Automatic surgical scene segmentation is fundamental for facilitating cognitive intelligence in the modern operating theatre. Previous works rely on conventional aggregation modules (e.g., dilated convolution, convolutional LSTM), which only make use of the local context. In this paper, we propose a novel framework STswinCL that explores the complementary intra- and inter-video relations to boost segmentation performance, by progressively capturing the global context. We firstly develop a hierarchy Transformer to capture intra-video relation that includes richer spatial and temporal cues from neighbor pixels and previous frames. A joint space-time window shift scheme is proposed to efficiently aggregate these two cues into each pixel embedding. Then, we explore inter-video relation via pixel-to-pixel contrastive learning, which well structures the global embedding space. A multi-source contrast training objective is developed to group the pixel embeddings across videos with the ground-truth guidance, which is crucial for learning the global property of the whole data. We extensively validate our approach on two public surgical video benchmarks, including EndoVis18 Challenge and CaDIS dataset. Experimental results demonstrate the promising performance of our method, which consistently exceeds previous state-of-the-art approaches. Code is available at https://github.com/YuemingJin/STswinCL.
Autonomous vehicles utilize urban scene segmentation to understand the real world like a human and react accordingly. Semantic segmentation of normal scenes has experienced a remarkable rise in accuracy on conventional benchmarks. However, a significant portion of real-life accidents features abnormal scenes, such as those with object deformations, overturns, and unexpected traffic behaviors. Since even small mis-segmentation of driving scenes can lead to serious threats to human lives, the robustness of such models in accident scenarios is an extremely important factor in ensuring safety of intelligent transportation systems. In this paper, we propose a Multi-source Meta-learning Unsupervised Domain Adaptation (MMUDA) framework, to improve the generalization of segmentation transformers to extreme accident scenes. In MMUDA, we make use of Multi-Domain Mixed Sampling to augment the images of multiple-source domains (normal scenes) with the target data appearances (abnormal scenes). To train our model, we intertwine and study a meta-learning strategy in the multi-source setting for robustifying the segmentation results. We further enhance the segmentation backbone (SegFormer) with a HybridASPP decoder design, featuring large window attention spatial pyramid pooling and strip pooling, to efficiently aggregate long-range contextual dependencies. Our approach achieves a mIoU score of 46.97% on the DADA-seg benchmark, surpassing the previous state-of-the-art model by more than 7.50%. Code will be made publicly available at https://github.com/xinyu-laura/MMUDA.
Current visual detectors, though impressive within their training distribution, often fail to parse out-of-distribution scenes into their constituent entities. Recent test-time adaptation methods use auxiliary self-supervised losses to adapt the network parameters to each test example independently and have shown promising results towards generalization outside the training distribution for the task of image classification. In our work, we find evidence that these losses are insufficient for the task of scene decomposition, without also considering architectural inductive biases. Recent slot-centric generative models attempt to decompose scenes into entities in a self-supervised manner by reconstructing pixels. Drawing upon these two lines of work, we propose Slot-TTA, a semi-supervised slot-centric scene decomposition model that at test time is adapted per scene through gradient descent on reconstruction or cross-view synthesis objectives. We evaluate Slot-TTA across multiple input modalities, images or 3D point clouds, and show substantial out-of-distribution performance improvements against state-of-the-art supervised feed-forward detectors, and alternative test-time adaptation methods.
Self-supervised learning has drawn attention through its effectiveness in learning in-domain representations with no ground-truth annotations; in particular, it is shown that properly designed pretext tasks (e.g., contrastive prediction task) bring significant performance gains for downstream tasks (e.g., classification task). Inspired from this, we tackle video scene segmentation, which is a task of temporally localizing scene boundaries in a video, with a self-supervised learning framework where we mainly focus on designing effective pretext tasks. In our framework, we discover a pseudo-boundary from a sequence of shots by splitting it into two continuous, non-overlapping sub-sequences and leverage the pseudo-boundary to facilitate the pre-training. Based on this, we introduce three novel boundary-aware pretext tasks: 1) Shot-Scene Matching (SSM), 2) Contextual Group Matching (CGM) and 3) Pseudo-boundary Prediction (PP); SSM and CGM guide the model to maximize intra-scene similarity and inter-scene discrimination while PP encourages the model to identify transitional moments. Through comprehensive analysis, we empirically show that pre-training and transferring contextual representation are both critical to improving the video scene segmentation performance. Lastly, we achieve the new state-of-the-art on the MovieNet-SSeg benchmark. The code is available at https://github.com/kakaobrain/bassl.
