Following the tracking-by-detection paradigm, multiple object tracking deals with challenging scenarios, occlusions or even missing detections; the priority is often given to quality measures instead of speed, and a good trade-off between the two is hard to achieve. Based on recent work, we propose a fast, light-weight tracker able to predict targets position and reidentify them at once, when it is usually done with two sequential steps. To do so, we combine a bounding box regressor with a target-oriented appearance learner in a newly designed and unified architecture. This way, our tracker can infer the targets' image pose but also provide us with a confidence level about target identity. Most of the time, it is also common to filter out the detector outputs with a preprocessing step, throwing away precious information about what has been seen in the image. We propose a tracks management strategy able to balance efficiently between detection and tracking outputs and their associated likelihoods. Simply put, we spotlight a full siamese based single object tracker able to predict both position and appearance features at once with a light-weight and all-in-one architecture, within a balanced overall multi-target management strategy. We demonstrate the efficiency and speed of our system w.r.t the literature on the well-known MOT17 challenge benchmark, and bring to the fore qualitative evaluations as well as state-of-the-art quantitative results.

DeepFashion2 dataset raises a new challenge for a keypoint detection task. It contains 13 categories with a different number of keypoints, 294 in total. Direct prediction of all keypoints leads to huge memory consumption, slow training, and inference speed. This paper presents a study of keypoint grouping approach and how it affects performance on the example of CenterNet architecture. We propose a simple and efficient automatic grouping technique and apply it to DeepFashion2 fashion landmark task and MS COCO Human Pose task. It allows reducing memory consumption up to 30%, decreasing inference time up to 30%, and training time up to 26% without compromising accuracy.

We propose a deep neural network-based method to recover dynamic online trajectories from offline handwritten Japanese kanji character images. It is a challenging task since Japanese kanji characters consist of multiple strokes. Our proposed model has three main components: Convolutional Neural Network-based encoder, Long Short-Term Memory Network-based decoder with an attention layer, and Gaussian Mixture Model (GMM). The encoder focuses on feature extraction while the decoder refers to the extracted features and generates time-sequences of GMM parameters. The attention layer is the key component for trajectory recovery. The GMM provides robustness to style variations so that the proposed model does not overfit to training samples. In the experiments, the proposed method is evaluated by both visual verification and handwritten character recognition. This is the first attempt to use online recovered trajectories to help improve the performance of offline handwriting recognition. Although the visual verification reveals some problems, the recognition experiments demonstrate the effect of trajectory recovery in improving the accuracy of offline handwritten character recognition when online recognition of the recovered trajectories are combined.

Human-object interaction recognition aims for identifying the relationship between a human subject and an object. Researchers incorporate global scene context into the early layers of deep Convolutional Neural Networks as a solution. They report a significant increase in the performance since generally interactions are correlated with the scene (i.e. riding bicycle on the city street). However, this approach leads to the following problems. It increases the network size in the early layers, therefore not efficient. It leads to noisy filter responses when the scene is irrelevant, therefore not accurate. It only leverages scene context whereas human-object interactions offer a multitude of contexts, therefore incomplete. To circumvent these issues, in this work, we propose Self-Selective Context (SSC). SSC operates on the joint appearance of human-objects and context to bring the most discriminative context(s) into play for recognition. We devise novel contextual features that model the locality of human-object interactions and show that SSC can seamlessly integrate with the State-of-the-art interaction recognition models. Our experiments show that SSC leads to an important increase in interaction recognition performance, while using much fewer parameters.

Substantial efforts have been made on improving the generalization abilities of deep neural networks (DNNs) in order to obtain better performances without introducing more parameters. On the other hand, meta-learning approaches exhibit powerful generalization on new tasks in few-shot learning. Intuitively, few-shot learning is more challenging than the standard supervised learning as each target class only has a very few or no training samples. The natural question that arises is whether the meta-learning idea can be used for improving the generalization of DNNs on the standard supervised learning. In this paper, we propose a novel meta-learning based training procedure (MLTP) for DNNs and demonstrate that the meta-learning idea can indeed improve the generalization abilities of DNNs. MLTP simulates the meta-training process by considering a batch of training samples as a task. The key idea is that the gradient descent step for improving the current task performance should also improve a new task performance, which is ignored by the current standard procedure for training neural networks. MLTP also benefits from all the existing training techniques such as dropout, weight decay, and batch normalization. We evaluate MLTP by training a variety of small and large neural networks on three benchmark datasets, i.e., CIFAR-10, CIFAR-100, and Tiny ImageNet. The experimental results show a consistently improved generalization performance on all the DNNs with different sizes, which verifies the promise of MLTP and demonstrates that the meta-learning idea is indeed able to improve the generalization of DNNs on the standard supervised learning.

