This study follows many classical approaches to multi-object tracking (MOT) that model the problem using dynamic graphical data structures, and adapts this formulation to make it amenable to modern neural networks. Our main contributions in this work are the creation of a framework based on dynamic undirected graphs that represent the data association problem over multiple timesteps, and a message passing graph neural network (MPNN) that operates on these graphs to produce the desired likelihood for every association therein. We also provide solutions and propositions for the computational problems that need to be addressed to create a memory-efficient, real-time, online algorithm that can reason over multiple timesteps, correct previous mistakes, update beliefs, and handle missed/false detections. To demonstrate the efficacy of our approach, we only use the 2D box location and object category ID to construct the descriptor for each object instance. Despite this, our model performs on par with state-of-the-art approaches that make use of additional sensors, as well as multiple hand-crafted and/or learned features. This illustrates that given the right problem formulation and model design, raw bounding boxes (and their kinematics) from any off-the-shelf detector are sufficient to achieve competitive tracking results on challenging MOT benchmarks.
Precise segmentation of objects is an important problem in tasks like class-agnostic object proposal generation or instance segmentation. Deep learning-based systems usually generate segmentations of objects based on coarse feature maps, due to the inherent downsampling in CNNs. This leads to segmentation boundaries not adhering well to the object boundaries in the image. To tackle this problem, we introduce a new superpixel-based refinement approach on top of the state-of-the-art object proposal system AttentionMask. The refinement utilizes superpixel pooling for feature extraction and a novel superpixel classifier to determine if a high precision superpixel belongs to an object or not. Our experiments show an improvement of up to 26.0% in terms of average recall compared to original AttentionMask. Furthermore, qualitative and quantitative analyses of the segmentations reveal significant improvements in terms of boundary adherence for the proposed refinement compared to various deep learning-based state-of-the-art object proposal generation systems.
Predicting the future interaction of objects when they come into contact with their environment is key for autonomous agents to take intelligent and anticipatory actions. This paper presents a perception framework that fuses visual and tactile feedback to make predictions about the expected motion of objects in dynamic scenes. Visual information captures object properties such as 3D shape and location, while tactile information provides critical cues about interaction forces and resulting object motion when it makes contact with the environment. Utilizing a novel See-Through-your-Skin (STS) sensor that provides high resolution multimodal sensing of contact surfaces, our system captures both the visual appearance and the tactile properties of objects. We interpret the dual stream signals from the sensor using a Multimodal Variational Autoencoder (MVAE), allowing us to capture both modalities of contacting objects and to develop a mapping from visual to tactile interaction and vice-versa. Additionally, the perceptual system can be used to infer the outcome of future physical interactions, which we validate through simulated and real-world experiments in which the resting state of an object is predicted from given initial conditions.
Autonomous vehicles are conceived to provide safe and secure services by validating the safety standards as indicated by SOTIF-ISO/PAS-21448 (Safety of the intended functionality). Keeping in this context, the perception of the environment plays an instrumental role in conjunction with localization, planning and control modules. As a pivotal algorithm in the perception stack, object detection provides extensive insights into the autonomous vehicle's surroundings. Camera and Lidar are extensively utilized for object detection among different sensor modalities, but these exteroceptive sensors have limitations in resolution and adverse weather conditions. In this work, radar-based object detection is explored provides a counterpart sensor modality to be deployed and used in adverse weather conditions. The radar gives complex data; for this purpose, a channel boosting feature ensemble method with transformer encoder-decoder network is proposed. The object detection task using radar is formulated as a set prediction problem and evaluated on the publicly available dataset in both good and good-bad weather conditions. The proposed method's efficacy is extensively evaluated using the COCO evaluation metric, and the best-proposed model surpasses its state-of-the-art counterpart method by $12.55\%$ and $12.48\%$ in both good and good-bad weather conditions.
We introduce the first metric for evaluating disentanglement at individual hierarchy levels of a structured latent representation. Applied to object-centric generative models, this offers a systematic, unified approach to evaluating (i) object separation between latent slots (ii) disentanglement of object properties inside individual slots (iii) disentanglement of intrinsic and extrinsic object properties. We theoretically show that for structured representations, our framework gives stronger guarantees of selecting a good model than previous disentanglement metrics. Experimentally, we demonstrate that viewing object compositionality as a disentanglement problem addresses several issues with prior visual metrics of object separation. As a core technical component, we present the first representation probing algorithm handling slot permutation invariance.
