1.MeMViT: Memory-Augmented Multiscale Vision Transformer for Efficient Long-Term Video Recognition ⬇️
While today's video recognition systems parse snapshots or short clips accurately, they cannot connect the dots and reason across a longer range of time yet. Most existing video architectures can only process <5 seconds of a video without hitting the computation or memory bottlenecks.
In this paper, we propose a new strategy to overcome this challenge. Instead of trying to process more frames at once like most existing methods, we propose to process videos in an online fashion and cache "memory" at each iteration. Through the memory, the model can reference prior context for long-term modeling, with only a marginal cost. Based on this idea, we build MeMViT, a Memory-augmented Multiscale Vision Transformer, that has a temporal support 30x longer than existing models with only 4.5% more compute; traditional methods need >3,000% more compute to do the same. On a wide range of settings, the increased temporal support enabled by MeMViT brings large gains in recognition accuracy consistently. MeMViT obtains state-of-the-art results on the AVA, EPIC-Kitchens-100 action classification, and action anticipation datasets. Code and models will be made publicly available.
2.Learning Pixel Trajectories with Multiscale Contrastive Random Walks ⬇️
A range of video modeling tasks, from optical flow to multiple object tracking, share the same fundamental challenge: establishing space-time correspondence. Yet, approaches that dominate each space differ. We take a step towards bridging this gap by extending the recent contrastive random walk formulation to much denser, pixel-level space-time graphs. The main contribution is introducing hierarchy into the search problem by computing the transition matrix between two frames in a coarse-to-fine manner, forming a multiscale contrastive random walk when extended in time. This establishes a unified technique for self-supervised learning of optical flow, keypoint tracking, and video object segmentation. Experiments demonstrate that, for each of these tasks, the unified model achieves performance competitive with strong self-supervised approaches specific to that task. Project site: this https URL
3.Omnivore: A Single Model for Many Visual Modalities ⬇️
Prior work has studied different visual modalities in isolation and developed separate architectures for recognition of images, videos, and 3D data. Instead, in this paper, we propose a single model which excels at classifying images, videos, and single-view 3D data using exactly the same model parameters. Our 'Omnivore' model leverages the flexibility of transformer-based architectures and is trained jointly on classification tasks from different modalities. Omnivore is simple to train, uses off-the-shelf standard datasets, and performs at-par or better than modality-specific models of the same size. A single Omnivore model obtains 86.0% on ImageNet, 84.1% on Kinetics, and 67.1% on SUN RGB-D. After finetuning, our models outperform prior work on a variety of vision tasks and generalize across modalities. Omnivore's shared visual representation naturally enables cross-modal recognition without access to correspondences between modalities. We hope our results motivate researchers to model visual modalities together.
4.Revisiting Weakly Supervised Pre-Training of Visual Perception Models ⬇️
Model pre-training is a cornerstone of modern visual recognition systems. Although fully supervised pre-training on datasets like ImageNet is still the de-facto standard, recent studies suggest that large-scale weakly supervised pre-training can outperform fully supervised approaches. This paper revisits weakly-supervised pre-training of models using hashtag supervision with modern versions of residual networks and the largest-ever dataset of images and corresponding hashtags. We study the performance of the resulting models in various transfer-learning settings including zero-shot transfer. We also compare our models with those obtained via large-scale self-supervised learning. We find our weakly-supervised models to be very competitive across all settings, and find they substantially outperform their self-supervised counterparts. We also include an investigation into whether our models learned potentially troubling associations or stereotypes. Overall, our results provide a compelling argument for the use of weakly supervised learning in the development of visual recognition systems. Our models, Supervised Weakly through hashtAGs (SWAG), are available publicly.
5.Stitch it in Time: GAN-Based Facial Editing of Real Videos ⬇️
The ability of Generative Adversarial Networks to encode rich semantics within their latent space has been widely adopted for facial image editing. However, replicating their success with videos has proven challenging. Sets of high-quality facial videos are lacking, and working with videos introduces a fundamental barrier to overcome - temporal coherency. We propose that this barrier is largely artificial. The source video is already temporally coherent, and deviations from this state arise in part due to careless treatment of individual components in the editing pipeline. We leverage the natural alignment of StyleGAN and the tendency of neural networks to learn low frequency functions, and demonstrate that they provide a strongly consistent prior. We draw on these insights and propose a framework for semantic editing of faces in videos, demonstrating significant improvements over the current state-of-the-art. Our method produces meaningful face manipulations, maintains a higher degree of temporal consistency, and can be applied to challenging, high quality, talking head videos which current methods struggle with.
6.DIVA-DAF: A Deep Learning Framework for Historical Document Image Analysis ⬇️
In this paper, we introduce a new deep learning framework called DIVA-DAF. We have developed this framework to support our research on historical document image analysis tasks and to develop techniques to reduce the need for manually-labeled ground truth. We want to apply self-supervised learning techniques and use different kinds of training data. Our new framework aids us in performing rapid prototyping and reproducible experiments. We present a first semantic segmentation experiment on DIVA-HisDB using our framework, achieving state-of-the-art results. The DIVA-DAF framework is open-source, and we encourage other research groups to use it for their experiments.
