In this paper, we introduce a new concept of incorporating factorized flow maps as mid-level representations, for bridging the perception and the control modules in modular learning based robotic frameworks. To investigate the advantages of factorized flow maps and examine their interplay with the other types of mid-level representations, we further develop a configurable framework, along with four different environments that contain both static and dynamic objects, for analyzing the impacts of factorized optical flow maps on the performance of deep reinforcement learning agents. Based on this framework, we report our experimental results on various scenarios, and offer a set of analyses to justify our hypothesis. Finally, we validate flow factorization in real world scenarios.
Temporal sentence grounding aims to localize a target segment in an untrimmed video semantically according to a given sentence query. Most previous works focus on learning frame-level features of each whole frame in the entire video, and directly match them with the textual information. Such frame-level feature extraction leads to the obstacles of these methods in distinguishing ambiguous video frames with complicated contents and subtle appearance differences, thus limiting their performance. In order to differentiate fine-grained appearance similarities among consecutive frames, some state-of-the-art methods additionally employ a detection model like Faster R-CNN to obtain detailed object-level features in each frame for filtering out the redundant background contents. However, these methods suffer from missing motion analysis since the object detection module in Faster R-CNN lacks temporal modeling. To alleviate the above limitations, in this paper, we propose a novel Motion- and Appearance-guided 3D Semantic Reasoning Network (MA3SRN), which incorporates optical-flow-guided motion-aware, detection-based appearance-aware, and 3D-aware object-level features to better reason the spatial-temporal object relations for accurately modelling the activity among consecutive frames. Specifically, we first develop three individual branches for motion, appearance, and 3D encoding separately to learn fine-grained motion-guided, appearance-guided, and 3D-aware object features, respectively. Then, both motion and appearance information from corresponding branches are associated to enhance the 3D-aware features for the final precise grounding. Extensive experiments on three challenging datasets (ActivityNet Caption, Charades-STA and TACoS) demonstrate that the proposed MA3SRN model achieves a new state-of-the-art.
Multi-modal based speech separation has exhibited a specific advantage on isolating the target character in multi-talker noisy environments. Unfortunately, most of current separation strategies prefer a straightforward fusion based on feature learning of each single modality, which is far from sufficient consideration of inter-relationships between modalites. Inspired by learning joint feature representations from audio and visual streams with attention mechanism, in this study, a novel cross-modal fusion strategy is proposed to benefit the whole framework with semantic correlations between different modalities. To further improve audio-visual speech separation, the dense optical flow of lip motion is incorporated to strengthen the robustness of visual representation. The evaluation of the proposed work is performed on two public audio-visual speech separation benchmark datasets. The overall improvement of the performance has demonstrated that the additional motion network effectively enhances the visual representation of the combined lip images and audio signal, as well as outperforming the baseline in terms of all metrics with the proposed cross-modal fusion.
Data is the driving force of machine learning, with the amount and quality of training data often being more important for the performance of a system than architecture and training details. But collecting, processing and annotating real data at scale is difficult, expensive, and frequently raises additional privacy, fairness and legal concerns. Synthetic data is a powerful tool with the potential to address these shortcomings: 1) it is cheap 2) supports rich ground-truth annotations 3) offers full control over data and 4) can circumvent or mitigate problems regarding bias, privacy and licensing. Unfortunately, software tools for effective data generation are less mature than those for architecture design and training, which leads to fragmented generation efforts. To address these problems we introduce Kubric, an open-source Python framework that interfaces with PyBullet and Blender to generate photo-realistic scenes, with rich annotations, and seamlessly scales to large jobs distributed over thousands of machines, and generating TBs of data. We demonstrate the effectiveness of Kubric by presenting a series of 13 different generated datasets for tasks ranging from studying 3D NeRF models to optical flow estimation. We release Kubric, the used assets, all of the generation code, as well as the rendered datasets for reuse and modification.
Mixed reality applications often require virtual objects that are partly occluded by real objects. However, previous research and commercial products have limitations in terms of performance and efficiency. To address these challenges, we propose a novel depth contour occlusion (DCO) algorithm. The proposed method is based on the sensitivity of contour occlusion and a binocular stereoscopic vision device. In this method, a depth contour map is combined with a sparse depth map obtained from a two-stage adaptive filter area stereo matching algorithm and the depth contour information of the objects extracted by a digital image stabilisation optical flow method. We also propose a quadratic optimisation model with three constraints to generate an accurate dense map of the depth contour for high-quality real-virtual occlusion. The whole process is accelerated by GPU. To evaluate the effectiveness of the algorithm, we demonstrate a time con-sumption statistical analysis for each stage of the DCO algorithm execution. To verify the relia-bility of the real-virtual occlusion effect, we conduct an experimental analysis on single-sided, enclosed, and complex occlusions; subsequently, we compare it with the occlusion method without quadratic optimisation. With our GPU implementation for real-time DCO, the evaluation indicates that applying the presented DCO algorithm can enhance the real-time performance and the visual quality of real-virtual occlusion.