As part of the development of an educational tool that can help students achieve fluency in American Sign Language (ASL) through independent and interactive practice with immediate feedback, this paper introduces a near real-time system to recognize grammatical errors in continuous signing videos without necessarily identifying the entire sequence of signs. Our system automatically recognizes if a performance of ASL sentences contains grammatical errors made by ASL students. We first recognize the ASL grammatical elements including both manual gestures and nonmanual signals independently from multiple modalities (i.e. hand gestures, facial expressions, and head movements) by 3D-ResNet networks. Then the temporal boundaries of grammatical elements from different modalities are examined to detect ASL grammatical mistakes by using a sliding window-based approach. We have collected a dataset of continuous sign language, ASL-HW-RGBD, covering different aspects of ASL grammars for training and testing. Our system is able to recognize grammatical elements on ASL-HW-RGBD from manual gestures, facial expressions, and head movements and successfully detect 8 ASL grammatical mistakes.

Deep Learning systems have proven to be extremely successful for image recognition tasks for which significant amounts of training data is available, e.g., on the famous ImageNet dataset. We demonstrate that for robotics applications with cheap camera equipment, the low image quality, however, influences the classification accuracy, and freely available data bases cannot be exploited in a straight forward way to train classifiers to be used on a robot. As a solution we propose to train a network on degrading the quality images in order to mimic specific low quality imaging systems. Numerical experiments demonstrate that classification networks trained by using images produced by our quality degrading network along with the high quality images outperform classification networks trained only on high quality data when used on a real robot system, while being significantly easier to use than competing zero-shot domain adaptation techniques.

Since we were babies, we intuitively develop the ability to correlate the input from different cognitive sensors such as vision, audio, and text. However, in machine learning, this cross-modal learning is a nontrivial task because different modalities have no homogeneous properties. Previous works discover that there should be bridges among different modalities. From neurology and psychology perspective, humans have the capacity to link one modality with another one, e.g., associating a picture of a bird with the only hearing of its singing and vice versa. Is it possible for machine learning algorithms to recover the scene given the audio signal? In this paper, we propose a novel Cascade Attention-Guided Residue GAN (CAR-GAN), aiming at reconstructing the scenes given the corresponding audio signals. Particularly, we present a residue module to mitigate the gap between different modalities progressively. Moreover, a cascade attention guided network with a novel classification loss function is designed to tackle the cross-modal learning task. Our model keeps consistency in the high-level semantic label domain and is able to balance two different modalities. The experimental results demonstrate that our model achieves the state-of-the-art cross-modal audio-visual generation on the challenging Sub-URMP dataset.

Automated driving is regarded as the most promising technology for improving road safety in the future. In this context, connected vehicles have an important role regarding their ability to perform cooperative maneuvers for challenging traffic situations. We propose a benchmark for automated cooperative maneuvers. The targeted cooperative maneuver is the vehicle lane merge where a vehicle on the acceleration lane merges into the traffic of a motorway. The benchmark enables the evaluation of vehicle localization approaches as well as the study of cooperative maneuvers. It consists of temporally synchronized multi-view video streams, highly accurate camera calibration, and ground truth vehicle descriptions, including position, heading, speed, and shape. For benchmark generation, the lane merge maneuver is performed by human drivers on a test track, resulting in 120 lane merge data sets with various traffic situations and video recording conditions.

Generative adversarial networks (GANs) are deep neural networks that allow us to sample from an arbitrary probability distribution without explicitly estimating the distri- bution. There is a generator that takes a latent vector as input and transforms it into a valid sample from the distribution. There is also a discriminator that is trained to discriminate such fake samples from true samples of the distribution; at the same time, the generator is trained to generate fakes that the discriminator cannot tell apart from the true samples. Instead of learning a global generator, a recent approach involves training multiple generators each responsible from one part of the distribution. In this work, we review such approaches and propose the hierarchical mixture of generators, inspired from the hierarchical mixture of experts model, that learns a tree structure implementing a hierarchical clustering with soft splits in the decision nodes and local generators in the leaves. Since the generators are combined softly, the whole model is continuous and can be trained using gradient-based optimization, just like the original GAN model. Our experiments on five image data sets, namely, MNIST, FashionMNIST, UTZap50K, Oxford Flowers, and CelebA, show that our proposed model generates samples of high quality and diversity in terms of popular GAN evaluation metrics. The learned hierarchical structure also leads to knowledge extraction.