Existing spatial object recommendation algorithms generally treat objects identically when ranking them. However, spatial objects often cover different levels of spatial granularity and thereby are heterogeneous. For example, one user may prefer to be recommended a region (say Manhattan), while another user might prefer a venue (say a restaurant). Even for the same user, preferences can change at different stages of data exploration. In this paper, we study how to support top-k spatial object recommendations at varying levels of spatial granularity, enabling spatial objects at varying granularity, such as a city, suburb, or building, as a Point of Interest (POI). To solve this problem, we propose the use of a POI tree, which captures spatial containment relationships between POIs. We design a novel multi-task learning model called MPR (short for Multi-level POI Recommendation), where each task aims to return the top-k POIs at a certain spatial granularity level. Each task consists of two subtasks: (i) attribute-based representation learning; (ii) interaction-based representation learning. The first subtask learns the feature representations for both users and POIs, capturing attributes directly from their profiles. The second subtask incorporates user-POI interactions into the model. Additionally, MPR can provide insights into why certain recommendations are being made to a user based on three types of hints: user-aspect, POI-aspect, and interaction-aspect. We empirically validate our approach using two real-life datasets, and show promising performance improvements over several state-of-the-art methods.
Tiny object classification problem exists in many machine learning applications like medical imaging or remote sensing, where the object of interest usually occupies a small region of the whole image. It is challenging to design an efficient machine learning model with respect to tiny object of interest. Current neural network structures are unable to deal with tiny object efficiently because they are mainly developed for images featured by large scale objects. However, in quantum physics, there is a great theoretical foundation guiding us to analyze the target function for image classification regarding to specific objects size ratio. In our work, we apply Tensor Networks to solve this arising tough machine learning problem. First, we summarize the previous work that connects quantum spin model to image classification and bring the theory into the scenario of tiny object classification. Second, we propose using 2D multi-scale entanglement renormalization ansatz (MERA) to classify tiny objects in image. In the end, our experimental results indicate that tensor network models are effective for tiny object classification problem and potentially will beat state-of-the-art. Our codes will be available online https://github.com/timqqt/MERA_Image_Classification.
Confidence calibration is a major concern when applying artificial neural networks in safety-critical applications. Since most research in this area has focused on classification in the past, confidence calibration in the scope of object detection has gained more attention only recently. Based on previous work, we study the miscalibration of object detection models with respect to image location and box scale. Our main contribution is to additionally consider the impact of box selection methods like non-maximum suppression to calibration. We investigate the default intrinsic calibration of object detection models and how it is affected by these post-processing techniques. For this purpose, we distinguish between black-box calibration with non-maximum suppression and white-box calibration with raw network outputs. Our experiments reveal that post-processing highly affects confidence calibration. We show that non-maximum suppression has the potential to degrade initially well-calibrated predictions, leading to overconfident and thus miscalibrated models.
The challenging task of multi-object tracking (MOT) requires simultaneous reasoning about track initialization, identity, and spatio-temporal trajectories. We formulate this task as a frame-to-frame set prediction problem and introduce TrackFormer, an end-to-end trainable MOT approach based on an encoder-decoder Transformer architecture. Our model achieves data association between frames via attention by evolving a set of track predictions through a video sequence. The Transformer decoder initializes new tracks from static object queries and autoregressively follows existing tracks in space and time with the conceptually new and identity preserving track queries. Both query types benefit from self- and encoder-decoder attention on global frame-level features, thereby omitting any additional graph optimization or modeling of motion and/or appearance. TrackFormer introduces a new tracking-by-attention paradigm and while simple in its design is able to achieve state-of-the-art performance on the task of multi-object tracking (MOT17 and MOT20) and segmentation (MOTS20). The code is available at https://github.com/timmeinhardt/trackformer .
Pretraining on large labeled datasets is a prerequisite to achieve good performance in many computer vision tasks like 2D object recognition, video classification etc. However, pretraining is not widely used for 3D recognition tasks where state-of-the-art methods train models from scratch. A primary reason is the lack of large annotated datasets because 3D data is both difficult to acquire and time consuming to label. We present a simple self-supervised pertaining method that can work with any 3D data - single or multiview, indoor or outdoor, acquired by varied sensors, without 3D registration. We pretrain standard point cloud and voxel based model architectures, and show that joint pretraining further improves performance. We evaluate our models on 9 benchmarks for object detection, semantic segmentation, and object classification, where they achieve state-of-the-art results and can outperform supervised pretraining. We set a new state-of-the-art for object detection on ScanNet (69.0% mAP) and SUNRGBD (63.5% mAP). Our pretrained models are label efficient and improve performance for classes with few examples.
