The handling of long videos with complex and occluded sequences has recently emerged as a new challenge in the video instance segmentation (VIS) community. However, existing methods have limitations in addressing this challenge. We argue that the biggest bottleneck in current approaches is the discrepancy between training and inference. To effectively bridge this gap, we propose a Generalized framework for VIS, namely GenVIS, that achieves state-of-the-art performance on challenging benchmarks without designing complicated architectures or requiring extra post-processing. The key contribution of GenVIS is the learning strategy, which includes a query-based training pipeline for sequential learning with a novel target label assignment. Additionally, we introduce a memory that effectively acquires information from previous states. Thanks to the new perspective, which focuses on building relationships between separate frames or clips, GenVIS can be flexibly executed in both online and semi-online manner. We evaluate our approach on popular VIS benchmarks, achieving state-of-the-art results on YouTube-VIS 2019/2021/2022 and Occluded VIS (OVIS). Notably, we greatly outperform the state-of-the-art on the long VIS benchmark (OVIS), improving 5.6 AP with ResNet-50 backbone. Code is available at https://github.com/miranheo/GenVIS.
For an autonomous vehicle it is essential to observe the ongoing dynamics of a scene and consequently predict imminent future scenarios to ensure safety to itself and others. This can be done using different sensors and modalities. In this paper we investigate the usage of optical flow for predicting future semantic segmentations. To do so we propose a model that forecasts flow fields autoregressively. Such predictions are then used to guide the inference of a learned warping function that moves instance segmentations on to future frames. Results on the Cityscapes dataset demonstrate the effectiveness of optical-flow methods.
The game of Jenga represents an inspiring benchmark for developing innovative manipulation solutions for complex tasks. Indeed, it encouraged the study of novel robotics methods to successfully extract blocks from the tower. A Jenga game round undoubtedly embeds many traits of complex industrial or surgical manipulation tasks, requiring a multi-step strategy, the combination of visual and tactile data, and the highly precise motion of the robotic arm to perform a single block extraction. In this work, we propose a novel, cost-effective architecture for playing Jenga with e.Do, a 6-DOF anthropomorphic manipulator manufactured by Comau, a standard depth camera, and an inexpensive monodirectional force sensor. Our solution focuses on a visual-based control strategy to accurately align the end-effector with the desired block, enabling block extraction by pushing. To this aim, we train an instance segmentation deep learning model on a synthetic custom dataset to segment each piece of the Jenga tower, allowing visual tracking of the desired block's pose during the motion of the manipulator. We integrate the visual-based strategy with a 1D force sensor to detect whether the block can be safely removed by identifying a force threshold value. Our experimentation shows that our low-cost solution allows e.DO to precisely reach removable blocks and perform up to 14 consecutive extractions in a row.
3D object detection is a critical task in autonomous driving. Recently multi-modal fusion-based 3D object detection methods, which combine the complementary advantages of LiDAR and camera, have shown great performance improvements over mono-modal methods. However, so far, no methods have attempted to utilize the instance-level contextual image semantics to guide the 3D object detection. In this paper, we propose a simple and effective Painting Adaptive Instance-prior for 3D object detection (PAI3D) to fuse instance-level image semantics flexibly with point cloud features. PAI3D is a multi-modal sequential instance-level fusion framework. It first extracts instance-level semantic information from images, the extracted information, including objects categorical label, point-to-object membership and object position, are then used to augment each LiDAR point in the subsequent 3D detection network to guide and improve detection performance. PAI3D outperforms the state-of-the-art with a large margin on the nuScenes dataset, achieving 71.4 in mAP and 74.2 in NDS on the test split. Our comprehensive experiments show that instance-level image semantics contribute the most to the performance gain, and PAI3D works well with any good-quality instance segmentation models and any modern point cloud 3D encoders, making it a strong candidate for deployment on autonomous vehicles.