7.End-to-end Generative Pretraining for Multimodal Video Captioning ⬇️
Recent video and language pretraining frameworks lack the ability to generate sentences. We present Multimodal Video Generative Pretraining (MV-GPT), a new pretraining framework for learning from unlabelled videos which can be effectively used for generative tasks such as multimodal video captioning. Unlike recent video-language pretraining frameworks, our framework trains both a multimodal video encoder and a sentence decoder jointly. To overcome the lack of captions in unlabelled videos, we leverage the future utterance as an additional text source and propose a bidirectional generation objective -- we generate future utterances given the present mulitmodal context, and also the present utterance given future observations. With this objective, we train an encoder-decoder model end-to-end to generate a caption from raw pixels and transcribed speech directly. Our model achieves state-of-the-art performance for multimodal video captioning on four standard benchmarks, as well as for other video understanding tasks such as VideoQA, video retrieval and action classification.
8.Watermarking Pre-trained Encoders in Contrastive Learning ⬇️
Contrastive learning has become a popular technique to pre-train image encoders, which could be used to build various downstream classification models in an efficient way. This process requires a large amount of data and computation resources. Hence, the pre-trained encoders are an important intellectual property that needs to be carefully protected. It is challenging to migrate existing watermarking techniques from the classification tasks to the contrastive learning scenario, as the owner of the encoder lacks the knowledge of the downstream tasks which will be developed from the encoder in the future. We propose the \textit{first} watermarking methodology for the pre-trained encoders. We introduce a task-agnostic loss function to effectively embed into the encoder a backdoor as the watermark. This backdoor can still exist in any downstream models transferred from the encoder. Extensive evaluations over different contrastive learning algorithms, datasets, and downstream tasks indicate our watermarks exhibit high effectiveness and robustness against different adversarial operations.
9.CP-Net: Contour-Perturbed Reconstruction Network for Self-Supervised Point Cloud Learning ⬇️
Self-supervised learning has not been fully explored for point cloud analysis. Current frameworks are mainly based on point cloud reconstruction. Given only 3D coordinates, such approaches tend to learn local geometric structures and contours, while failing in understanding high level semantic content. Consequently, they achieve unsatisfactory performance in downstream tasks such as classification, segmentation, etc. To fill this gap, we propose a generic Contour-Perturbed Reconstruction Network (CP-Net), which can effectively guide self-supervised reconstruction to learn semantic content in the point cloud, and thus promote discriminative power of point cloud representation. First, we introduce a concise contour-perturbed augmentation module for point cloud reconstruction. With guidance of geometry disentangling, we divide point cloud into contour and content components. Subsequently, we perturb the contour components and preserve the content components on the point cloud. As a result, self supervisor can effectively focus on semantic content, by reconstructing the original point cloud from such perturbed one. Second, we use this perturbed reconstruction as an assistant branch, to guide the learning of basic reconstruction branch via a distinct dual-branch consistency loss. In this case, our CP-Net not only captures structural contour but also learn semantic content for discriminative downstream tasks. Finally, we perform extensive experiments on a number of point cloud benchmarks. Part segmentation results demonstrate that our CP-Net (81.5% of mIoU) outperforms the previous self-supervised models, and narrows the gap with the fully-supervised methods. For classification, we get a competitive result with the fully-supervised methods on ModelNet40 (92.5% accuracy) and ScanObjectNN (87.9% accuracy). The codes and models will be released afterwards.
10.HumanIBR: High Quality Image-based Rendering of Challenging Human Performers using Sparse Views ⬇️
In this paper, we introduce HumanIBR, a method that addresses the challenge of novel view rendering of human performers that wear clothes with complex patterns using a sparse set of camera views. Some recent works have achieved remarkable rendering quality on humans that wear pure clothes using sparse views, but if the clothes have complex color patterns, the rendering quality is still very low. To this end, the proposed HumanIBR uses a human reconstruction net with pixel-aligned spatial transformer and a render net that uses geometry-guided pixel-wise feature integration to achieve to goal of high quality human reconstruction and rendering. The designed pixel-aligned spatial transformer calculates the correlations between the input views, producing human reconstruction results with high-frequency details presented in the input views. Based on the reconstruction, the geometry-guided pixel-wise visibility reasoning provides a guidance for multi-view feature integration, enabling the render net to render high quality images on novel views. Unlike previous neural rendering works that always need to train or fine-tune a separate network for each scene or human, our method is a general framework that is able to generalize to novel humans. Experiments show that our approach outperforms all the prior general or human-specific works on both synthetic data and real-world data.
11.WPPNets: Unsupervised CNN Training with Wasserstein Patch Priors for Image Superresolution ⬇️
We introduce WPPNets, which are CNNs trained by a new unsupervised loss function for image superresolution of materials microstructures. Instead of requiring access to a large database of registered high- and low-resolution images, we only assume to know a large database of low resolution images, the forward operator and one high-resolution reference image. Then, we propose a loss function based on the Wasserstein patch prior which measures the Wasserstein-2 distance between the patch distributions of the predictions and the reference image. We demonstrate by numerical examples that WPPNets outperform other methods with similar assumptions. In particular, we show that WPPNets are much more stable under inaccurate knowledge or perturbations of the forward operator. This enables us to use them in real-world applications, where neither a large database of registered data nor the exact forward operator are given.
