the-art performance in the weakly-supervised panoptic segmentation problem. Codes are available at https://github.com/BraveGroup/PSPS.git.

Xiaoyi Dong, Jianmin Bao, Ting Zhang, Dongdong Chen, Weiming Zhang, Lu Yuan, Dong Chen, Fang Wen, Nenghai Yu

We propose bootstrapped masked autoencoders (BootMAE), a new approach for vision BERT pretraining. BootMAE improves the original masked autoencoders (MAE) with two core designs: 1) momentum encoder that provides online feature as extra BERT prediction targets; 2) target-aware decoder that tries to reduce the pressure on the encoder to memorize target-specific information in BERT pretraining. The first design is motivated by the observation that using a pretrained MAE to extract the features as the BERT prediction target for masked tokens can achieve better pretraining performance. Therefore, we add a momentum encoder in parallel with the original MAE encoder, which bootstraps the pretraining performance by using its own representation as the BERT prediction target. In the second design, we introduce target-specific information (e.g., pixel values of unmasked patches) from the encoder directly to the decoder to reduce the pressure on the encoder of memorizing the target-specific information. Thus, the encoder focuses on semantic modeling, which is the goal of BERT pretraining, and does not need to waste its capacity in memorizing the information of unmasked tokens related to the prediction target. Through extensive experiments, our BootMAE achieves 84.2% Top-1 accuracy on ImageNet-1K with ViT-B backbone, outperforming MAE by +0.8% under the same pre-training epochs. BootMAE also gets +1.0 mIoU improvements on semantic segmentation on ADE20K and +1.3 box AP, +1.4 mask AP improvement on object detection and segmentation on COCO dataset. Code is released at https://github.com/LightDXY/BootMAE

Mehmet Aygün, Oisin Mac Aodha

We explore semantic correspondence estimation through the lens of unsupervised learning. We thoroughly evaluate several recently proposed unsupervised methods across multiple challenging datasets using a standardized evaluation protocol where we vary factors such as the backbone architecture, the pre-training strategy, and the pre-training and finetuning datasets. To better understand the failure modes of these methods, and in order to provide a clearer path for improvement, we provide a new diagnostic framework along with a new performance metric that is better suited to the semantic matching task. Finally, we introduce a new unsupervised correspondence approach which utilizes the strength of pre-trained features while encouraging better matches during training. This results in significantly better matching performance compared to current state-of-the-art methods.

Tuan Ngo, Khoi Nguyen

This paper introduces a new problem in 3D point cloud: few-shot instance segmentation. Given a few annotated point clouds exemplified a target class, our goal is to segment all instances of this target class in a query point cloud. This problem has a wide range of practical applications where point-wise instance segmentation annotation is prohibitively expensive to collect. To address this problem, we present Geodesic-Former -- the first geodesic-guided transformer for 3D point cloud instance segmentation. The key idea is to leverage the geodesic distance to tackle the density imbalance of LiDAR 3D point clouds. The LiDAR 3D point clouds are dense near the object surface and sparse or empty elsewhere making the Euclidean distance less effective to distinguish different objects. The geodesic distance, on the other hand, is more suitable since it encodes the scene’s geometry which can be used as a guiding signal for the attention mechanism in a transformer decoder to generate kernels representing distinct features of instances. These kernels are then used in a dynamic convolution to obtain the final instance masks. To evaluate Geodesic-Former on the new task, we propose new splits of the two common 3D point cloud instance segmentation datasets: ScannetV2 and S3DIS. Geodesic-Former consistently outperforms strong baselines adapted from state-of-the-art 3D point cloud instance segmentation approaches with a significant margin. Code is available at https://github.com/VinAIResearch/GeoFormer.

Zihan Lin, Zilei Wang, Yixin Zhang

Deep networks have been shown to suffer from catastrophic forgetting. In this work, we try to alleviate this phenomenon in the field of continual semantic segmentation (CSS). We observe that two main problems lie in existing arts. First, attention is only paid to designing constraints for encoder (i.e., the backbone of segmentation network) or output probabilities. But we find that forgetting also happens in the decoder head and harms the performance greatly. Second, old and new knowledge are entangled in intermediate features when learning new categories, making existing practices hard to balance between plasticity and stability. On these bases, we propose a framework driven by two novel constraints to address the aforementioned problems. First, a structure preserving loss is applied to the decoder’s output to maintain the discriminative power of old classes from two different granularities in embedding space. Second, a feature projection module is adopted to disentangle the process of preserving old knowledge from learning new classes. Extensive evaluations on VOC2012 and ADE20K datasets show the effectiveness of our approach, which significantly outperforms existing state-of-the-art CSS methods.