We launch EVA, a vision-centric foundation model to explore the limits of visual representation at scale using only publicly accessible data. EVA is a vanilla ViT pre-trained to reconstruct the masked out image-text aligned vision features conditioned on visible image patches. Via this pretext task, we can efficiently scale up EVA to one billion parameters, and sets new records on a broad range of representative vision downstream tasks, such as image recognition, video action recognition, object detection, instance segmentation and semantic segmentation without heavy supervised training. Moreover, we observe quantitative changes in scaling EVA result in qualitative changes in transfer learning performance that are not present in other models. For instance, EVA takes a great leap in the challenging large vocabulary instance segmentation task: our model achieves almost the same state-of-the-art performance on LVISv1.0 dataset with over a thousand categories and COCO dataset with only eighty categories. Beyond a pure vision encoder, EVA can also serve as a vision-centric, multi-modal pivot to connect images and text. We find initializing the vision tower of a giant CLIP from EVA can greatly stabilize the training and outperform the training from scratch counterpart with much fewer samples and less compute, providing a new direction for scaling up and accelerating the costly training of multi-modal foundation models. To facilitate future research, we release all the code and models at https://github.com/baaivision/EVA.
In comparative neuroanatomy, the characterization of brain cytoarchitecture is critical to a better understanding of brain structure and function, as it helps to distill information on the development, evolution, and distinctive features of different populations. The automatic segmentation of individual brain cells is a primary prerequisite and yet remains challenging. A new method (MR-NOM) was developed for the instance segmentation of cells in Nissl-stained histological images of the brain. MR-NOM exploits a multi-scale approach to deliberately over-segment the cells into superpixels and subsequently merge them via a classifier based on shape, structure, and intensity features. The method was tested on images of the cerebral cortex, proving successful in dealing with cells of varying characteristics that partially touch or overlap, showing better performance than two state-of-the-art methods.
Heavily relying on 3D annotations limits the real-world application of 3D object detection. In this paper, we propose a method that does not demand any 3D annotation, while being able to predict fully oriented 3D bounding boxes. Our method, called Recursive Cross-View (RCV), utilizes the three-view principle to convert 3D detection into multiple 2D detection tasks, requiring only a subset of 2D labels. We propose a recursive paradigm, in which instance segmentation and 3D bounding box generation by Cross-View are implemented recursively until convergence. Specifically, our proposed method involves the use of a frustum for each 2D bounding box, which is then followed by the recursive paradigm that ultimately generates a fully oriented 3D box, along with its corresponding class and score. Note that, class and score are given by the 2D detector. Estimated on the SUN RGB-D and KITTI datasets, our method outperforms existing image-based approaches. To justify that our method can be quickly used to new tasks, we implement it on two real-world scenarios, namely 3D human detection and 3D hand detection. As a result, two new 3D annotated datasets are obtained, which means that RCV can be viewed as a (semi-) automatic 3D annotator. Furthermore, we deploy RCV on a depth sensor, which achieves detection at 7 fps on a live RGB-D stream. RCV is the first 3D detection method that yields fully oriented 3D boxes without consuming 3D labels.