Recent research has revealed that deep neural networks are highly vulnerable to adversarial examples. In this paper, different from most adversarial attacks which directly modify pixels in spatial domain, we propose a novel black-box attack in frequency domain, named as f-mixup, based on the property of natural images and perception disparity between human-visual system (HVS) and convolutional neural networks (CNNs): First, natural images tend to have the bulk of their Fourier spectrums concentrated on the low frequency domain; Second, HVS is much less sensitive to high frequencies while CNNs can utilize both low and high frequency information to make predictions. Extensive experiments are conducted and show that deeper CNNs tend to concentrate more on the high frequency domain, which may explain the contradiction between robustness and accuracy. In addition, we compared f-mixup with existing attack methods and observed that our approach possesses great advantages. Finally, we show that f-mixup can be also incorporated in training to make deep CNNs defensible against a kind of perturbations effectively.

Image stitching describes the process of reconstruction of a high resolution image from combining multiple images. Using a scanning electron microscope as the image source, individual images will show patterns in a nm dimension whereas the combined image may cover an area of several mm2. The recovery of the physical layout of modern semiconductor products manufactured in advanced technologies nodes down to 22 nm requires a perfect stitching process with no deviation with respect to the original design data, as any stitching error will result in failures during the reconstruction of the electrical design. In addition, the recovery of the complete design requires the acquisition of all individual layers of a semiconductor device which represent a 3D structure with interconnections defining error limits on the stitching error for each individual scanned image mosaic. An advanced stitching and alignment process is presented enabling a true geometrical layout recovery in nanoscale dimensions which is also applied and evaluated on other use cases from biological applications.

Super-resolution aims at increasing the resolution and level of detail within an image. The current state of the art in general single-image super-resolution is held by nESRGAN+,which injects a Gaussian noise after each residual layer at training time. In this paper, we harness evolutionary methods to improve nESRGAN+ by optimizing the noise injection at inference time. More precisely, we use Diagonal CMA to optimize the injected noise according to a novel criterion combining quality assessment and realism. Our results are validated by the PIRM perceptual score and a human study. Our method outperforms nESRGAN+ on several standard super-resolution datasets. More generally, our approach can be used to optimize any method based on noise injection.

In this paper, we propose a method to detect various types of lane markings and road boundaries simultaneously from a monocular camera image. This method detects lane markings and road boundaries in IPM images obtained by the Inverse Perspective Mapping of input images. First, bounding boxes surrounding a lane marking or the road boundary are extracted by the object detection network. At the same time, these areas are labelled with a solid line, a dashed line, a zebra line, a curb, a grass, a sidewall and so on. Next, in each bounding box, lane marking boundaries or road boundaries are drawn by the image-to-image translation network. We use YOLOv3 for the object detection and pix2pix for the image translation. We create our own datasets including various types of lane markings and road boundaries and evaluate our approach using these datasets qualitatively and quantitatively.

Recently, Facial expression synthesis has shown remarkable advances with the advent of Generative Adversarial Networks (GANs). However, these GAN-based approaches mostly generate photo-realistic results as long as the target data distribution is close to the training data distribution. The quality of GANs results significantly degrades when testing images are from a slightly different distribution. In this work, we propose a pixel-based facial expression synthesis method. Recent work has shown that facial expression synthesis changes only local regions of faces. In the proposed method, each output pixel observes only one input pixel. The proposed method achieves generalization capability by leveraging only few hundred images. Experimental results demonstrate that the proposed method performs comparably with the recent GANs on in-dataset images and significantly outperforms on in the wild images. In addition, the proposed method is faster and it also achieves significantly better performance with two orders of magnitudes lesser computational and storage cost as compared to state-of-the-art GAN-based methods.

In this paper, a multi-task learning framework has been proposed for solving and improving traditional Mongolian words recognition. To be specific, a sequence-to-sequence model with attention mechanism was utilized to accomplish the task of recognition. Therein, the attention mechanism is designed to fulfill the task of glyph segmentation during the process of recognition. Although the glyph segmentation is an implicit operation, the information of glyph segmentation can be integrated into the process of recognition. After that, the two tasks can be accomplished simultaneously under the framework of multi-task learning. By this way, adjacent image frames can be decoded into a glyph more precisely, which results in improving not only the performance of words recognition but also the accuracy of character segmentation. Experimental results demonstrate that the proposed multi-task learning based scheme outperforms the conventional glyph segmentation-based method and various segmentation-free (i.e. holistic recognition) methods.