Existing point-cloud based 3D object detectors use convolution-like operators to process information in a local neighbourhood with fixed-weight kernels and aggregate global context hierarchically. However, non-local neural networks and self-attention for 2D vision have shown that explicitly modeling long-range interactions can lead to more robust and competitive models. In this paper, we propose two variants of self-attention for contextual modeling in 3D object detection by augmenting convolutional features with self-attention features. We first incorporate the pairwise self-attention mechanism into the current state-of-the-art BEV, voxel and point-based detectors and show consistent improvement over strong baseline models of up to 1.5 3D AP while simultaneously reducing their parameter footprint and computational cost by 15-80% and 30-50%, respectively, on the KITTI validation set. We next propose a self-attention variant that samples a subset of the most representative features by learning deformations over randomly sampled locations. This not only allows us to scale explicit global contextual modeling to larger point-clouds, but also leads to more discriminative and informative feature descriptors. Our method can be flexibly applied to most state-of-the-art detectors with increased accuracy and parameter and compute efficiency. We show our proposed method improves 3D object detection performance on KITTI, nuScenes and Waymo Open datasets. Code is available at https://github.com/AutoVision-cloud/SA-Det3D.
We present a simple yet effective progressive self-guided loss function to facilitate deep learning-based salient object detection (SOD) in images. The saliency maps produced by the most relevant works still suffer from incomplete predictions due to the internal complexity of salient objects. Our proposed progressive self-guided loss simulates a morphological closing operation on the model predictions for progressively creating auxiliary training supervisions to step-wisely guide the training process. We demonstrate that this new loss function can guide the SOD model to highlight more complete salient objects step-by-step and meanwhile help to uncover the spatial dependencies of the salient object pixels in a region growing manner. Moreover, a new feature aggregation module is proposed to capture multi-scale features and aggregate them adaptively by a branch-wise attention mechanism. Benefiting from this module, our SOD framework takes advantage of adaptively aggregated multi-scale features to locate and detect salient objects effectively. Experimental results on several benchmark datasets show that our loss function not only advances the performance of existing SOD models without architecture modification but also helps our proposed framework to achieve state-of-the-art performance.
Segmenting objects in videos is a fundamental computer vision task. The current deep learning based paradigm offers a powerful, but data-hungry solution. However, current datasets are limited by the cost and human effort of annotating object masks in videos. This effectively limits the performance and generalization capabilities of existing video segmentation methods. To address this issue, we explore weaker form of bounding box annotations. We introduce a method for generating segmentation masks from per-frame bounding box annotations in videos. To this end, we propose a spatio-temporal aggregation module that effectively mines consistencies in the object and background appearance across multiple frames. We use our resulting accurate masks for weakly supervised training of video object segmentation (VOS) networks. We generate segmentation masks for large scale tracking datasets, using only their bounding box annotations. The additional data provides substantially better generalization performance leading to state-of-the-art results in both the VOS and more challenging tracking domain.
Object detection and semantic segmentation are two of the most widely adopted deep learning algorithms in agricultural applications. One of the major sources of variability in image quality acquired in the outdoors for such tasks is changing lighting condition that can alter the appearance of the objects or the contents of the entire image. While transfer learning and data augmentation to some extent reduce the need for large amount of data to train deep neural networks, the large variety of cultivars and the lack of shared datasets in agriculture makes wide-scale field deployments difficult. In this paper, we present a high throughput robust active lighting-based camera system that generates consistent images in all lighting conditions. We detail experiments that show the consistency in images quality leading to relatively fewer images to train deep neural networks for the task of object detection. We further present results from field experiment under extreme lighting conditions where images without active lighting significantly lack to provide consistent results. The experimental results show that on average, deep nets for object detection trained on consistent data required nearly four times less data to achieve similar level of accuracy. This proposed work could potentially provide pragmatic solutions to computer vision needs in agriculture.
Semantic segmentation is a hot topic in computer vision where the most challenging tasks of object detection and recognition have been handling by the success of semantic segmentation approaches. We propose a concept of object-by-object learning technique to detect 11 types of facial skin lesions using semantic segmentation methods. Detecting individual skin lesion in a dense group is a challenging task, because of ambiguities in the appearance of the visual data. We observe that there exist co-occurrent visual relations between object classes (e.g., wrinkle and age spot, or papule and whitehead, etc.). In fact, rich contextual information significantly helps to handle the issue. Therefore, we propose REthinker blocks that are composed of the locally constructed convLSTM/Conv3D layers and SE module as a one-shot attention mechanism whose responsibility is to increase network's sensitivity in the local and global contextual representation that supports to capture ambiguously appeared objects and co-occurrence interactions between object classes. Experiments show that our proposed model reached MIoU of 79.46% on the test of a prepared dataset, representing a 15.34% improvement over Deeplab v3+ (MIoU of 64.12%).