Universal Image Segmentation is not a new concept. Past attempts to unify image segmentation in the last decades include scene parsing, panoptic segmentation, and, more recently, new panoptic architectures. However, such panoptic architectures do not truly unify image segmentation because they need to be trained individually on the semantic, instance, or panoptic segmentation to achieve the best performance. Ideally, a truly universal framework should be trained only once and achieve SOTA performance across all three image segmentation tasks. To that end, we propose OneFormer, a universal image segmentation framework that unifies segmentation with a multi-task train-once design. We first propose a task-conditioned joint training strategy that enables training on ground truths of each domain (semantic, instance, and panoptic segmentation) within a single multi-task training process. Secondly, we introduce a task token to condition our model on the task at hand, making our model task-dynamic to support multi-task training and inference. Thirdly, we propose using a query-text contrastive loss during training to establish better inter-task and inter-class distinctions. Notably, our single OneFormer model outperforms specialized Mask2Former models across all three segmentation tasks on ADE20k, CityScapes, and COCO, despite the latter being trained on each of the three tasks individually with three times the resources. With new ConvNeXt and DiNAT backbones, we observe even more performance improvement. We believe OneFormer is a significant step towards making image segmentation more universal and accessible. To support further research, we open-source our code and models at https://github.com/SHI-Labs/OneFormer
Compared to the great progress of large-scale vision transformers (ViTs) in recent years, large-scale models based on convolutional neural networks (CNNs) are still in an early state. This work presents a new large-scale CNN-based foundation model, termed InternImage, which can obtain the gain from increasing parameters and training data like ViTs. Different from the recent CNNs that focus on large dense kernels, InternImage takes deformable convolution as the core operator, so that our model not only has the large effective receptive field required for downstream tasks such as detection and segmentation, but also has the adaptive spatial aggregation conditioned by input and task information. As a result, the proposed InternImage reduces the strict inductive bias of traditional CNNs and makes it possible to learn stronger and more robust patterns with large-scale parameters from massive data like ViTs. The effectiveness of our model is proven on challenging benchmarks including ImageNet, COCO, and ADE20K. It is worth mentioning that InternImage-H achieved a new record 65.4 mAP on COCO test-dev and 62.9 mIoU on ADE20K, outperforming current leading CNNs and ViTs. The code will be released at https://github.com/OpenGVLab/InternImage.
We present Polite Teacher, a simple yet effective method for the task of semi-supervised instance segmentation. The proposed architecture relies on the Teacher-Student mutual learning framework. To filter out noisy pseudo-labels, we use confidence thresholding for bounding boxes and mask scoring for masks. The approach has been tested with CenterMask, a single-stage anchor-free detector. Tested on the COCO 2017 val dataset, our architecture significantly (approx. +8 pp. in mask AP) outperforms the baseline at different supervision regimes. To the best of our knowledge, this is one of the first works tackling the problem of semi-supervised instance segmentation and the first one devoted to an anchor-free detector.
Generally, microscopy image analysis in biology relies on the segmentation of individual nuclei, using a dedicated stained image, to identify individual cells. However stained nuclei have drawbacks like the need for sample preparation, and specific equipment on the microscope but most importantly, and as it is in most cases, the nuclear stain is not relevant to the biological questions of interest but is solely used for the segmentation task. In this study, we used non-stained brightfield images for nuclei segmentation with the advantage that they can be acquired on any microscope from both live or fixed samples and do not necessitate specific sample preparation. Nuclei semantic segmentation from brightfield images was obtained, on four distinct cell lines with U-Net-based architectures. We tested systematically deep pre-trained encoders to identify the best performing in combination with the different neural network architectures used. Additionally, two distinct and effective strategies were employed for instance segmentation, followed by thorough instance evaluation. We obtained effective semantic and instance segmentation of nuclei in brightfield images from standard test sets as well as from very diverse biological contexts triggered upon treatment with various small molecule inhibitor. The code used in this study was made public to allow further use by the community.
By contextualizing the kernel as global as possible, Modern ConvNets have shown great potential in computer vision tasks. However, recent progress on \textit{multi-order game-theoretic interaction} within deep neural networks (DNNs) reveals the representation bottleneck of modern ConvNets, where the expressive interactions have not been effectively encoded with the increased kernel size. To tackle this challenge, we propose a new family of modern ConvNets, dubbed MogaNet, for discriminative visual representation learning in pure ConvNet-based models with favorable complexity-performance trade-offs. MogaNet encapsulates conceptually simple yet effective convolutions and gated aggregation into a compact module, where discriminative features are efficiently gathered and contextualized adaptively. MogaNet exhibits great scalability, impressive efficiency of parameters, and competitive performance compared to state-of-the-art ViTs and ConvNets on ImageNet and various downstream vision benchmarks, including COCO object detection, ADE20K semantic segmentation, 2D\&3D human pose estimation, and video prediction. Notably, MogaNet hits 80.0\% and 87.8\% accuracy with 5.2M and 181M parameters on ImageNet-1K, outperforming ParC-Net and ConvNeXt-L, while saving 59\% FLOPs and 17M parameters, respectively. The source code is available at \url{https://github.com/Westlake-AI/MogaNet}.
