Expressions and facial action units (AUs) are two levels of facial behavior descriptors. Expression auxiliary information has been widely used to improve the AU detection performance. However, most existing expression representations can only describe pre-determined discrete categories (e.g., Angry, Disgust, Happy, Sad, etc.) and cannot capture subtle expression transformations like AUs. In this paper, we propose a novel fine-grained \textsl{Global Expression representation Encoder} to capture subtle and continuous facial movements, to promote AU detection. To obtain such a global expression representation, we propose to train an expression embedding model on a large-scale expression dataset according to global expression similarity. Moreover, considering the local definition of AUs, it is essential to extract local AU features. Therefore, we design a \textsl{Local AU Features Module} to generate local facial features for each AU. Specifically, it consists of an AU feature map extractor and a corresponding AU mask extractor. First, the two extractors transform the global expression representation into AU feature maps and masks, respectively. Then, AU feature maps and their corresponding AU masks are multiplied to generate AU masked features focusing on local facial region. Finally, the AU masked features are fed into an AU classifier for judging the AU occurrence. Extensive experiment results demonstrate the superiority of our proposed method. Our method validly outperforms previous works and achieves state-of-the-art performances on widely-used face datasets, including BP4D, DISFA, and BP4D+.
Since Facial Action Unit (AU) annotations require domain expertise, common AU datasets only contain a limited number of subjects. As a result, a crucial challenge for AU detection is addressing identity overfitting. We find that AUs and facial expressions are highly associated, and existing facial expression datasets often contain a large number of identities. In this paper, we aim to utilize the expression datasets without AU labels to facilitate AU detection. Specifically, we develop a novel AU detection framework aided by the Global-Local facial Expressions Embedding, dubbed GLEE-Net. Our GLEE-Net consists of three branches to extract identity-independent expression features for AU detection. We introduce a global branch for modeling the overall facial expression while eliminating the impacts of identities. We also design a local branch focusing on specific local face regions. The combined output of global and local branches is firstly pre-trained on an expression dataset as an identity-independent expression embedding, and then finetuned on AU datasets. Therefore, we significantly alleviate the issue of limited identities. Furthermore, we introduce a 3D global branch that extracts expression coefficients through 3D face reconstruction to consolidate 2D AU descriptions. Finally, a Transformer-based multi-label classifier is employed to fuse all the representations for AU detection. Extensive experiments demonstrate that our method significantly outperforms the state-of-the-art on the widely-used DISFA, BP4D and BP4D+ datasets.
Deep learning for image super-resolution (SR) has been investigated by numerous researchers in recent years. Most of the works concentrate on effective block designs and improve the network representation but lack interpretation. There are also iterative optimization-inspired networks for image SR, which take the solution step as a whole without giving an explicit optimization step. This paper proposes an unfolding iterative shrinkage thresholding algorithm (ISTA) inspired network for interpretable image SR. Specifically, we analyze the problem of image SR and propose a solution based on the ISTA method. Inspired by the mathematical analysis, the ISTA block is developed to conduct the optimization in an end-to-end manner. To make the exploration more effective, a multi-scale exploitation block and multi-scale attention mechanism are devised to build the ISTA block. Experimental results show the proposed ISTA-inspired restoration network (ISTAR) achieves competitive or better performances than other optimization-inspired works with fewer parameters and lower computation complexity.
This paper studies real-time motion planning and control for ball bumping motion with quadruped robots. To enable the quadruped to bump the flying ball with different initializations, we develop a nonlinear trajectory optimization-based planning scheme that jointly identifies the take-off time and state to achieve accurate ball hitting during the flight phase. Such a planning scheme employs a two-dimensional single rigid body model that achieves a satisfactory balance between accuracy and efficiency for the highly time-sensitive task. To precisely execute the planned motion, the tracking controller needs to incorporate the strict time-state constraint imposed on the take-off and ball-hitting events. To this end, we develop an improved model predictive controller that respects the critical time-state constraints. The proposed planning and control framework is validated with a real Aliengo robot. Experiments show that the problem planning approach can be computed in approximately 60ms on average, enabling successful accomplishment of the ball bumping motion with various initializations in real time.
Current lane detection methods are struggling with the invisibility lane issue caused by heavy shadows, severe road mark degradation, and serious vehicle occlusion. As a result, discriminative lane features can be barely learned by the network despite elaborate designs due to the inherent invisibility of lanes in the wild. In this paper, we target at finding an enhanced feature space where the lane features are distinctive while maintaining a similar distribution of lanes in the wild. To achieve this, we propose a novel Repainting and Imitating Learning (RIL) framework containing a pair of teacher and student without any extra data or extra laborious labeling. Specifically, in the repainting step, an enhanced ideal virtual lane dataset is built in which only the lane regions are repainted while non-lane regions are kept unchanged, maintaining the similar distribution of lanes in the wild. The teacher model learns enhanced discriminative representation based on the virtual data and serves as the guidance for a student model to imitate. In the imitating learning step, through the scale-fusing distillation module, the student network is encouraged to generate features that mimic the teacher model both on the same scale and cross scales. Furthermore, the coupled adversarial module builds the bridge to connect not only teacher and student models but also virtual and real data, adjusting the imitating learning process dynamically. Note that our method introduces no extra time cost during inference and can be plug-and-play in various cutting-edge lane detection networks. Experimental results prove the effectiveness of the RIL framework both on CULane and TuSimple for four modern lane detection methods. The code and model will be available soon.