A powerful simulator highly decreases the need for real-world tests when training and evaluating autonomous vehicles. Data-driven simulators flourished with the recent advancement of conditional Generative Adversarial Networks (cGANs), providing high-fidelity images. The main challenge is synthesizing photorealistic images while following given constraints. In this work, we propose to improve the quality of generated images by rethinking the discriminator architecture. The focus is on the class of problems where images are generated given semantic inputs, such as scene segmentation maps or human body poses. We build on successful cGAN models to propose a new semantically-aware discriminator that better guides the generator. We aim to learn a shared latent representation that encodes enough information to jointly do semantic segmentation, content reconstruction, along with a coarse-to-fine grained adversarial reasoning. The achieved improvements are generic and simple enough to be applied to any architecture of conditional image synthesis. We demonstrate the strength of our method on the scene, building, and human synthesis tasks across three different datasets. The code is available at https://github.com/vita-epfl/SemDisc.
The robustness of semantic segmentation on edge cases of traffic scene is a vital factor for the safety of intelligent transportation. However, most of the critical scenes of traffic accidents are extremely dynamic and previously unseen, which seriously harm the performance of semantic segmentation methods. In addition, the delay of the traditional camera during high-speed driving will further reduce the contextual information in the time dimension. Therefore, we propose to extract dynamic context from event-based data with a higher temporal resolution to enhance static RGB images, even for those from traffic accidents with motion blur, collisions, deformations, overturns, etc. Moreover, in order to evaluate the segmentation performance in traffic accidents, we provide a pixel-wise annotated accident dataset, namely DADA-seg, which contains a variety of critical scenarios from traffic accidents. Our experiments indicate that event-based data can provide complementary information to stabilize semantic segmentation under adverse conditions by preserving fine-grained motion of fast-moving foreground (crash objects) in accidents. Our approach achieves +8.2% performance gain on the proposed accident dataset, exceeding more than 20 state-of-the-art semantic segmentation methods. The proposal has been demonstrated to be consistently effective for models learned on multiple source databases including Cityscapes, KITTI-360, BDD, and ApolloScape.
Geometric feature learning for 3D surfaces is critical for many applications in computer graphics and 3D vision. However, deep learning currently lags in hierarchical modeling of 3D surfaces due to the lack of required operations and/or their efficient implementations. In this paper, we propose a series of modular operations for effective geometric feature learning from 3D triangle meshes. These operations include novel mesh convolutions, efficient mesh decimation and associated mesh (un)poolings. Our mesh convolutions exploit spherical harmonics as orthonormal bases to create continuous convolutional filters. The mesh decimation module is GPU-accelerated and able to process batched meshes on-the-fly, while the (un)pooling operations compute features for up/down-sampled meshes. We provide open-source implementation of these operations, collectively termed Picasso. Picasso supports heterogeneous mesh batching and processing. Leveraging its modular operations, we further contribute a novel hierarchical neural network for perceptual parsing of 3D surfaces, named PicassoNet++. It achieves highly competitive performance for shape analysis and scene segmentation on prominent 3D benchmarks. The code, data and trained models are available at https://github.com/EnyaHermite/Picasso.
Point cloud processing is a challenging task due to its sparsity and irregularity. Prior works introduce delicate designs on either local feature aggregator or global geometric architecture, but few combine both advantages. We propose Dual-Scale Point Cloud Recognition with High-frequency Fusion (DSPoint) to extract local-global features by concurrently operating on voxels and points. We reverse the conventional design of applying convolution on voxels and attention to points. Specifically, we disentangle point features through channel dimension for dual-scale processing: one by point-wise convolution for fine-grained geometry parsing, the other by voxel-wise global attention for long-range structural exploration. We design a co-attention fusion module for feature alignment to blend local-global modalities, which conducts inter-scale cross-modality interaction by communicating high-frequency coordinates information. Experiments and ablations on widely-adopted ModelNet40, ShapeNet, and S3DIS demonstrate the state-of-the-art performance of our DSPoint.
