Temporal action detection (TAD) aims to determine the semantic label and the boundaries of every action instance in an untrimmed video. It is a fundamental and challenging task in video understanding and significant progress has been made. Previous methods involve multiple stages or networks and hand-designed rules or operations, which fall short in efficiency and flexibility. In this paper, we propose an end-to-end framework for TAD upon Transformer, termed \textit{TadTR}, which maps a set of learnable embeddings to action instances in parallel. TadTR is able to adaptively extract temporal context information required for making action predictions, by selectively attending to a sparse set of snippets in a video. As a result, it simplifies the pipeline of TAD and requires lower computation cost than previous detectors, while preserving remarkable detection performance. TadTR achieves state-of-the-art performance on HACS Segments (+3.35% average mAP). As a single-network detector, TadTR runs 10$\times$ faster than its comparable competitor. It outperforms existing single-network detectors by a large margin on THUMOS14 (+5.0% average mAP) and ActivityNet (+7.53% average mAP). When combined with other detectors, it reports 54.1% mAP at IoU=0.5 on THUMOS14, and 34.55% average mAP on ActivityNet-1.3. Our code will be released at \url{https://github.com/xlliu7/TadTR}.
We address the challenging problem of Natural Language Comprehension beyond plain-text documents by introducing the TILT neural network architecture which simultaneously learns layout information, visual features, and textual semantics. Contrary to previous approaches, we rely on a decoder capable of solving all problems involving natural language. The layout is represented as an attention bias and complemented with contextualized visual information, while the core of our model is a pretrained encoder-decoder Transformer. We trained our network on real-world documents with different layouts, such as tables, figures, and forms. Our novel approach achieves state-of-the-art in extracting information from documents and answering questions, demanding layout understanding (DocVQA, CORD, WikiOps, SROIE). At the same time, we simplify the process by employing an end-to-end model.
Random exploration is one of the main mechanisms through which reinforcement learning (RL) finds well-performing policies. However, it can lead to undesirable or catastrophic outcomes when learning online in safety-critical environments. In fact, safe learning is one of the major obstacles towards real-world agents that can learn during deployment. One way of ensuring that agents respect hard limitations is to explicitly configure boundaries in which they can operate. While this might work in some cases, we do not always have clear a-priori information which states and actions can lead dangerously close to hazardous states. Here, we present an approach where an additional policy can override the main policy and offer a safer alternative action. In our instinct-regulated RL (IR^2L) approach, an "instinctual" network is trained to recognize undesirable situations, while guarding the learning policy against entering them. The instinct network is pre-trained on a single task where it is safe to make mistakes, and transferred to environments in which learning a new task safely is critical. We demonstrate IR^2L in the OpenAI Safety gym domain, in which it receives a significantly lower number of safety violations during training than a baseline RL approach while reaching similar task performance.
Unsupervised learning methods have recently shown their competitiveness against supervised training. Typically, these methods use a single objective to train the entire network. But one distinct advantage of unsupervised over supervised learning is that the former possesses more variety and freedom in designing the objective. In this work, we explore new dimensions of unsupervised learning by proposing the Progressive Stage-wise Learning (PSL) framework. For a given unsupervised task, we design multilevel tasks and define different learning stages for the deep network. Early learning stages are forced to focus on lowlevel tasks while late stages are guided to extract deeper information through harder tasks. We discover that by progressive stage-wise learning, unsupervised feature representation can be effectively enhanced. Our extensive experiments show that PSL consistently improves results for the leading unsupervised learning methods.
Detecting pedestrians is a crucial task in autonomous driving systems to ensure the safety of drivers and pedestrians. The technologies involved in these algorithms must be precise and reliable, regardless of environment conditions. Relying solely on RGB cameras may not be enough to recognize road environments in situations where cameras cannot capture scenes properly. Some approaches aim to compensate for these limitations by combining RGB cameras with TOF sensors, such as LIDARs. However, there are few works that address this problem using exclusively the 3D geometric information provided by LIDARs. In this paper, we propose a PointNet++ based architecture to detect pedestrians in dense 3D point clouds. The aim is to explore the potential contribution of geometric information alone in pedestrian detection systems. We also present a semi-automatic labeling system that transfers pedestrian and non-pedestrian labels from RGB images onto the 3D domain. The fact that our datasets have RGB registered with point clouds enables label transferring by back projection from 2D bounding boxes to point clouds, with only a light manual supervision to validate results. We train PointNet++ with the geometry of the resulting 3D labelled clusters. The evaluation confirms the effectiveness of the proposed method, yielding precision and recall values around 98%.