12.SPAMs: Structured Implicit Parametric Models ⬇️
Parametric 3D models have formed a fundamental role in modeling deformable objects, such as human bodies, faces, and hands; however, the construction of such parametric models requires significant manual intervention and domain expertise. Recently, neural implicit 3D representations have shown great expressibility in capturing 3D shape geometry. We observe that deformable object motion is often semantically structured, and thus propose to learn Structured-implicit PArametric Models (SPAMs) as a deformable object representation that structurally decomposes non-rigid object motion into part-based disentangled representations of shape and pose, with each being represented by deep implicit functions. This enables a structured characterization of object movement, with part decomposition characterizing a lower-dimensional space in which we can establish coarse motion correspondence. In particular, we can leverage the part decompositions at test time to fit to new depth sequences of unobserved shapes, by establishing part correspondences between the input observation and our learned part spaces; this guides a robust joint optimization between the shape and pose of all parts, even under dramatic motion sequences. Experiments demonstrate that our part-aware shape and pose understanding lead to state-of-the-art performance in reconstruction and tracking of depth sequences of complex deforming object motion. We plan to release models to the public at this https URL.
13.GeoFill: Reference-Based Image Inpainting of Scenes with Complex Geometry ⬇️
Reference-guided image inpainting restores image pixels by leveraging the content from another reference image. The previous state-of-the-art, TransFill, warps the source image with multiple homographies, and fuses them together for hole filling. Inspired by structure from motion pipelines and recent progress in monocular depth estimation, we propose a more principled approach that does not require heuristic planar assumptions. We leverage a monocular depth estimate and predict relative pose between cameras, then align the reference image to the target by a differentiable 3D reprojection and a joint optimization of relative pose and depth map scale and offset. Our approach achieves state-of-the-art performance on both RealEstate10K and MannequinChallenge dataset with large baselines, complex geometry and extreme camera motions. We experimentally verify our approach is also better at handling large holes.
14.Deep Unsupervised Contrastive Hashing for Large-Scale Cross-Modal Text-Image Retrieval in Remote Sensing ⬇️
Due to the availability of large-scale multi-modal data (e.g., satellite images acquired by different sensors, text sentences, etc) archives, the development of cross-modal retrieval systems that can search and retrieve semantically relevant data across different modalities based on a query in any modality has attracted great attention in RS. In this paper, we focus our attention on cross-modal text-image retrieval, where queries from one modality (e.g., text) can be matched to archive entries from another (e.g., image). Most of the existing cross-modal text-image retrieval systems require a high number of labeled training samples and also do not allow fast and memory-efficient retrieval due to their intrinsic characteristics. These issues limit the applicability of the existing cross-modal retrieval systems for large-scale applications in RS. To address this problem, in this paper we introduce a novel deep unsupervised cross-modal contrastive hashing (DUCH) method for RS text-image retrieval. The proposed DUCH is made up of two main modules: 1) feature extraction module (which extracts deep representations of the text-image modalities); and 2) hashing module (which learns to generate cross-modal binary hash codes from the extracted representations). Within the hashing module, we introduce a novel multi-objective loss function including: i) contrastive objectives that enable similarity preservation in both intra- and inter-modal similarities; ii) an adversarial objective that is enforced across two modalities for cross-modal representation consistency; iii) binarization objectives for generating representative hash codes. Experimental results show that the proposed DUCH outperforms state-of-the-art unsupervised cross-modal hashing methods on two multi-modal (image and text) benchmark archives in RS. Our code is publicly available at this https URL.
15.A Computational Model for Machine Thinking ⬇️
A machine thinking model is proposed in this report based on recent advances of computer vision and the recent results of neuroscience devoted to brain understanding. We deliver the result of machine thinking in the form of sentences of natural-language or drawn sketches either informative or decisional. This result is obtained from a reasoning performed on new acquired data and memorized data.
16.What can we learn from misclassified ImageNet images? ⬇️
Understanding the patterns of misclassified ImageNet images is particularly important, as it could guide us to design deep neural networks (DNN) that generalize better. However, the richness of ImageNet imposes difficulties for researchers to visually find any useful patterns of misclassification. Here, to help find these patterns, we propose "Superclassing ImageNet dataset". It is a subset of ImageNet which consists of 10 superclasses, each containing 7-116 related subclasses (e.g., 52 bird types, 116 dog types). By training neural networks on this dataset, we found that: (i) Misclassifications are rarely across superclasses, but mainly among subclasses within a superclass. (ii) Ensemble networks trained each only on subclasses of a given superclass perform better than the same network trained on all subclasses of all superclasses. Hence, we propose a two-stage Super-Sub framework, and demonstrate that: (i) The framework improves overall classification performance by 3.3%, by first inferring a superclass using a generalist superclass-level network, and then using a specialized network for final subclass-level classification. (ii) Although the total parameter storage cost increases to a factor N+1 for N superclasses compared to using a single network, with finetuning, delta and quantization aware training techniques this can be reduced to 0.2N+1. Another advantage of this efficient implementation is that the memory cost on the GPU during inference is equivalent to using only one network. The reason is we initiate each subclass-level network through addition of small parameter variations (deltas) to the superclass-level network. (iii) Finally, our framework promises to be more scalable and generalizable than the common alternative of simply scaling up a vanilla network in size, since very large networks often suffer from overfitting and gradient vanishing.