In this research work, we have demonstrated the application of Mask-RCNN (Regional Convolutional Neural Network), a deep-learning algorithm for computer vision and specifically object detection, to semiconductor defect inspection domain. Stochastic defect detection and classification during semiconductor manufacturing has grown to be a challenging task as we continuously shrink circuit pattern dimensions (e.g., for pitches less than 32 nm). Defect inspection and analysis by state-of-the-art optical and e-beam inspection tools is generally driven by some rule-based techniques, which in turn often causes to misclassification and thereby necessitating human expert intervention. In this work, we have revisited and extended our previous deep learning-based defect classification and detection method towards improved defect instance segmentation in SEM images with precise extent of defect as well as generating a mask for each defect category/instance. This also enables to extract and calibrate each segmented mask and quantify the pixels that make up each mask, which in turn enables us to count each categorical defect instances as well as to calculate the surface area in terms of pixels. We are aiming at detecting and segmenting different types of inter-class stochastic defect patterns such as bridge, break, and line collapse as well as to differentiate accurately between intra-class multi-categorical defect bridge scenarios (as thin/single/multi-line/horizontal/non-horizontal) for aggressive pitches as well as thin resists (High NA applications). Our proposed approach demonstrates its effectiveness both quantitatively and qualitatively.
There is limited understanding of the information captured by deep spatiotemporal models in their intermediate representations. For example, while evidence suggests that action recognition algorithms are heavily influenced by visual appearance in single frames, no quantitative methodology exists for evaluating such static bias in the latent representation compared to bias toward dynamics. We tackle this challenge by proposing an approach for quantifying the static and dynamic biases of any spatiotemporal model, and apply our approach to three tasks, action recognition, automatic video object segmentation (AVOS) and video instance segmentation (VIS). Our key findings are: (i) Most examined models are biased toward static information. (ii) Some datasets that are assumed to be biased toward dynamics are actually biased toward static information. (iii) Individual channels in an architecture can be biased toward static, dynamic or a combination of the two. (iv) Most models converge to their culminating biases in the first half of training. We then explore how these biases affect performance on dynamically biased datasets. For action recognition, we propose StaticDropout, a semantically guided dropout that debiases a model from static information toward dynamics. For AVOS, we design a better combination of fusion and cross connection layers compared with previous architectures.
Frozen pretrained models have become a viable alternative to the pretraining-then-finetuning paradigm for transfer learning. However, with frozen models there are relatively few parameters available for adapting to downstream tasks, which is problematic in computer vision where tasks vary significantly in input/output format and the type of information that is of value. In this paper, we present a study of frozen pretrained models when applied to diverse and representative computer vision tasks, including object detection, semantic segmentation and video action recognition. From this empirical analysis, our work answers the questions of what pretraining task fits best with this frozen setting, how to make the frozen setting more flexible to various downstream tasks, and the effect of larger model sizes. We additionally examine the upper bound of performance using a giant frozen pretrained model with 3 billion parameters (SwinV2-G) and find that it reaches competitive performance on a varied set of major benchmarks with only one shared frozen base network: 60.0 box mAP and 52.2 mask mAP on COCO object detection test-dev, 57.6 val mIoU on ADE20K semantic segmentation, and 81.7 top-1 accuracy on Kinetics-400 action recognition. With this work, we hope to bring greater attention to this promising path of freezing pretrained image models.