Jiayuan Zhou, Lijun Wang, Huchuan Lu, Kaining Huang, Xinchu Shi, Bocong Liu

RGB-D salient object detection (SOD) enjoys significant advantages in understanding 3D geometry of the scene. However, the geometry information conveyed by depth maps are mostly under-explored in existing RGB-D SOD methods. In this paper, we propose a new framework to address this issue. We augment the input image with multiple different views rendered using the depth maps, and cast the conventional single-view RGB-D SOD into a multi-view setting. Since different views captures complementary context of the 3D scene, the accuracy can be significantly improved through multi-view aggregation. We further design a multi-view saliency detection network (MVSalNet), which firstly performs saliency prediction for each view separately and incorporates multi-view outputs through a fusion model to produce final saliency prediction. A dynamic filtering module is also designed to facilitate more effective and flexible feature extraction. Extensive experiments on 6 widely used datasets demonstrate that our approach compares favorably against state-of-the-art approaches.

Joakim Johnander, Johan Edstedt, Michael Felsberg, Fahad Shahbaz Khan, Martin Danelljan

Few-shot segmentation is a challenging dense prediction task, which entails segmenting a novel query image given only a small annotated support set. The key problem is thus to design a method that aggregates detailed information from the support set, while being robust to large variations in appearance and context. To this end, we propose a few-shot segmentation method based on dense Gaussian process (GP) regression. Given the support set, our dense GP learns the mapping from local deep image features to mask values, capable of capturing complex appearance distributions. Furthermore, it provides a principled means of capturing uncertainty, which serves as another powerful cue for the final segmentation, obtained by a CNN decoder. Instead of a one-dimensional mask output, we further exploit the end-to-end learning capabilities of our approach to learn a high-dimensional output space for the GP. Our approach sets a new state-of-the-art on the PASCAL-5i and COCO-20i benchmarks, achieving an absolute gain of +8.4 mIoU in the COCO-20i 5-shot setting. Furthermore, the segmentation quality of our approach scales gracefully when increasing the support set size, while achieving robust cross-dataset transfer.

Zhixuan Li, Weining Ye, Tingting Jiang, Tiejun Huang

Amodal instance segmentation aims to predict the complete mask of the occluded instance, including both visible and invisible regions. Existing 2D AIS methods learn and predict the complete silhouettes of target instances in 2D space. However, masks in 2D space are only some observations and samples from the 3D model in different viewpoints and thus can not represent the real complete physical shape of the instances. With the 2D masks learned, 2D amodal methods are hard to generalize to new viewpoints not included in the training dataset. To tackle these problems, we are motivated by observations that (1) a 2D amodal mask is the projection of a 3D complete model, and (2) the 3D complete model can be recovered and reconstructed from the occluded 2D object instances. This paper builds a bridge to link the 2D occluded instances with the 3D complete models by 3D reconstruction and utilizes 3D shape prior for 2D AIS. To deal with the diversity of 3D shapes, our method is pretrained on large 3D reconstruction datasets for high-quality results. And we adopt the unsupervised 3D reconstruction method to avoid relying on 3D annotations. In this approach, our method can reconstruct 3D models from occluded 2D object instances and generalize to new unseen 2D viewpoints of the 3D object. Experiments demonstrate that our method outperforms all existing 2D AIS methods. Our code will be released.