Recent efforts in multi-domain learning for semantic segmentation attempt to learn multiple geographical datasets in a universal, joint model. A simple fine-tuning experiment performed sequentially on three popular road scene segmentation datasets demonstrates that existing segmentation frameworks fail at incrementally learning on a series of visually disparate geographical domains. When learning a new domain, the model catastrophically forgets previously learned knowledge. In this work, we pose the problem of multi-domain incremental learning for semantic segmentation. Given a model trained on a particular geographical domain, the goal is to (i) incrementally learn a new geographical domain, (ii) while retaining performance on the old domain, (iii) given that the previous domain's dataset is not accessible. We propose a dynamic architecture that assigns universally shared, domain-invariant parameters to capture homogeneous semantic features present in all domains, while dedicated domain-specific parameters learn the statistics of each domain. Our novel optimization strategy helps achieve a good balance between retention of old knowledge (stability) and acquiring new knowledge (plasticity). We demonstrate the effectiveness of our proposed solution on domain incremental settings pertaining to real-world driving scenes from roads of Germany (Cityscapes), the United States (BDD100k), and India (IDD).
The fully convolutional network (FCN) has achieved tremendous success in dense visual recognition tasks, such as scene segmentation. The last layer of FCN is typically a global classifier (1x1 convolution) to recognize each pixel to a semantic label. We empirically show that this global classifier, ignoring the intra-class distinction, may lead to sub-optimal results. In this work, we present a conditional classifier to replace the traditional global classifier, where the kernels of the classifier are generated dynamically conditioned on the input. The main advantages of the new classifier consist of: (i) it attends on the intra-class distinction, leading to stronger dense recognition capability; (ii) the conditional classifier is simple and flexible to be integrated into almost arbitrary FCN architectures to improve the prediction. Extensive experiments demonstrate that the proposed classifier performs favourably against the traditional classifier on the FCN architecture. The framework equipped with the conditional classifier (called CondNet) achieves new state-of-the-art performances on two datasets. The code and models are available at https://git.io/CondNet.
In recent years, using a deep convolutional neural network (CNN) as a feature encoder (or backbone) is the most commonly observed architectural pattern in several computer vision methods, and semantic segmentation is no exception. The two major drawbacks of this architectural pattern are: (i) the networks often fail to capture small classes such as wall, fence, pole, traffic light, traffic sign, and bicycle, which are crucial for autonomous vehicles to make accurate decisions. (ii) due to the arbitrarily increasing depth, the networks require massive labeled data and additional regularization techniques to converge and to prevent the risk of over-fitting, respectively. While regularization techniques come at minimal cost, the collection of labeled data is an expensive and laborious process. In this work, we address these two drawbacks by proposing a novel lightweight architecture named point-wise dense flow network (PDFNet). In PDFNet, we employ dense, residual, and multiple shortcut connections to allow a smooth gradient flow to all parts of the network. The extensive experiments on Cityscapes and CamVid benchmarks demonstrate that our method significantly outperforms baselines in capturing small classes and in few-data regimes. Moreover, our method achieves considerable performance in classifying out-of-the training distribution samples, evaluated on Cityscapes to KITTI dataset.
Semantic segmentation of remotely sensed urban scene images is required in a wide range of practical applications, such as land cover mapping, urban change detection, environmental protection, and economic assessment.Driven by rapid developments in deep learning technologies, the convolutional neural network (CNN) has dominated semantic segmentation for many years. CNN adopts hierarchical feature representation, demonstrating strong capabilities for local information extraction. However, the local property of the convolution layer limits the network from capturing the global context. Recently, as a hot topic in the domain of computer vision, Transformer has demonstrated its great potential in global information modelling, boosting many vision-related tasks such as image classification, object detection, and particularly semantic segmentation. In this paper, we propose a Transformer-based decoder and construct a UNet-like Transformer (UNetFormer) for real-time urban scene segmentation. For efficient segmentation, the UNetFormer selects the lightweight ResNet18 as the encoder and develops an efficient global-local attention mechanism to model both global and local information in the decoder. Extensive experiments reveal that our method not only runs faster but also produces higher accuracy compared with state-of-the-art lightweight models. Specifically, the proposed UNetFormer achieved 67.8% and 52.4% mIoU on the UAVid and LoveDA datasets, respectively, while the inference speed can achieve up to 322.4 FPS with a 512x512 input on a single NVIDIA GTX 3090 GPU. In further exploration, the proposed Transformer-based decoder combined with a Swin Transformer encoder also achieves the state-of-the-art result (91.3% F1 and 84.1% mIoU) on the Vaihingen dataset. The source code will be freely available at https://github.com/WangLibo1995/GeoSeg.