Circle representation has recently been introduced as a medical imaging optimized representation for more effective instance object detection on ball-shaped medical objects. With its superior performance on instance detection, it is appealing to extend the circle representation to instance medical object segmentation. In this work, we propose CircleSnake, a simple end-to-end circle contour deformation-based segmentation method for ball-shaped medical objects. Compared to the prevalent DeepSnake method, our contribution is three-fold: (1) We replace the complicated bounding box to octagon contour transformation with a computation-free and consistent bounding circle to circle contour adaption for segmenting ball-shaped medical objects; (2) Circle representation has fewer degrees of freedom (DoF=2) as compared with the octagon representation (DoF=8), thus yielding a more robust segmentation performance and better rotation consistency; (3) To the best of our knowledge, the proposed CircleSnake method is the first end-to-end circle representation deep segmentation pipeline method with consistent circle detection, circle contour proposal, and circular convolution. The key innovation is to integrate the circular graph convolution with circle detection into an end-to-end instance segmentation framework, enabled by the proposed simple and consistent circle contour representation. Glomeruli are used to evaluate the performance of the benchmarks. From the results, CircleSnake increases the average precision of glomerular detection from 0.559 to 0.614. The Dice score increased from 0.804 to 0.849. The code has been released: https://github.com/hrlblab/CircleSnake
In Video Instance Segmentation (VIS), current approaches either focus on the quality of the results, by taking the whole video as input and processing it offline; or on speed, by handling it frame by frame at the cost of competitive performance. In this work, we propose an online method that is on par with the performance of the offline counterparts. We introduce a message-passing graph neural network that encodes objects and relates them through time. We additionally propose a novel module to fuse features from the feature pyramid network with residual connections. Our model, trained end-to-end, achieves state-of-the-art performance on the YouTube-VIS dataset within the online methods. Further experiments on DAVIS demonstrate the generalization capability of our model to the video object segmentation task. Code is available at: \url{https://github.com/caganselim/TLTM}
Vision Transformers (ViTs) have recently become the state-of-the-art across many computer vision tasks. In contrast to convolutional networks (CNNs), ViTs enable global information sharing even within shallow layers of a network, i.e., among high-resolution features. However, this perk was later overlooked with the success of pyramid architectures such as Swin Transformer, which show better performance-complexity trade-offs. In this paper, we present a simple and efficient add-on component (termed GrafT) that considers global dependencies and multi-scale information throughout the network, in both high- and low-resolution features alike. It has the flexibility of branching out at arbitrary depths and shares most of the parameters and computations of the backbone. GrafT shows consistent gains over various well-known models which includes both hybrid and pure Transformer types, both homogeneous and pyramid structures, and various self-attention methods. In particular, it largely benefits mobile-size models by providing high-level semantics. On the ImageNet-1k dataset, GrafT delivers +3.9%, +1.4%, and +1.9% top-1 accuracy improvement to DeiT-T, Swin-T, and MobileViT-XXS, respectively. Our code and models will be made available.
We address the problem of detecting and erasing furniture from a wide angle photograph of a room. Inpainting large regions of an indoor scene often results in geometric inconsistencies of background elements within the inpaint mask. To address this problem, we utilize perceptual information (e.g. instance segmentation, and room layout) to produce a geometrically consistent empty version of a room. We share important details to make this system viable, such as per-plane inpainting, automatic rectification, and texture refinement. We provide detailed ablation along with qualitative examples, justifying our design choices. We show an application of our system by removing real furniture from a room and redecorating it with virtual furniture.
Stroke is the basic element of Chinese character and stroke extraction has been an important and long-standing endeavor. Existing stroke extraction methods are often handcrafted and highly depend on domain expertise due to the limited training data. Moreover, there are no standardized benchmarks to provide a fair comparison between different stroke extraction methods, which, we believe, is a major impediment to the development of Chinese character stroke understanding and related tasks. In this work, we present the first public available Chinese Character Stroke Extraction (CCSE) benchmark, with two new large-scale datasets: Kaiti CCSE (CCSE-Kai) and Handwritten CCSE (CCSE-HW). With the large-scale datasets, we hope to leverage the representation power of deep models such as CNNs to solve the stroke extraction task, which, however, remains an open question. To this end, we turn the stroke extraction problem into a stroke instance segmentation problem. Using the proposed datasets to train a stroke instance segmentation model, we surpass previous methods by a large margin. Moreover, the models trained with the proposed datasets benefit the downstream font generation and handwritten aesthetic assessment tasks. We hope these benchmark results can facilitate further research. The source code and datasets are publicly available at: https://github.com/lizhaoliu-Lec/CCSE.