Lingzhi Zhang, Yuqian Zhou, Connelly Barnes, Sohrab Amirghodsi, Zhe Lin, Eli Shechtman, Jianbo Shi

Image inpainting is an essential task for multiple practical applications like object removal and image editing. Deep GAN-based models greatly improve the inpainting performance in structures and textures within the hole, but might also generate unexpected artifacts like broken structures or color blobs. Users perceive these artifacts to judge the effectiveness of inpainting models, and retouch these imperfect areas to inpaint again in a typical retouching workflow. Inspired by this workflow, we propose a new learning task of automatic segmentation of inpainting perceptual artifacts, and apply the model for inpainting model evaluation and iterative refinement. Specifically, we first construct a new inpainting artifacts dataset by manually annotating perceptual artifacts in the results of state-of-the-art inpainting models. Then we train advanced segmentation networks on this dataset to reliably localize inpainting artifacts within inpainted images. Second, we propose a new interpretable evaluation metric called Perceptual Artifact Ratio (PAR), which is the ratio of objectionable inpainted regions to the entire inpainted area. PAR demonstrates a strong correlation with real user preference. Finally, we further apply the generated masks for iterative image inpainting by combining our approach with multiple recent inpainting methods. Extensive experiments demonstrate the consistent decrease of artifact regions and inpainting quality improvement across the different methods.

Wencan Cheng, Jong Hwan Ko

Scene flow estimation, which extracts point-wise motion between scenes, is becoming a crucial task in many computer vision tasks. However, all of the existing estimation methods utilize only the unidirectional features, restricting the accuracy and generality. This paper presents a novel scene flow estimation architecture using bidirectional flow embedding layers. The proposed bidirectional layer learns features along both forward and backward directions, enhancing the estimation performance. In addition, hierarchical feature extraction and warping improve the performance and reduce computational overhead. Experimental results show that the proposed architecture achieved a new state-of-the-art record by outperforming other approaches with large margin in both FlyingThings3D and KITTI benchmarks. Codes are available at https://github.com/cwc1260/BiFlow.

Dongwan Kim, Yi-Hsuan Tsai, Yumin Suh, Masoud Faraki, Sparsh Garg, Manmohan Chandraker, Bohyung Han

While it is desirable to train segmentation models on an aggregation of multiple datasets, a major challenge is that the label space of each dataset may be in conflict with one another. To tackle this challenge, we propose UniSeg, an effective and model-agnostic approach to automatically train segmentation models across multiple datasets with heterogeneous label spaces, without requiring any manual relabeling efforts. Specifically, we introduce two new ideas that account for conflicting and co-occurring labels to achieve better generalization performance in unseen domains. First, we identify a gradient conflict in training incurred by mismatched label spaces and propose a class-independent binary cross-entropy loss to alleviate such label conflicts. Second, we propose a loss function that considers class-relationships across datasets for a better multi-dataset training scheme.xtensive quantitative and qualitative analyses on road-scene datasets show that UniSeg improves over multi-dataset baselines, especially on unseen datasets, e.g., achieving more than 8%p gain in IoU on KITTI. Furthermore, UniSeg achieves 39.4% IoU on the WildDash2 public benchmark, making it one of the strongest submissions in the zero-shot setting. Our project page is available at https://www.nec-labs.com/ mas/UniSeg.

Xiangtai Li, Shilin Xu, Yibo Yang, Guangliang Cheng, Yunhai Tong, Dacheng Tao

Panoptic Part Segmentation (PPS) aims to unify panoptic segmentation and part segmentation into one task. Previous work mainly utilizes separated approaches to handle thing, stuff, and part predictions individually without performing any shared computation and task association. In this work, we aim to unify these tasks at the architectural level, designing the first end-to-end unified method named Panoptic-PartFormer. In particular, motivated by the recent progress in Vision Transformer, we model things, stuff, and part as object queries and directly learn to optimize the all three predictions as unified mask prediction and classification problem. We design a decoupled decoder to generate part feature and thing/stuff feature respectively. Then we propose to utilize all the queries and corresponding features to perform reasoning jointly and iteratively. The final mask can be obtained via inner product between queries and the corresponding features. The extensive ablation studies and analysis prove the effectiveness of our framework. Our Panoptic-PartFormer achieves the new state-of-the-art results on both Cityscapes PPS and Pascal Context PPS datasets with around 70\% GFlops and 50\% parameters decrease. Given its effectiveness and conceptual simplicity, we hope the Panoptic-PartFormer can serve as a strong baseline and aid future research in PPS. Our code and models will be available at \url{https://github.com/lxtGH/Panoptic-PartFormer}.