Transparent objects, such as glass walls and doors, constitute architectural obstacles hindering the mobility of people with low vision or blindness. For instance, the open space behind glass doors is inaccessible, unless it is correctly perceived and interacted with. However, traditional assistive technologies rarely cover the segmentation of these safety-critical transparent objects. In this paper, we build a wearable system with a novel dual-head Transformer for Transparency (Trans4Trans) perception model, which can segment general- and transparent objects. The two dense segmentation results are further combined with depth information in the system to help users navigate safely and assist them to negotiate transparent obstacles. We propose a lightweight Transformer Parsing Module (TPM) to perform multi-scale feature interpretation in the transformer-based decoder. Benefiting from TPM, the double decoders can perform joint learning from corresponding datasets to pursue robustness, meanwhile maintain efficiency on a portable GPU, with negligible calculation increase. The entire Trans4Trans model is constructed in a symmetrical encoder-decoder architecture, which outperforms state-of-the-art methods on the test sets of Stanford2D3D and Trans10K-v2 datasets, obtaining mIoU of 45.13% and 75.14%, respectively. Through a user study and various pre-tests conducted in indoor and outdoor scenes, the usability and reliability of our assistive system have been extensively verified. Meanwhile, the Tran4Trans model has outstanding performances on driving scene datasets. On Cityscapes, ACDC, and DADA-seg datasets corresponding to common environments, adverse weather, and traffic accident scenarios, mIoU scores of 81.5%, 76.3%, and 39.2% are obtained, demonstrating its high efficiency and robustness for real-world transportation applications.
Semantic segmentation aims to predict pixel-level labels. It has become a popular task in various computer vision applications. While fully supervised segmentation methods have achieved high accuracy on large-scale vision datasets, they are unable to generalize on a new test environment or a new domain well. In this work, we first introduce a new Un-aligned Domain Score to measure the efficiency of a learned model on a new target domain in unsupervised manner. Then, we present the new Bijective Maximum Likelihood(BiMaL) loss that is a generalized form of the Adversarial Entropy Minimization without any assumption about pixel independence. We have evaluated the proposed BiMaL on two domains. The proposed BiMaL approach consistently outperforms the SOTA methods on empirical experiments on "SYNTHIA to Cityscapes", "GTA5 to Cityscapes", and "SYNTHIA to Vistas".
Identifying unexpected objects on roads in semantic segmentation (e.g., identifying dogs on roads) is crucial in safety-critical applications. Existing approaches use images of unexpected objects from external datasets or require additional training (e.g., retraining segmentation networks or training an extra network), which necessitate a non-trivial amount of labor intensity or lengthy inference time. One possible alternative is to use prediction scores of a pre-trained network such as the max logits (i.e., maximum values among classes before the final softmax layer) for detecting such objects. However, the distribution of max logits of each predicted class is significantly different from each other, which degrades the performance of identifying unexpected objects in urban-scene segmentation. To address this issue, we propose a simple yet effective approach that standardizes the max logits in order to align the different distributions and reflect the relative meanings of max logits within each predicted class. Moreover, we consider the local regions from two different perspectives based on the intuition that neighboring pixels share similar semantic information. In contrast to previous approaches, our method does not utilize any external datasets or require additional training, which makes our method widely applicable to existing pre-trained segmentation models. Such a straightforward approach achieves a new state-of-the-art performance on the publicly available Fishyscapes Lost & Found leaderboard with a large margin. Our code is publicly available at this $\href{https://github.com/shjung13/Standardized-max-logits}{link}$.
We study the problem of inferring an object-centric scene representation from a single image, aiming to derive a representation that explains the image formation process, captures the scene's 3D nature, and is learned without supervision. Most existing methods on scene decomposition lack one or more of these characteristics, due to the fundamental challenge in integrating the complex 3D-to-2D image formation process into powerful inference schemes like deep networks. In this paper, we propose unsupervised discovery of Object Radiance Fields (uORF), integrating recent progresses in neural 3D scene representations and rendering with deep inference networks for unsupervised 3D scene decomposition. Trained on multi-view RGB images without annotations, uORF learns to decompose complex scenes with diverse, textured background from a single image. We show that uORF enables novel tasks, such as scene segmentation and editing in 3D, and it performs well on these tasks and on novel view synthesis on three datasets.