Graphical models allow to describe the interplay among variables of a system through a compact representation, suitable when relations evolve over time. For example, in a biological setting, genes interact differently depending on external environmental or metabolic factors. To incorporate this dynamics a viable strategy is to estimate a sequence of temporally related graphs assuming similarity among samples in different time points. While adjacent time points may direct the analysis towards a robust estimate of the underlying graph, the resulting model will not incorporate long-term or recurrent temporal relationships. In this work we propose a dynamical network inference model that leverages on kernels to consider general temporal patterns (such as circadian rhythms or seasonality). We show how our approach may also be exploited when the recurrent patterns are unknown, by coupling the network inference with a clustering procedure that detects possibly non-consecutive similar networks. Such clusters are then used to build similarity kernels. The convexity of the functional is determined by whether we impose or infer the kernel. In the first case, the optimisation algorithm exploits efficiently proximity operators with closed-form solutions. In the other case, we resort to an alternating minimisation procedure which jointly learns the temporal kernel and the underlying network. Extensive analysis on synthetic data shows the efficacy of our models compared to state-of-the-art methods. Finally, we applied our approach on two real-world applications to show how considering long-term patterns is fundamental to have insights on the behaviour of a complex system.

In this paper, we propose a novel network training mechanism called "dynamic channel propagation" to prune the model during the training period. In particular, we pick up a specific group of channels in each convolutional layer to participate in the forward propagation in training time according to the significance level of channel, which is defined as channel utility. The utility values with respect to all selected channels are updated simultaneously with the error back-propagation process and will constantly change. Furthermore, when the training ends, channels with high utility values are retained whereas those with low utility values are discarded. Hence, our proposed method trains and prunes neural networks simultaneously. We empirically evaluate our novel training method on various representative benchmark datasets and advanced convolutional neural network (CNN) architectures, including VGGNet and ResNet. The experiment results verify superior performance and robust effectiveness of our approach.

Recent researches demonstrate that correlation modeling plays a key role in high-performance multi-label classification methods. However, existing methods do not take full advantage of correlation information, especially correlations in feature and label spaces of each image, which limits the performance of correlation-based multi-label classification methods. With more correlations considered, in this study, a Fully Associative Network (FAN) is proposed for fully exploiting correlation information, which involves both visual feature and label correlations. Specifically, FAN introduces a robust covariance pooling to summarize convolution features as global image representation for capturing feature correlation in the multi-label task. Moreover, it constructs an effective label correlation matrix based on a re-weighted scheme, which is fed into a graph convolution network for capturing label correlation. Then, correlation between covariance representations (i.e., feature correlation ) and the outputs of GCN (i.e., label correlation) are modeled for final prediction. Experimental results on two datasets illustrate the effectiveness and efficiency of our proposed FAN compared with state-of-the-art methods.

Decision trees excel at interpretability of their prediction results. To achieve required prediction accuracies, however, often large ensembles of decision trees -- random forests -- are considered, reducing interpretability due to large size. Additionally, their size slows down inference on modern hardware and restricts their applicability in low-memory embedded devices. We introduce \emph{Decision Snippet Features}, which are obtained from small subtrees that appear frequently in trained random forests. We subsequently show that linear models on top of these features achieve comparable and sometimes even better predictive performance than the original random forest, while reducing the model size by up to two orders of magnitude.

In recent years, surveillance cameras are widely deployed in public places, and the general crime rate has been reduced significantly due to these ubiquitous devices. Usually, these cameras provide cues and evidence after crimes were conducted, while they are rarely used to prevent or stop criminal activities in time. It is both time and labor consuming to manually monitor a large amount of video data from surveillance cameras. Therefore, automatically recognizing violent behaviors from video signals becomes essential. In this paper, we summarize several existing video datasets for violence detection and propose a new video dataset with 2,000 videos all captured by surveillance cameras in real-world scenes. Also, we present a new method that utilizes both the merits of 3D-CNNs and optical flow, namely Flow Gated Network. The proposed approach obtains an accuracy of 87.25% on the test set of our proposed RWF-2000 database. The proposed database and source codes of this paper are currently open to access.