Xinyi Wu, Zhenyao Wu, Jin Wan, Lili Ju, Song Wang

Video semantic segmentation is a fundamental and important task in computer vision, and it usually requires large-scale labeled data for training deep neural network models. To avoid laborious manual labeling, domain adaptive video segmentation approaches were recently introduced by transferring the knowledge from the source domain of self-labeled simulated videos to the target domain of unlabeled real-world videos. However, it leads to an interesting question -- while video-to-video adaptation is a natural idea, does the source data require to be videos? In this paper, we argue that it is not necessary to transfer temporal knowledge since the temporal continuity of video segmentation in the target domain can be estimated and enforced without reference to videos in the source domain. This motivates a new framework of Image-to-Video Domain Adaptive Semantic Segmentation (I2VDA), where the source domain is a set of images without temporal information. Under this setting, we bridge the domain gap via adversarial training based on only the spatial knowledge, and develop a novel temporal augmentation strategy, through which the temporal consistency in the target domain is well-exploited and learned. In addition, we introduce a new training scheme by leveraging a proxy network to produce pseudo-labels on-the-fly, which is very effective to improve the stability of adversarial training. Experimental results on two synthetic-to-real scenarios show that the proposed I2VDA method can achieve even better performance on video semantic segmentation than existing state-of-the-art video-to-video domain adaption approaches.

Jingkang Yang, Yi Zhe Ang, Zujin Guo, Kaiyang Zhou, Wayne Zhang, Ziwei Liu

Existing research addresses scene graph generation (SGG), a critical technology to scene understanding in images, from the detection perspective, i.e., objects are detected using bounding boxes followed by prediction of their pairwise relationships. We argue that such a paradigm would cause several problems that impede the progress of the field. For instance, bounding box-based labels in current datasets usually contain redundant information like hairs and miss some background information that is crucial to the understanding of context. In this work, we introduce panoptic scene graph generation (PSG task), a new problem that requires the model to generate more comprehensive scene graph representations based on panoptic segmentations rather than rigid bounding boxes. A high-quality PSG dataset, which contains 51k well-annotated images from COCO and Visual Genome, is created for the community to keep track of the progress. For benchmarking, we build three two-stage models, which are modified from current state-of-the-arts in SGG, and another one-stage model called PSGTR, which is based on the efficient Transformer-based detector, i.e., DETR. We further propose an approach called PSGFormer, which achieves significant improvements with two novel extensions over PSGTR: 1) separate modeling of objects and relations in the form of queries in two Transformer decoders, and 2) a prompting-like interaction mechanism. In the end, we share insights on open challenges and future directions.

Jonathan Kahana, Yedid Hoshen

Learning representations of images that are invariant to sensitive or unwanted attributes is important for many tasks including bias removal and cross domain retrieval. Here, our objective is to learn representations that are invariant to the domain (sensitive attribute) for which labels are provided, while being informative over all other image attributes, which are unlabeled. We present a new approach, proposing a new domain-wise contrastive objective for ensuring invariant representations. This objective crucially restricts negative image pairs to be drawn from the same domain, which enforces domain invariance whereas the standard contrastive objective does not. This domain-wise objective is insufficient on its own as it suffers from shortcut solutions resulting in feature suppression. We overcome this issue by a combination of a reconstruction constraint, image augmentations and initialization with pre-trained weights. Our analysis shows that the choice of augmentations is important, and that a misguided choice of augmentations can harm the invariance and informativeness objectives. In an extensive evaluation, our method convincingly outperforms the state-of-the-art in terms of representation invariance, representation informativeness, and training speed. Furthermore, we find that in some cases our method can achieve excellent results even without the reconstruction constraint, leading to a much faster and resource efficient training.

Fabian Mentzer, Eirikur Agustsson, Johannes Ballé, David Minnen, Nick Johnston, George Toderici

We present the first neural video compression method based on generative adversarial networks (GANs). Our approach significantly outperforms previous neural and non-neural video compression methods in a user study, setting a new state-of-the-art in visual quality for neural methods. We show that the GAN loss is crucial to obtain this high visual quality. Two components make the GAN loss effective: we i) synthesize detail by conditioning the generator on a latent extracted from the warped previous reconstruction to then ii) propagate this detail with high-quality flow. We find that user studies are required to compare methods, i.e., none of our quantitative metrics were able to predict all studies. We present the network design choices in detail, and ablate them with user studies.