17.AirPose: Multi-View Fusion Network for Aerial 3D Human Pose and Shape Estimation ⬇️
In this letter, we present a novel markerless 3D human motion capture (MoCap) system for unstructured, outdoor environments that uses a team of autonomous unmanned aerial vehicles (UAVs) with on-board RGB cameras and computation. Existing methods are limited by calibrated cameras and off-line processing. Thus, we present the first method (AirPose) to estimate human pose and shape using images captured by multiple extrinsically uncalibrated flying cameras. AirPose itself calibrates the cameras relative to the person instead of relying on any pre-calibration. It uses distributed neural networks running on each UAV that communicate viewpoint-independent information with each other about the person (i.e., their 3D shape and articulated pose). The person's shape and pose are parameterized using the SMPL-X body model, resulting in a compact representation, that minimizes communication between the UAVs. The network is trained using synthetic images of realistic virtual environments, and fine-tuned on a small set of real images. We also introduce an optimization-based post-processing method (AirPose$^{+}$) for offline applications that require higher MoCap quality. We make our method's code and data available for research at this https URL. A video describing the approach and results is available at this https URL.
18.The Elements of Temporal Sentence Grounding in Videos: A Survey and Future Directions ⬇️
Temporal sentence grounding in videos (TSGV), a.k.a., natural language video localization (NLVL) or video moment retrieval (VMR), aims to retrieve a temporal moment that semantically corresponds to a language query from an untrimmed video. Connecting computer vision and natural language, TSGV has drawn significant attention from researchers in both communities. This survey attempts to provide a summary of fundamental concepts in TSGV and current research status, as well as future research directions. As the background, we present a common structure of functional components in TSGV, in a tutorial style: from feature extraction from raw video and language query, to answer prediction of the target moment. Then we review the techniques for multimodal understanding and interaction, which is the key focus of TSGV for effective alignment between the two modalities. We construct a taxonomy of TSGV techniques and elaborate methods in different categories with their strengths and weaknesses. Lastly, we discuss issues with the current TSGV research and share our insights about promising research directions.
19.Predicting Vegetation Stratum Occupancy from Airborne LiDAR Data with Deep Learning ⬇️
We propose a new deep learning-based method for estimating the occupancy of vegetation strata from airborne 3D LiDAR point clouds. Our model predicts rasterized occupancy maps for three vegetation strata corresponding to lower, medium, and higher cover. Our weakly-supervised training scheme allows our network to only be supervised with vegetation occupancy values aggregated over cylindrical plots containing thousands of points. Such ground truth is easier to produce than pixel-wise or point-wise annotations. Our method outperforms handcrafted and deep learning baselines in terms of precision by up to 30%, while simultaneously providing visual and interpretable predictions. We provide an open-source implementation along with a dataset of 199 agricultural plots to train and evaluate weakly supervised occupancy regression algorithms.
20.TerViT: An Efficient Ternary Vision Transformer ⬇️
Vision transformers (ViTs) have demonstrated great potential in various visual tasks, but suffer from expensive computational and memory cost problems when deployed on resource-constrained devices. In this paper, we introduce a ternary vision transformer (TerViT) to ternarize the weights in ViTs, which are challenged by the large loss surface gap between real-valued and ternary parameters. To address the issue, we introduce a progressive training scheme by first training 8-bit transformers and then TerViT, and achieve a better optimization than conventional methods. Furthermore, we introduce channel-wise ternarization, by partitioning each matrix to different channels, each of which is with an unique distribution and ternarization interval. We apply our methods to popular DeiT and Swin backbones, and extensive results show that we can achieve competitive performance. For example, TerViT can quantize Swin-S to 13.1MB model size while achieving above 79% Top-1 accuracy on ImageNet dataset.
21.Lightweight Salient Object Detection in Optical Remote Sensing Images via Feature Correlation ⬇️
Salient object detection in optical remote sensing images (ORSI-SOD) has been widely explored for understanding ORSIs. However, previous methods focus mainly on improving the detection accuracy while neglecting the cost in memory and computation, which may hinder their real-world applications. In this paper, we propose a novel lightweight ORSI-SOD solution, named CorrNet, to address these issues. In CorrNet, we first lighten the backbone (VGG-16) and build a lightweight subnet for feature extraction. Then, following the coarse-to-fine strategy, we generate an initial coarse saliency map from high-level semantic features in a Correlation Module (CorrM). The coarse saliency map serves as the location guidance for low-level features. In CorrM, we mine the object location information between high-level semantic features through the cross-layer correlation operation. Finally, based on low-level detailed features, we refine the coarse saliency map in the refinement subnet equipped with Dense Lightweight Refinement Blocks, and produce the final fine saliency map. By reducing the parameters and computations of each component, CorrNet ends up having only 4.09M parameters and running with 21.09G FLOPs. Experimental results on two public datasets demonstrate that our lightweight CorrNet achieves competitive or even better performance compared with 26 state-of-the-art methods (including 16 large CNN-based methods and 2 lightweight methods), and meanwhile enjoys the clear memory and run time efficiency. The code and results of our method are available at this https URL.
22.Domain Generalization via Frequency-based Feature Disentanglement and Interaction ⬇️
Data out-of-distribution is a meta-challenge for all statistical learning algorithms that strongly rely on the i.i.d. assumption. It leads to unavoidable labor costs and confidence crises in realistic applications. For that, domain generalization aims at mining domain-irrelevant knowledge from multiple source domains that can generalize to unseen target domains with unknown distributions. In this paper, leveraging the image frequency domain, we uniquely work with two key observations: (i) the high-frequency information of images depict object edge structure, which is naturally consistent across different domains, and (ii) the low-frequency component retains object smooth structure but are much more domain-specific. Motivated by these insights, we introduce (i) an encoder-decoder structure for high-frequency and low-frequency feature disentangling, (ii) an information interaction mechanism that ensures helpful knowledge from both two parts can cooperate effectively, and (iii) a novel data augmentation technique that works on the frequency domain for encouraging robustness of the network. The proposed method obtains state-of-the-art results on three widely used domain generalization benchmarks (Digit-DG, Office-Home, and PACS).