The SoccerNet 2022 challenges were the second annual video understanding challenges organized by the SoccerNet team. In 2022, the challenges were composed of 6 vision-based tasks: (1) action spotting, focusing on retrieving action timestamps in long untrimmed videos, (2) replay grounding, focusing on retrieving the live moment of an action shown in a replay, (3) pitch localization, focusing on detecting line and goal part elements, (4) camera calibration, dedicated to retrieving the intrinsic and extrinsic camera parameters, (5) player re-identification, focusing on retrieving the same players across multiple views, and (6) multiple object tracking, focusing on tracking players and the ball through unedited video streams. Compared to last year's challenges, tasks (1-2) had their evaluation metrics redefined to consider tighter temporal accuracies, and tasks (3-6) were novel, including their underlying data and annotations. More information on the tasks, challenges and leaderboards are available on https://www.soccer-net.org. Baselines and development kits are available on https://github.com/SoccerNet.
Few-shot classification requires deep neural networks to learn generalized representations only from limited training images, which is challenging but significant in low-data regimes. Recently, CLIP-based methods have shown promising few-shot performance benefited from the contrastive language-image pre-training. Based on this point, we question if the large-scale pre-training can alleviate the few-shot data deficiency and also assist the representation learning by the pre-learned knowledge. In this paper, we propose CoMo, a Collaboration of pre-trained Models that incorporates diverse prior knowledge from various pre-training paradigms for better few-shot learning. Our CoMo includes: CLIP's language-contrastive knowledge, DINO's vision-contrastive knowledge, and DALL-E's language-generative knowledge. Specifically, CoMo works in two aspects: few-shot data expansion and diverse knowledge ensemble. For one, we generate synthetic images via zero-shot DALL-E to enrich the few-shot training data without any manpower. For the other, we introduce a learnable Multi-Knowledge Adapter (MK-Adapter) to adaptively blend the predictions from CLIP and DINO. By such collaboration, CoMo can fully unleash the potential of different pre-training methods and unify them to perform state-of-the-art for few-shot classification. We conduct extensive experiments on 11 datasets to demonstrate the superiority and generalization ability of our approach.
This paper studies the class of scenario-based safety testing algorithms in the black-box safety testing configuration. For algorithms sharing the same state-action set coverage with different sampling distributions, it is commonly believed that prioritizing the exploration of high-risk state-actions leads to a better sampling efficiency. Our proposal disputes the above intuition by introducing an impossibility theorem that provably shows all safety testing algorithms of the aforementioned difference perform equally well with the same expected sampling efficiency. Moreover, for testing algorithms covering different sets of state-actions, the sampling efficiency criterion is no longer applicable as different algorithms do not necessarily converge to the same termination condition. We then propose a testing aggressiveness definition based on the almost safe set concept along with an unbiased and efficient algorithm that compares the aggressiveness between testing algorithms. Empirical observations from the safety testing of bipedal locomotion controllers and vehicle decision-making modules are also presented to support the proposed theoretical implications and methodologies.
Open-world object detection, as a more general and challenging goal, aims to recognize and localize objects described by arbitrary category names. The recent work GLIP formulates this problem as a grounding problem by concatenating all category names of detection datasets into sentences, which leads to inefficient interaction between category names. This paper presents DetCLIP, a paralleled visual-concept pre-training method for open-world detection by resorting to knowledge enrichment from a designed concept dictionary. To achieve better learning efficiency, we propose a novel paralleled concept formulation that extracts concepts separately to better utilize heterogeneous datasets (i.e., detection, grounding, and image-text pairs) for training. We further design a concept dictionary~(with descriptions) from various online sources and detection datasets to provide prior knowledge for each concept. By enriching the concepts with their descriptions, we explicitly build the relationships among various concepts to facilitate the open-domain learning. The proposed concept dictionary is further used to provide sufficient negative concepts for the construction of the word-region alignment loss\, and to complete labels for objects with missing descriptions in captions of image-text pair data. The proposed framework demonstrates strong zero-shot detection performances, e.g., on the LVIS dataset, our DetCLIP-T outperforms GLIP-T by 9.9% mAP and obtains a 13.5% improvement on rare categories compared to the fully-supervised model with the same backbone as ours.
Universal domain adaptation (UniDA) aims to transfer the knowledge of common classes from source domain to target domain without any prior knowledge on the label set, which requires to distinguish the unknown samples from the known ones in the target domain. Like the traditional unsupervised domain adaptation problem, the misalignment between two domains exists due to the biased and less-discriminative embedding. Recent methods proposed to complete the domain misalignment by clustering target samples with the nearest neighbors or the prototypes. However, it is dangerous to do so since we do not have any prior knowledge about the distributions of unknown samples which can magnify the misalignment especially when the unknown set is big. Meanwhile, other existing classifier-based methods could easily produce overconfident predictions of unknown samples because of the supervised objective in source domain leading the whole model to be biased towards the common classes in the target domain. Therefore, we propose a novel non-parameter unknown samples detection method based on mapping the samples in the original feature space into a reliable linear sub-space which makes data points more sparse to reduce the misalignment between unknown samples and source samples. Moreover, unlike the recent methods applying extra parameters to improve the classification of unknown samples, this paper well balances the confidence values of both known and unknown samples through an unknown-adaptive margin loss which can control the gradient updating of the classifier learning on supervised source samples depending on the confidence level of detected unknown samples at current step. Finally, experiments on four public datasets demonstrate that our method significantly outperforms existing state-of-the-art methods.