Norman Mu, Alexander Kirillov, David Wagner, Saining Xie

Recent work has shown that self-supervised pre-training leads to improvements over supervised learning on challenging visual recognition tasks. CLIP, an exciting new approach to learning with language supervision, demonstrates promising performance on a wide variety of benchmarks. In this work, we explore whether self-supervised learning can aid in the use of language supervision for visual representation learning with Vision Transformers. We introduce SLIP, a multi-task learning framework for combining self-supervised learning and CLIP pre-training. After pre-training, we thoroughly evaluate representation quality and compare performance to both CLIP and self-supervised learning under three distinct settings: zero-shot transfer, linear classification, and end-to-end finetuning. Across ImageNet and a battery of additional datasets, we find that SLIP improves accuracy by a large margin. We validate our results further with experiments on different model sizes, training schedules, and pre-training datasets. Our findings show that SLIP enjoys the best of both worlds: better performance than self-supervision (+8.1% linear accuracy) and language supervision (+5.2% zero-shot accuracy). Our code is available at: github.com/facebookresearch/SLIP.

Mingfu Liang, Jiahuan Zhou, Wei Wei, Ying Wu

The subpopulation shifting challenge, known as some subpopulations of a category that are not seen during training, severely limits the classification performance of the state-of-the-art convolutional neural networks. Thus, to mitigate this practical issue, we explore incremental subpopulation learning (ISL) to adapt the original model via incrementally learning the unseen subpopulations without retaining the seen population data. However, striking a great balance between subpopulation learning and seen population forgetting is the main challenge in ISL but is not well studied by existing approaches. These incremental learners simply use a pre-defined and fixed hyperparameter to balance the learning objective and forgetting regularization, but their learning is usually biased towards either side in the long run. In this paper, we propose a novel two-stage learning scheme to explicitly disentangle the acquisition and forgetting for achieving a better balance between subpopulation learning and seen population forgetting: in the first “gain-acquisition” stage, we progressively learn a new classifier based on the margin-enforce loss, which enforces the hard samples and population to have a larger weight for classifier updating and avoid uniformly updating all the population; in the second “counter-forgetting” stage, we search for the proper combination of the new and old classifiers by optimizing a novel objective based on proxies of forgetting and acquisition. We benchmark the representative and state-of-the-art non-exemplar-based incremental learning methods on a large-scale subpopulation shifting dataset for the first time. Under almost all the challenging ISL protocols, we significantly outperform other methods by a large margin, demonstrating our superiority to alleviate the subpopulation shifting problem.

Liyuan Wang, Xingxing Zhang, Qian Li, Jun Zhu, Yi Zhong

Continual learning requires incremental compatibility with a sequence of tasks. However, the design of model architecture remains an open question: In general, learning all tasks with a shared set of parameters suffers from severe interference between tasks; while learning each task with a dedicated parameter subspace is limited by scalability. In this work, we theoretically analyze the generalization errors for learning plasticity and memory stability in continual learning, which can be uniformly upper-bounded by (1) discrepancy between task distributions, (2) flatness of loss landscape and (3) cover of parameter space. Then, inspired by the robust biological learning system that processes sequential experiences with multiple parallel compartments, we propose Cooperation of Small Continual Learners (CoSCL) as a general strategy for continual learning. Specifically, we present an architecture with a fixed number of narrower sub-networks to learn all incremental tasks in parallel, which can naturally reduce the two errors through improving the three components of the upper bound. To strengthen this advantage, we encourage to cooperate these sub-networks by penalizing the difference of predictions made by their feature representations. With a fixed parameter budget, CoSCL can improve a variety of representative continual learning approaches by a large margin (e.g., up to 10.64% on CIFAR-100-SC, 9.33% on CIFAR-100-RS, 11.45% on CUB-200-2011 and 6.72% on Tiny-ImageNet) and achieve the new state-of-the-art performance. Our code is available at https://github.com/lywang3081/CoSCL.