23.A Joint Morphological Profiles and Patch Tensor Change Detection for Hyperspectral Imagery ⬇️
Multi-temporal hyperspectral images can be used to detect changed information, which has gradually attracted researchers' attention. However, traditional change detection algorithms have not deeply explored the relevance of spatial and spectral changed features, which leads to low detection accuracy. To better excavate both spectral and spatial information of changed features, a joint morphology and patch-tensor change detection (JMPT) method is proposed. Initially, a patch-based tensor strategy is adopted to exploit similar property of spatial structure, where the non-overlapping local patch image is reshaped into a new tensor cube, and then three-order Tucker decompositon and image reconstruction strategies are adopted to obtain more robust multi-temporal hyperspectral datasets. Meanwhile, multiple morphological profiles including max-tree and min-tree are applied to extract different attributes of multi-temporal images. Finally, these results are fused to general a final change detection map. Experiments conducted on two real hyperspectral datasets demonstrate that the proposed detector achieves better detection performance.
24.PRMI: A Dataset of Minirhizotron Images for Diverse Plant Root Study ⬇️
Understanding a plant's root system architecture (RSA) is crucial for a variety of plant science problem domains including sustainability and climate adaptation. Minirhizotron (MR) technology is a widely-used approach for phenotyping RSA non-destructively by capturing root imagery over time. Precisely segmenting roots from the soil in MR imagery is a critical step in studying RSA features. In this paper, we introduce a large-scale dataset of plant root images captured by MR technology. In total, there are over 72K RGB root images across six different species including cotton, papaya, peanut, sesame, sunflower, and switchgrass in the dataset. The images span a variety of conditions including varied root age, root structures, soil types, and depths under the soil surface. All of the images have been annotated with weak image-level labels indicating whether each image contains roots or not. The image-level labels can be used to support weakly supervised learning in plant root segmentation tasks. In addition, 63K images have been manually annotated to generate pixel-level binary masks indicating whether each pixel corresponds to root or not. These pixel-level binary masks can be used as ground truth for supervised learning in semantic segmentation tasks. By introducing this dataset, we aim to facilitate the automatic segmentation of roots and the research of RSA with deep learning and other image analysis algorithms.
25.CELESTIAL: Classification Enabled via Labelless Embeddings with Self-supervised Telescope Image Analysis Learning ⬇️
A common class of problems in remote sensing is scene classification, a fundamentally important task for natural hazards identification, geographic image retrieval, and environment monitoring. Recent developments in this field rely label-dependent supervised learning techniques which is antithetical to the 35 petabytes of unlabelled satellite imagery in NASA GIBS. To solve this problem, we establish CELESTIAL-a self-supervised learning pipeline for effectively leveraging sparsely-labeled satellite imagery. This pipeline successfully adapts SimCLR, an algorithm that first learns image representations on unlabelled data and then fine-tunes this knowledge on the provided labels. Our results show CELESTIAL requires only a third of the labels that the supervised method needs to attain the same accuracy on an experimental dataset. The first unsupervised tier can enable applications such as reverse image search for NASA Worldview (i.e. searching similar atmospheric phenomenon over years of unlabelled data with minimal samples) and the second supervised tier can lower the necessity of expensive data annotation significantly. In the future, we hope we can generalize the CELESTIAL pipeline to other data types, algorithms, and applications.
26.Self-supervised Video Representation Learning with Cascade Positive Retrieval ⬇️
Self-supervised video representation learning has been shown to effectively improve downstream tasks such as video retrieval and action recognition. In this paper, we present the Cascade Positive Retrieval (CPR) that successively mines positive examples w.r.t. the query for contrastive learning in a cascade of stages. Specifically, CPR exploits multiple views of a query example in different modalities, where an alternative view may help find another positive example dissimilar in the query view. We explore the effects of possible CPR configurations in ablations including the number of mining stages, the top similar example selection ratio in each stage, and progressive training with an incremental number of the final Top-k selection. The overall mining quality is measured to reflect the recall across training set classes. CPR reaches a median class mining recall of 83.3%, outperforming previous work by 5.5%. Implementation-wise, CPR is complementary to pretext tasks and can be easily applied to previous work. In the evaluation of pretraining on UCF101, CPR consistently improves existing work and even achieves state-of-the-art R@1 of 56.7% and 24.4% in video retrieval as well as 83.8% and 54.8% in action recognition on UCF101 and HMDB51. For transfer from large video dataset Kinetics400 to UCF101 and HDMB, CPR benefits existing work, showing competitive Top-1 accuracies of 85.1% and 57.4% despite pretraining at a lower resolution and frame sampling rate. The code will be released soon for reproducing the results. The code is available at this https URL.
27.GASCN: Graph Attention Shape Completion Network ⬇️
Shape completion, the problem of inferring the complete geometry of an object given a partial point cloud, is an important problem in robotics and computer vision. This paper proposes the Graph Attention Shape Completion Network (GASCN), a novel neural network model that solves this problem. This model combines a graph-based model for encoding local point cloud information with an MLP-based architecture for encoding global information. For each completed point, our model infers the normal and extent of the local surface patch which is used to produce dense yet precise shape completions. We report experiments that demonstrate that GASCN outperforms standard shape completion methods on a standard benchmark drawn from the Shapenet dataset.