When providing the boundary conditions for hydrological flood models and estimating the associated risk, interpolating precipitation at very high temporal resolutions (e.g. 5 minutes) is essential not to miss the cause of flooding in local regions. In this paper, we study optical flow-based interpolation of globally available weather radar images from satellites. The proposed approach uses deep neural networks for the interpolation of multiple video frames, while terrain information is combined with temporarily coarse-grained precipitation radar observation as inputs for self-supervised training. An experiment with the Meteonet radar precipitation dataset for the flood risk simulation in Aude, a department in Southern France (2018), demonstrated the advantage of the proposed method over a linear interpolation baseline, with up to 20% error reduction.
Video Panoptic Segmentation (VPS) aims to generate coherent panoptic segmentation and track the identities of all pixels across video frames. Existing methods predominantly utilize the trained instance embedding to keep the consistency of panoptic segmentation. However, they inevitably struggle to cope with the challenges of small objects, similar appearance but inconsistent identities, occlusion, and strong instance contour deformations. To address these problems, we present HybridTracker, a lightweight and joint tracking model attempting to eliminate the limitations of the single tracker. HybridTracker performs pixel tracker and instance tracker in parallel to obtain the association matrices, which are fused into a matching matrix. In the instance tracker, we design a differentiable matching layer, ensuring the stability of inter-frame matching. In the pixel tracker, we compute the dice coefficient of the same instance of different frames given the estimated optical flow, forming the Intersection Over Union (IoU) matrix. We additionally propose mutual check and temporal consistency constraints during inference to settle the occlusion and contour deformation challenges. Comprehensive experiments show that HybridTracker achieves superior performance than state-of-the-art methods on Cityscapes-VPS and VIPER datasets.
Structured illumination microscopy (SIM) is an optical super-resolution technique that enables live-cell imaging beyond the diffraction limit. Reconstruction of SIM data is prone to artefacts, which becomes problematic when imaging highly dynamic samples because previous methods rely on the assumption that samples are static. We propose a new transformer-based reconstruction method, VSR-SIM, that uses shifted 3-dimensional window multi-head attention in addition to channel attention mechanism to tackle the problem of video super-resolution (VSR) in SIM. The attention mechanisms are found to capture motion in sequences without the need for common motion estimation techniques such as optical flow. We take an approach to training the network that relies solely on simulated data using videos of natural scenery with a model for SIM image formation. We demonstrate a use case enabled by VSR-SIM referred to as rolling SIM imaging, which increases temporal resolution in SIM by a factor of 9. Our method can be applied to any SIM setup enabling precise recordings of dynamic processes in biomedical research with high temporal resolution.
Weakly supervised instance segmentation has gained popularity because it reduces high annotation cost of pixel-level masks required for model training. Recent approaches for weakly supervised instance segmentation detect and segment objects using appearance information obtained from a static image. However, it poses the challenge of identifying objects with a non-discriminatory appearance. In this study, we address this problem by using motion information from image sequences. We propose a two-stream encoder that leverages appearance and motion features extracted from images and optical flows. Additionally, we propose a novel pairwise loss that considers both appearance and motion information to supervise segmentation. We conducted extensive evaluations on the YouTube-VIS 2019 benchmark dataset. Our results demonstrate that the proposed method improves the Average Precision of the state-of-the-art method by 3.1.
Optical flow estimation is a basic task in self-driving and robotics systems, which enables to temporally interpret traffic scenes. Autonomous vehicles clearly benefit from the ultra-wide Field of View (FoV) offered by 360{\deg} panoramic sensors. However, due to the unique imaging process of panoramic cameras, models designed for pinhole images do not directly generalize satisfactorily to 360{\deg} panoramic images. In this paper, we put forward a novel network framework--PanoFlow, to learn optical flow for panoramic images. To overcome the distortions introduced by equirectangular projection in panoramic transformation, we design a Flow Distortion Augmentation (FDA) method, which contains radial flow distortion (FDA-R) or equirectangular flow distortion (FDA-E). We further look into the definition and properties of cyclic optical flow for panoramic videos, and hereby propose a Cyclic Flow Estimation (CFE) method by leveraging the cyclicity of spherical images to infer 360{\deg} optical flow and converting large displacement to relatively small displacement. PanoFlow is applicable to any existing flow estimation method and benefits from the progress of narrow-FoV flow estimation. In addition, we create and release a synthetic panoramic dataset FlowScape based on CARLA to facilitate training and quantitative analysis. PanoFlow achieves state-of-the-art performance on the public OmniFlowNet and the established FlowScape benchmarks. Our proposed approach reduces the End-Point-Error (EPE) on FlowScape by 27.3%. On OmniFlowNet, PanoFlow achieves a 55.5% error reduction from the best published result. We also qualitatively validate our method via a collection vehicle and a public real-world OmniPhotos dataset, indicating strong potential and robustness for real-world navigation applications. Code and dataset are publicly available at https://github.com/MasterHow/PanoFlow.