3D point-clouds and 2D images are different visual representations of the physical world. While human vision can understand both representations, computer vision models designed for 2D image and 3D point-cloud understanding are quite different. Our paper explores the potential of transferring 2D model architectures and weights to understand 3D point-clouds, by empirically investigating the feasibility of the transfer, the benefits of the transfer, and shedding light on why the transfer works. We discover that we can indeed use the same architecture and pretrained weights of a neural net model to understand both images and point-clouds. Specifically, we transfer the image-pretrained model to a point-cloud model by copying or inflating the weights. We find that finetuning the transformed image-pretrained models (FIP) with minimal efforts -- only on input, output, and normalization layers -- can achieve competitive performance on 3D point-cloud classification, beating a wide range of point-cloud models that adopt task-specific architectures and use a variety of tricks. When finetuning the whole model, the performance improves even further. Meanwhile, FIP improves data efficiency, reaching up to 10.0 top-1 accuracy percent on few-shot classification. It also speeds up the training of point-cloud models by up to 11.1x for a target accuracy (e.g., 90 % accuracy). Lastly, we provide an explanation of the image to point-cloud transfer from the aspect of neural collapse. The code is available at: \url{https://github.com/chenfengxu714/image2point}.
Despite rapid progress in scene segmentation in recent years, 3D segmentation methods are still limited when there is severe occlusion. The key challenge is estimating the segment boundaries of (partially) occluded objects, which are inherently ambiguous when considering only a single frame. In this work, we propose Multihypothesis Segmentation Tracking (MST), a novel method for volumetric segmentation in changing scenes, which allows scene ambiguity to be tracked and our estimates to be adjusted over time as we interact with the scene. Two main innovations allow us to tackle this difficult problem: 1) A novel way to sample possible segmentations from a segmentation tree; and 2) A novel approach to fusing tracking results with multiple segmentation estimates. These methods allow MST to track the segmentation state over time and incorporate new information, such as new objects being revealed. We evaluate our method on several cluttered tabletop environments in simulation and reality. Our results show that MST outperforms baselines in all tested scenes.
Domain adaptation is to transfer the shared knowledge learned from the source domain to a new environment, i.e., target domain. One common practice is to train the model on both labeled source-domain data and unlabeled target-domain data. Yet the learned models are usually biased due to the strong supervision of the source domain. Most researchers adopt the early-stopping strategy to prevent over-fitting, but when to stop training remains a challenging problem since the lack of the target-domain validation set. In this paper, we propose one efficient bootstrapping method, called Adaboost Student, explicitly learning complementary models during training and liberating users from empirical early stopping. Adaboost Student combines the deep model learning with the conventional training strategy, i.e., adaptive boosting, and enables interactions between learned models and the data sampler. We adopt one adaptive data sampler to progressively facilitate learning on hard samples and aggregate "weak" models to prevent over-fitting. Extensive experiments show that (1) Without the need to worry about the stopping time, AdaBoost Student provides one robust solution by efficient complementary model learning during training. (2) AdaBoost Student is orthogonal to most domain adaptation methods, which can be combined with existing approaches to further improve the state-of-the-art performance. We have achieved competitive results on three widely-used scene segmentation domain adaptation benchmarks.
Enhancing the generalization capability of deep neural networks to unseen domains is crucial for safety-critical applications in the real world such as autonomous driving. To address this issue, this paper proposes a novel instance selective whitening loss to improve the robustness of the segmentation networks for unseen domains. Our approach disentangles the domain-specific style and domain-invariant content encoded in higher-order statistics (i.e., feature covariance) of the feature representations and selectively removes only the style information causing domain shift. As shown in Fig. 1, our method provides reasonable predictions for (a) low-illuminated, (b) rainy, and (c) unseen structures. These types of images are not included in the training dataset, where the baseline shows a significant performance drop, contrary to ours. Being simple yet effective, our approach improves the robustness of various backbone networks without additional computational cost. We conduct extensive experiments in urban-scene segmentation and show the superiority of our approach to existing work. Our code is available at https://github.com/shachoi/RobustNet.