28.Experimental Large-Scale Jet Flames' Geometrical Features Extraction for Risk Management Using Infrared Images and Deep Learning Segmentation Methods ⬇️
Jet fires are relatively small and have the least severe effects among the diverse fire accidents that can occur in industrial plants; however, they are usually involved in a process known as the domino effect, that leads to more severe events, such as explosions or the initiation of another fire, making the analysis of such fires an important part of risk analysis. This research work explores the application of deep learning models in an alternative approach that uses the semantic segmentation of jet fires flames to extract main geometrical attributes, relevant for fire risk assessments. A comparison is made between traditional image processing methods and some state-of-the-art deep learning models. It is found that the best approach is a deep learning architecture known as UNet, along with its two improvements, Attention UNet and UNet++. The models are then used to segment a group of vertical jet flames of varying pipe outlet diameters to extract their main geometrical characteristics. Attention UNet obtained the best general performance in the approximation of both height and area of the flames, while also showing a statistically significant difference between it and UNet++. UNet obtained the best overall performance for the approximation of the lift-off distances; however, there is not enough data to prove a statistically significant difference between Attention UNet and UNet++. The only instance where UNet++ outperformed the other models, was while obtaining the lift-off distances of the jet flames with 0.01275 m pipe outlet diameter. In general, the explored models show good agreement between the experimental and predicted values for relatively large turbulent propane jet flames, released in sonic and subsonic regimes; thus, making these radiation zones segmentation models, a suitable approach for different jet flame risk management scenarios.
29.Estimating Egocentric 3D Human Pose in the Wild with External Weak Supervision ⬇️
Egocentric 3D human pose estimation with a single fisheye camera has drawn a significant amount of attention recently. However, existing methods struggle with pose estimation from in-the-wild images, because they can only be trained on synthetic data due to the unavailability of large-scale in-the-wild egocentric datasets. Furthermore, these methods easily fail when the body parts are occluded by or interacting with the surrounding scene. To address the shortage of in-the-wild data, we collect a large-scale in-the-wild egocentric dataset called Egocentric Poses in the Wild (EgoPW). This dataset is captured by a head-mounted fisheye camera and an auxiliary external camera, which provides an additional observation of the human body from a third-person perspective during training. We present a new egocentric pose estimation method, which can be trained on the new dataset with weak external supervision. Specifically, we first generate pseudo labels for the EgoPW dataset with a spatio-temporal optimization method by incorporating the external-view supervision. The pseudo labels are then used to train an egocentric pose estimation network. To facilitate the network training, we propose a novel learning strategy to supervise the egocentric features with the high-quality features extracted by a pretrained external-view pose estimation model. The experiments show that our method predicts accurate 3D poses from a single in-the-wild egocentric image and outperforms the state-of-the-art methods both quantitatively and qualitatively.
30.Learning-by-Novel-View-Synthesis for Full-Face Appearance-based 3D Gaze Estimation ⬇️
Despite recent advances in appearance-based gaze estimation techniques, the need for training data that covers the target head pose and gaze distribution remains a crucial challenge for practical deployment. This work examines a novel approach for synthesizing gaze estimation training data based on monocular 3D face reconstruction. Unlike prior works using multi-view reconstruction, photo-realistic CG models, or generative neural networks, our approach can manipulate and extend the head pose range of existing training data without any additional requirements. We introduce a projective matching procedure to align the reconstructed 3D facial mesh to the camera coordinate system and synthesize face images with accurate gaze labels. We also propose a mask-guided gaze estimation model and data augmentation strategies to further improve the estimation accuracy by taking advantage of the synthetic training data. Experiments using multiple public datasets show that our approach can significantly improve the estimation performance on challenging cross-dataset settings with non-overlapping gaze distributions.
31.The Role of Facial Expressions and Emotion in ASL ⬇️
There is little prior work on quantifying the relationships between facial expressions and emotionality in American Sign Language. In this final report, we provide two methods for studying these relationships through probability and prediction. Using a large corpus of natural signing manually annotated with facial features paired with lexical emotion datasets, we find that there exist many relationships between emotionality and the face, and that a simple classifier can predict what someone is saying in terms of broad emotional categories only by looking at the face.
32.Enhanced Performance of Pre-Trained Networks by Matched Augmentation Distributions ⬇️
There exists a distribution discrepancy between training and testing, in the way images are fed to modern CNNs. Recent work tried to bridge this gap either by fine-tuning or re-training the network at different resolutions. However re-training a network is rarely cheap and not always viable. To this end, we propose a simple solution to address the train-test distributional shift and enhance the performance of pre-trained models -- which commonly ship as a package with deep learning platforms \eg, PyTorch. Specifically, we demonstrate that running inference on the center crop of an image is not always the best as important discriminatory information may be cropped-off. Instead we propose to combine results for multiple random crops for a test image. This not only matches the train time augmentation but also provides the full coverage of the input image. We explore combining representation of random crops through averaging at different levels \ie, deep feature level, logit level, and softmax level. We demonstrate that, for various families of modern deep networks, such averaging results in better validation accuracy compared to using a single central crop per image. The softmax averaging results in the best performance for various pre-trained networks without requiring any re-training or fine-tuning whatsoever. On modern GPUs with batch processing, the paper's approach to inference of pre-trained networks, is essentially free as all images in a batch can all be processed at once.