Machine vision is a critical subsystem for enabling robots to be able to perform a variety of tasks in orchard environments. However, orchards are highly visually complex environments, and computer vision algorithms operating in them must be able to contend with variable lighting conditions and background noise. Past work on enabling deep learning algorithms to operate in these environments has typically required large amounts of hand-labeled data to train a deep neural network or physically controlling the conditions under which the environment is perceived. In this paper, we train a neural network system in simulation only using simulated RGB data and optical flow. This resulting neural network is able to perform foreground segmentation of branches in a busy orchard environment without additional real-world training or using any special setup or equipment beyond a standard camera. Our results show that our system is highly accurate and, when compared to a network using manually labeled RGBD data, achieves significantly more consistent and robust performance across environments that differ from the training set.
Most existing real-time deep models trained with each frame independently may produce inconsistent results across the temporal axis when tested on a video sequence. A few methods take the correlations in the video sequence into account,e.g., by propagating the results to the neighboring frames using optical flow or extracting frame representations using multi-frame information, which may lead to inaccurate results or unbalanced latency. In this work, we focus on improving the temporal consistency without introducing computation overhead in inference. To this end, we perform inference at each frame. Temporal consistency is achieved by learning from video frames with extra constraints during the training phase. introduced for inference. We propose several techniques to learn from the video sequence, including a temporal consistency loss and online/offline knowledge distillation methods. On the task of semantic video segmentation, weighing among accuracy, temporal smoothness, and efficiency, our proposed method outperforms keyframe-based methods and a few baseline methods that are trained with each frame independently, on datasets including Cityscapes, Camvid, and 300VW-Mask. We further apply our training method to video instance segmentation on YouTubeVISand develop an application of portrait matting in video sequences, by segmenting temporally consistent instance-level trimaps across frames. Experiments show superior qualitative and quantitative results. Code is available at: https://git.io/vidseg.
Image-based computational fluid dynamics have long played an important role in leveraging knowledge and understanding of several physical phenomena. In particular, probabilistic computational methods have opened the way to modelling the complex dynamics of systems in purely random turbulent motion. In the field of structural geology, a better understanding of the deformation and stress state both within the salt and the surrounding rocks is of great interest to characterize all kinds of subsurface long-terms energy-storage systems. The objective of this research is to determine the non-linear deformation of the salt boundary over time using a parallelized, stochastic filtering approach from x-ray computed tomography (CT) image time series depicting the evolution of salt structures triggered by gravity and under differential loading. This work represents a first step towards bringing together physical modeling and advanced stochastic image processing methods where model uncertainty is taken into account.
Automatic surgical instrument segmentation of endoscopic images is a crucial building block of many computer-assistance applications for minimally invasive surgery. So far, state-of-the-art approaches completely rely on the availability of a ground-truth supervision signal, obtained via manual annotation, thus expensive to collect at large scale. In this paper, we present FUN-SIS, a Fully-UNsupervised approach for binary Surgical Instrument Segmentation. FUN-SIS trains a per-frame segmentation model on completely unlabelled endoscopic videos, by solely relying on implicit motion information and instrument shape-priors. We define shape-priors as realistic segmentation masks of the instruments, not necessarily coming from the same dataset/domain as the videos. The shape-priors can be collected in various and convenient ways, such as recycling existing annotations from other datasets. We leverage them as part of a novel generative-adversarial approach, allowing to perform unsupervised instrument segmentation of optical-flow images during training. We then use the obtained instrument masks as pseudo-labels in order to train a per-frame segmentation model; to this aim, we develop a learning-from-noisy-labels architecture, designed to extract a clean supervision signal from these pseudo-labels, leveraging their peculiar noise properties. We validate the proposed contributions on three surgical datasets, including the MICCAI 2017 EndoVis Robotic Instrument Segmentation Challenge dataset. The obtained fully-unsupervised results for surgical instrument segmentation are almost on par with the ones of fully-supervised state-of-the-art approaches. This suggests the tremendous potential of the proposed method to leverage the great amount of unlabelled data produced in the context of minimally invasive surgery.