The paper presents an analysis of the latest developments in the field of stereo vision in the low-cost segment, both for prototypes and for industrial designs. We described the theory of stereo vision and presented information about cameras and data transfer protocols and their compatibility with various devices. The theory in the field of image processing for stereo vision processes is considered and the calibration process is described in detail. Ultimately, we presented the developed stereo vision system and provided the main points that need to be considered when developing such systems. The final, we presented software for adjusting stereo vision parameters in real-time in the python language in the Windows operating system.
Injection molding is one of the most popular manufacturing methods for the modeling of complex plastic objects. Faster numerical simulation of the technological process would allow for faster and cheaper design cycles of new products. In this work, we propose a baseline for a data processing pipeline that includes the extraction of data from Moldflow simulation projects and the prediction of the fill time and deflection distributions over 3-dimensional surfaces using machine learning models. We propose algorithms for engineering of features, including information of injector gates parameters that will mostly affect the time for plastic to reach the particular point of the form for fill time prediction, and geometrical features for deflection prediction. We propose and evaluate baseline machine learning models for fill time and deflection distribution prediction and provide baseline values of MSE and RMSE metrics. Finally, we measure the execution time of our solution and show that it significantly exceeds the time of simulation with Moldflow software: approximately 17 times and 14 times faster for mean and median total times respectively, comparing the times of all analysis stages for deflection prediction. Our solution has been implemented in a prototype web application that was approved by the management board of Fiat Chrysler Automobiles and Illogic SRL. As one of the promising applications of this surrogate modelling approach, we envision the use of trained models as a fast objective function in the task of optimization of technological parameters of the injection molding process (meaning optimal placement of gates), which could significantly aid engineers in this task, or even automate it.
Most classification models treat different object classes in parallel and the misclassifications between any two classes are treated equally. In contrast, human beings can exploit high-level information in making a prediction of an unknown object. Inspired by this observation, the paper proposes a super-class guided network (SGNet) to integrate the high-level semantic information into the network so as to increase its performance in inference. SGNet takes two-level class annotations that contain both super-class and finer class labels. The super-classes are higher-level semantic categories that consist of a certain amount of finer classes. A super-class branch (SCB), trained on super-class labels, is introduced to guide finer class prediction. At the inference time, we adopt two different strategies: Two-step inference (TSI) and direct inference (DI). TSI first predicts the super-class and then makes predictions of the corresponding finer class. On the other hand, DI directly generates predictions from the finer class branch (FCB). Extensive experiments have been performed on CIFAR-100 and MS COCO datasets. The experimental results validate the proposed approach and demonstrate its superior performance on image classification and object detection.
Existing approaches for unsupervised domain adaptive object detection perform feature alignment via adversarial training. While these methods achieve reasonable improvements in performance, they typically perform category-agnostic domain alignment, thereby resulting in negative transfer of features. To overcome this issue, in this work, we attempt to incorporate category information into the domain adaptation process by proposing Memory Guided Attention for Category-Aware Domain Adaptation (MeGA-CDA). The proposed method consists of employing category-wise discriminators to ensure category-aware feature alignment for learning domain-invariant discriminative features. However, since the category information is not available for the target samples, we propose to generate memory-guided category-specific attention maps which are then used to route the features appropriately to the corresponding category discriminator. The proposed method is evaluated on several benchmark datasets and is shown to outperform existing approaches.
The research of visual signal compression has a long history. Fueled by deep learning, exciting progress has been made recently. Despite achieving better compression performance, existing end-to-end compression algorithms are still designed towards better signal quality in terms of rate-distortion optimization. In this paper, we show that the design and optimization of network architecture could be further improved for compression towards machine vision. We propose an inverted bottleneck structure for end-to-end compression towards machine vision, which specifically accounts for efficient representation of the semantic information. Moreover, we quest the capability of optimization by incorporating the analytics accuracy into the optimization process, and the optimality is further explored with generalized rate-accuracy optimization in an iterative manner. We use object detection as a showcase for end-to-end compression towards machine vision, and extensive experiments show that the proposed scheme achieves significant BD-rate savings in terms of analysis performance. Moreover, the promise of the scheme is also demonstrated with strong generalization capability towards other machine vision tasks, due to the enabling of signal-level reconstruction.