33.Modeling and hexahedral meshing of arterial networks from centerlines ⬇️
Computational fluid dynamics (CFD) simulation provides valuable information on blood flow from the vascular geometry. However, it requires to extract accurate models of arteries from low resolution medical images, which remains challenging. Centerline-based representation is widely used to model large vascular networks with small vessels, as it enables manual editing and encodes the topological information. In this work, we propose an automatic method to generate an hexahedral mesh suitable for CFD directly from centerlines. The proposed method is an improvement of the state-of-the-art in terms of robustness, mesh quality and reproductibility.
Both the modeling and meshing tasks are addressed. A new vessel model based on penalized splines is proposed to overcome the limitations inherent to the centerline representation, such as noise and sparsity. Bifurcations are reconstructed using a physiologically accurate parametric model that we extended to planar n-furcations. Finally, a volume mesh with structured, hexahedral and flow oriented cells is produced from the proposed vascular network model.
The proposed method offers a better robustness and mesh quality than the state-of-the-art methods. As it combines both modeling and meshing techniques, it can be applied to edit the geometry and topology of vascular models effortlessly to study the impact on hemodynamics. We demonstrate the efficiency of our method by entirely meshing a dataset of 60 cerebral vascular networks. 92% of the vessels and 83% of the bifurcations where mesh without defects needing manual intervention, despite the challenging aspect of the input data. The source code will be released publicly.
34.Real-time Rendering for Integral Imaging Light Field Displays Based on a Voxel-Pixel Lookup Table ⬇️
A real-time elemental image array (EIA) generation method which does not sacrifice accuracy nor rely on high-performance hardware is developed, through raytracing and pre-stored voxel-pixel lookup table (LUT). Benefiting from both offline and online working flow, experiments verified the effectiveness.
35.Physically Embodied Deep Image Optimisation ⬇️
Physical sketches are created by learning programs to control a drawing robot. A differentiable rasteriser is used to optimise sets of drawing strokes to match an input image, using deep networks to provide an encoding for which we can compute a loss. The optimised drawing primitives can then be translated into G-code commands which command a robot to draw the image using drawing instruments such as pens and pencils on a physical support medium.
36.Towards deep observation: A systematic survey on artificial intelligence techniques to monitor fetus via Ultrasound Images ⬇️
Developing innovative informatics approaches aimed to enhance fetal monitoring is a burgeoning field of study in reproductive medicine. Several reviews have been conducted regarding Artificial intelligence (AI) techniques to improve pregnancy outcomes. They are limited by focusing on specific data such as mother's care during pregnancy. This systematic survey aims to explore how artificial intelligence (AI) can assist with fetal growth monitoring via Ultrasound (US) image. We used eight medical and computer science bibliographic databases, including PubMed, Embase, PsycINFO, ScienceDirect, IEEE explore, ACM Library, Google Scholar, and the Web of Science. We retrieved studies published between 2010 to 2021. Data extracted from studies were synthesized using a narrative approach. Out of 1269 retrieved studies, we included 107 distinct studies from queries that were relevant to the topic in the survey. We found that 2D ultrasound images were more popular (n=88) than 3D and 4D ultrasound images (n=19). Classification is the most used method (n=42), followed by segmentation (n=31), classification integrated with segmentation (n=16) and other miscellaneous such as object-detection, regression and reinforcement learning (n=18). The most common areas within the pregnancy domain were the fetus head (n=43), then fetus body (n=31), fetus heart (n=13), fetus abdomen (n=10), and lastly the fetus face (n=10). In the most recent studies, deep learning techniques were primarily used (n=81), followed by machine learning (n=16), artificial neural network (n=7), and reinforcement learning (n=2). AI techniques played a crucial role in predicting fetal diseases and identifying fetus anatomy structures during pregnancy. More research is required to validate this technology from a physician's perspective, such as pilot studies and randomized controlled trials on AI and its applications in a hospital setting.
37.ASL Video Corpora & Sign Bank: Resources Available through the American Sign Language Linguistic Research Project (ASLLRP) ⬇️
The American Sign Language Linguistic Research Project (ASLLRP) provides Internet access to high-quality ASL video data, generally including front and side views and a close-up of the face. The manual and non-manual components of the signing have been linguistically annotated using SignStream(R). The recently expanded video corpora can be browsed and searched through the Data Access Interface (DAI 2) we have designed; it is possible to carry out complex searches. The data from our corpora can also be downloaded; annotations are available in an XML export format. We have also developed the ASLLRP Sign Bank, which contains almost 6,000 sign entries for lexical signs, with distinct English-based glosses, with a total of 41,830 examples of lexical signs (in addition to about 300 gestures, over 1,000 fingerspelled signs, and 475 classifier examples). The Sign Bank is likewise accessible and searchable on the Internet; it can also be accessed from within SignStream(R) (software to facilitate linguistic annotation and analysis of visual language data) to make annotations more accurate and efficient. Here we describe the available resources. These data have been used for many types of research in linguistics and in computer-based sign language recognition from video; examples of such research are provided in the latter part of this article.