Label distributions in camera-trap images are highly imbalanced and long-tailed, resulting in neural networks tending to be biased towards head-classes that appear frequently. Although long-tail learning has been extremely explored to address data imbalances, few studies have been conducted to consider camera-trap characteristics, such as multi-domain and multi-frame setup. Here, we propose a unified framework and introduce two datasets for long-tailed camera-trap recognition. We first design domain experts, where each expert learns to balance imperfect decision boundaries caused by data imbalances and complement each other to generate domain-balanced decision boundaries. Also, we propose a flow consistency loss to focus on moving objects, expecting class activation maps of multi-frame matches the flow with optical flow maps for input images. Moreover, two long-tailed camera-trap datasets, WCS-LT and DMZ-LT, are introduced to validate our methods. Experimental results show the effectiveness of our framework, and proposed methods outperform previous methods on recessive domain samples.
Depth can provide useful geographical cues for salient object detection (SOD), and has been proven helpful in recent RGB-D SOD methods. However, existing video salient object detection (VSOD) methods only utilize spatiotemporal information and seldom exploit depth information for detection. In this paper, we propose a depth-cooperated trimodal network, called DCTNet for VSOD, which is a pioneering work to incorporate depth information to assist VSOD. To this end, we first generate depth from RGB frames, and then propose an approach to treat the three modalities unequally. Specifically, a multi-modal attention module (MAM) is designed to model multi-modal long-range dependencies between the main modality (RGB) and the two auxiliary modalities (depth, optical flow). We also introduce a refinement fusion module (RFM) to suppress noises in each modality and select useful information dynamically for further feature refinement. Lastly, a progressive fusion strategy is adopted after the refined features to achieve final cross-modal fusion. Experiments on five benchmark datasets demonstrate the superiority of our depth-cooperated model against 12 state-of-the-art methods, and the necessity of depth is also validated.
This work aims at generating captions for soccer videos using deep learning. In this context, this paper introduces a dataset, model, and triple-level evaluation. The dataset consists of 22k caption-clip pairs and three visual features (images, optical flow, inpainting) for ~500 hours of \emph{SoccerNet} videos. The model is divided into three parts: a transformer learns language, ConvNets learn vision, and a fusion of linguistic and visual features generates captions. The paper suggests evaluating generated captions at three levels: syntax (the commonly used evaluation metrics such as BLEU-score and CIDEr), meaning (the quality of descriptions for a domain expert), and corpus (the diversity of generated captions). The paper shows that the diversity of generated captions has improved (from 0.07 reaching 0.18) with semantics-related losses that prioritize selected words. Semantics-related losses and the utilization of more visual features (optical flow, inpainting) improved the normalized captioning score by 28\%. The web page of this work: https://sites.google.com/view/soccercaptioning}{https://sites.google.com/view/soccercaptioning
We present a frame interpolation algorithm that synthesizes multiple intermediate frames from two input images with large in-between motion. Recent methods use multiple networks to estimate optical flow or depth and a separate network dedicated to frame synthesis. This is often complex and requires scarce optical flow or depth ground-truth. In this work, we present a single unified network, distinguished by a multi-scale feature extractor that shares weights at all scales, and is trainable from frames alone. To synthesize crisp and pleasing frames, we propose to optimize our network with the Gram matrix loss that measures the correlation difference between feature maps. Our approach outperforms state-of-the-art methods on the Xiph large motion benchmark. We also achieve higher scores on Vimeo-90K, Middlebury and UCF101, when comparing to methods that use perceptual losses. We study the effect of weight sharing and of training with datasets of increasing motion range. Finally, we demonstrate our model's effectiveness in synthesizing high quality and temporally coherent videos on a challenging near-duplicate photos dataset. Codes and pre-trained models are available at https://film-net.github.io.
Contactless monitoring using thermal imaging has become increasingly proposed to monitor patient deterioration in hospital, most recently to detect fevers and infections during the COVID-19 pandemic. In this letter, we propose a novel method to estimate patient motion and observe clinical workload using a similar technical setup but combined with open source object detection algorithms (YOLOv4) and optical flow. Patient motion estimation was used to approximate patient agitation and sedation, while worker motion was used as a surrogate for caregiver workload. Performance was illustrated by comparing over 32000 frames from videos of patients recorded in an Intensive Care Unit, to clinical agitation scores recorded by clinical workers.
Estimating per-pixel motion between video frames, known as optical flow, is a long-standing problem in video understanding and analysis. Most contemporary optical flow techniques largely focus on addressing the cross-image matching with feature similarity, with few methods considering how to explicitly reason over the given scene for achieving a holistic motion understanding. In this work, taking a fresh perspective, we introduce a novel graph-based approach, called adaptive graph reasoning for optical flow (AGFlow), to emphasize the value of scene/context information in optical flow. Our key idea is to decouple the context reasoning from the matching procedure, and exploit scene information to effectively assist motion estimation by learning to reason over the adaptive graph. The proposed AGFlow can effectively exploit the context information and incorporate it within the matching procedure, producing more robust and accurate results. On both Sintel clean and final passes, our AGFlow achieves the best accuracy with EPE of 1.43 and 2.47 pixels, outperforming state-of-the-art approaches by 11.2% and 13.6%, respectively.