Recent Multiple Object Tracking (MOT) methods have gradually attempted to integrate object detection and instance re-identification (Re-ID) into a united network to form a one-stage solution. Typically, these methods use two separated branches within a single network to accomplish detection and Re-ID respectively without studying the inter-relationship between them, which inevitably impedes the tracking performance. In this paper, we propose an online multi-object tracking framework based on a hierarchical single-branch network to solve this problem. Specifically, the proposed single-branch network utilizes an improved Hierarchical Online In-stance Matching (iHOIM) loss to explicitly model the inter-relationship between object detection and Re-ID. Our novel iHOIM loss function unifies the objectives of the two sub-tasks and encourages better detection performance and feature learning even in extremely crowded scenes. Moreover, we propose to introduce the object positions, predicted by a motion model, as region proposals for subsequent object detection, where the intuition is that detection results and motion predictions can complement each other in different scenarios. Experimental results on MOT16 and MOT20 datasets show that we can achieve state-of-the-art tracking performance, and the ablation study verifies the effectiveness of each proposed component.
Multiple object tracking is to give each object an id in the video. The difficulty is how to match the predicted objects and detected objects in same frames. Matching features include appearance features, location features, etc. These features of the predicted object are basically based on some previous frames. However, few papers describe the relationship in the time domain between the previous frame features and the current frame features.In this paper, we proposed a time domain graph convolutional network for multiple objects tracking.The model is mainly divided into two parts, we first use convolutional neural network (CNN) to extract pedestrian appearance feature, which is a normal operation processing image in deep learning, then we use GCN to model some past frames' appearance feature to get the prediction appearance feature of the current frame. Due to this extension, we can get the pose features of the current frame according to the relationship between some frames in the past. Experimental evaluation shows that our extensions improve the MOTA by 1.3 on the MOT16, achieving overall competitive performance at high frame rates.
The task of estimating the 6D pose of an object from RGB images can be broken down into two main steps: an initial pose estimation step, followed by a refinement procedure to correctly register the object and its observation. In this paper, we propose a new method for 6D pose estimation refinement from RGB images. To achieve high accuracy of the final estimate, the observation and a rendered model need to be aligned. Our main insight is that after the initial pose estimate, it is important to pay attention to distinct spatial features of the object in order to improve the estimation accuracy during alignment. Furthermore, parts of the object that are occluded in the image should be given less weight during the alignment process. Most state-of-the-art refinement approaches do not allow for this fine-grained reasoning and can not fully leverage the structure of the problem. In contrast, we propose a novel neural network architecture built around a spatial attention mechanism that identifies and leverages information about spatial details during pose refinement. We experimentally show that this approach learns to attend to salient spatial features and learns to ignore occluded parts of the object, leading to better pose estimation across datasets. We conduct experiments on standard benchmark datasets for 6D pose estimation (LineMOD and Occlusion LineMOD) and outperform previous state-of-the-art methods.
We present STaR, a novel method that performs Self-supervised Tracking and Reconstruction of dynamic scenes with rigid motion from multi-view RGB videos without any manual annotation. Recent work has shown that neural networks are surprisingly effective at the task of compressing many views of a scene into a learned function which maps from a viewing ray to an observed radiance value via volume rendering. Unfortunately, these methods lose all their predictive power once any object in the scene has moved. In this work, we explicitly model rigid motion of objects in the context of neural representations of radiance fields. We show that without any additional human specified supervision, we can reconstruct a dynamic scene with a single rigid object in motion by simultaneously decomposing it into its two constituent parts and encoding each with its own neural representation. We achieve this by jointly optimizing the parameters of two neural radiance fields and a set of rigid poses which align the two fields at each frame. On both synthetic and real world datasets, we demonstrate that our method can render photorealistic novel views, where novelty is measured on both spatial and temporal axes. Our factored representation furthermore enables animation of unseen object motion.
Deep learning based object detectors are commonly deployed on mobile devices to solve a variety of tasks. For maximum accuracy, each detector is usually trained to solve one single specific task, and comes with a completely independent set of parameters. While this guarantees high performance, it is also highly inefficient, as each model has to be separately downloaded and stored. In this paper we address the question: can task-specific detectors be trained and represented as a shared set of weights, plus a very small set of additional weights for each task? The main contributions of this paper are the following: 1) we perform the first systematic study of parameter-efficient transfer learning techniques for object detection problems; 2) we propose a technique to learn a model patch with a size that is dependent on the difficulty of the task to be learned, and validate our approach on 10 different object detection tasks. Our approach achieves similar accuracy as previously proposed approaches, while being significantly more compact.