38.Homogenization of Existing Inertial-Based Datasets to Support Human Activity Recognition ⬇️
Several techniques have been proposed to address the problem of recognizing activities of daily living from signals. Deep learning techniques applied to inertial signals have proven to be effective, achieving significant classification accuracy. Recently, research in human activity recognition (HAR) models has been almost totally model-centric. It has been proven that the number of training samples and their quality are critical for obtaining deep learning models that both perform well independently of their architecture, and that are more robust to intraclass variability and interclass similarity. Unfortunately, publicly available datasets do not always contain hight quality data and a sufficiently large and diverse number of samples (e.g., number of subjects, type of activity performed, and duration of trials). Furthermore, datasets are heterogeneous among them and therefore cannot be trivially combined to obtain a larger set. The final aim of our work is the definition and implementation of a platform that integrates datasets of inertial signals in order to make available to the scientific community large datasets of homogeneous signals, enriched, when possible, with context information (e.g., characteristics of the subjects and device position). The main focus of our platform is to emphasise data quality, which is essential for training efficient models.
39.Convolutional Neural Networks for Spherical Signal Processing via Spherical Haar Tight Framelets ⬇️
In this paper, we develop a general theoretical framework for constructing Haar-type tight framelets on any compact set with a hierarchical partition. In particular, we construct a novel area-regular hierarchical partition on the 2-sphere and establish its corresponding spherical Haar tight framelets with directionality. We conclude by evaluating and illustrating the effectiveness of our area-regular spherical Haar tight framelets in several denoising experiments. Furthermore, we propose a convolutional neural network (CNN) model for spherical signal denoising which employs the fast framelet decomposition and reconstruction algorithms. Experiment results show that our proposed CNN model outperforms threshold methods, and processes strong generalization and robustness properties.
40.An Automated Robotic Arm: A Machine Learning Approach ⬇️
The term robot generally refers to a machine that looks and works in a way similar to a human. The modern industry is rapidly shifting from manual control of systems to automation, in order to increase productivity and to deliver quality products. Computer-based systems, though feasible for improving quality and productivity, are inflexible to work with, and the cost of such systems is significantly high. This led to the swift adoption of automated systems to perform industrial tasks. One such task of industrial significance is of picking and placing objects from one place to another. The implementation of automation in pick and place tasks helps to improve efficiency of system and also the performance. In this paper, we propose to demonstrate the designing and working of an automated robotic arm with the Machine Learning approach. The work uses Machine Learning approach for object identification detection and traversal, which is adopted with Tensor flow package for better and accurate results.
41.ROS georegistration: Aerial Multi-spectral Image Simulator for the Robot Operating System ⬇️
This article describes a software package called ROS georegistration intended for use with the Robot Operating System (ROS) and the Gazebo 3D simulation environment. ROSgeoregistration provides tools for the simulation, test and deployment of aerial georegistration algorithms and is made available with a link provided in the paper. A model creation package is provided which downloads multi-spectral images from the Google Earth Engine database and, if necessary, incorporates these images into a single, possibly very large, reference image. Additionally a Gazebo plugin which uses the real-time sensor pose and image formation model to generate simulated imagery using the specified reference image is provided along with related plugins for UAV relevant data. The novelty of this work is threefold: (1) this is the first system to link the massive multi-spectral imaging database of Google's Earth Engine to the Gazebo simulator, (2) this is the first example of a system that can simulate geospatially and radiometrically accurate imagery from multiple sensor views of the same terrain region, and (3) integration with other UAS tools creates a new holistic UAS simulation environment to support UAS system and subsystem development where real-world testing would generally be prohibitive. Sensed imagery and ground truth registration information is published to client applications which can receive imagery synchronously with telemetry from other payload sensors, e.g., IMU, GPS/GNSS, barometer, and windspeed sensor data. To highlight functionality, we demonstrate ROSgeoregistration for simulating Electro-Optical (EO) and Synthetic Aperture Radar (SAR) image sensors and an example use case for developing and evaluating image-based UAS position feedback, i.e., pose for image-based Guidance Navigation and Control (GNC) applications.
42.BLINC: Lightweight Bimodal Learning for Low-Complexity VVC Intra Coding ⬇️
The latest video coding standard, Versatile Video Coding (VVC), achieves almost twice coding efficiency compared to its predecessor, the High Efficiency Video Coding (HEVC). However, achieving this efficiency (for intra coding) requires 31x computational complexity compared to HEVC, making it challenging for low power and real-time applications. This paper, proposes a novel machine learning approach that jointly and separately employs two modalities of features, to simplify the intra coding decision. First a set of features are extracted that use the existing DCT core of VVC, to assess the texture characteristics, and forms the first modality of data. This produces high quality features with almost no overhead. The distribution of intra modes at the neighboring blocks is also used to form the second modality of data, which provides statistical information about the frame. Second, a two-step feature reduction method is designed that reduces the size of feature set, such that a lightweight model with a limited number of parameters can be used to learn the intra mode decision task. Third, three separate training strategies are proposed (1) an offline training strategy using the first (single) modality of data, (2) an online training strategy that uses the second (single) modality, and (3) a mixed online-offline strategy that uses bimodal learning. Finally, a low-complexity encoding algorithms is proposed based on the proposed learning strategies. Extensive experimental results show that the proposed methods can reduce up to 24% of encoding time, with a negligible loss of coding efficiency. Moreover, it is demonstrated how a bimodal learning strategy can boost the performance of learning. Lastly, the proposed method has a very low computational overhead (0.2%), and uses existing components of a VVC encoder, which makes it much more practical compared to competing solutions.