Few-shot object detection has rapidly progressed owing to the success of meta-learning strategies. However, the requirement of a fine-tuning stage in existing methods is timeconsuming and significantly hinders their usage in real-time applications such as autonomous exploration of low-power robots. To solve this problem, we present a brand new architecture, AirDet, which is free of fine-tuning by learning class agnostic relation with support images. Specifically, we propose a support-guided cross-scale (SCS) feature fusion network to generate object proposals, a global-local relation network (GLR) for shots aggregation, and a relation-based prototype embedding network (R-PEN) for precise localization. Exhaustive experiments are conducted on COCO and PASCAL VOC datasets, where surprisingly, AirDet achieves comparable or even better results than the exhaustively finetuned methods, reaching up to 40-60% improvements on the baseline. To our excitement, AirDet obtains favorable performance on multi-scale objects, especially the small ones. Furthermore, we present evaluation results on real-world exploration tests from the DARPA Subterranean Challenge, which strongly validate the feasibility of AirDet in robotics. The source code, pre-trained models, along with the real world data for exploration, will be made public.
Object encoding and identification are vital for robotic tasks such as autonomous exploration, semantic scene understanding, and re-localization. Previous approaches have attempted to either track objects or generate descriptors for object identification. However, such systems are limited to a "fixed" partial object representation from a single viewpoint. In a robot exploration setup, there is a requirement for a temporally "evolving" global object representation built as the robot observes the object from multiple viewpoints. Furthermore, given the vast distribution of unknown novel objects in the real world, the object identification process must be class-agnostic. In this context, we propose a novel temporal 3D object encoding approach, dubbed AirObject, to obtain global keypoint graph-based embeddings of objects. Specifically, the global 3D object embeddings are generated using a temporal convolutional network across structural information of multiple frames obtained from a graph attention-based encoding method. We demonstrate that AirObject achieves the state-of-the-art performance for video object identification and is robust to severe occlusion, perceptual aliasing, viewpoint shift, deformation, and scale transform, outperforming the state-of-the-art single-frame and sequential descriptors. To the best of our knowledge, AirObject is one of the first temporal object encoding methods.
Autonomous robots frequently need to detect "interesting" scenes to decide on further exploration, or to decide which data to share for cooperation. These scenarios often require fast deployment with little or no training data. Prior work considers "interestingness" based on data from the same distribution. Instead, we propose to develop a method that automatically adapts online to the environment to report interesting scenes quickly. To address this problem, we develop a novel translation-invariant visual memory and design a three-stage architecture for long-term, short-term, and online learning, which enables the system to learn human-like experience, environmental knowledge, and online adaption, respectively. With this system, we achieve an average of 20% higher accuracy than the state-of-the-art unsupervised methods in a subterranean tunnel environment. We show comparable performance to supervised methods for robot exploration scenarios showing the efficacy of our approach. We expect that the presented method will play an important role in the robotic interestingness recognition exploration tasks.
A recommender system predicts users' potential interests in items, where the core is to learn user/item embeddings. Nevertheless, it suffers from the data-sparsity issue, which the cross-domain recommendation can alleviate. However, most prior works either jointly learn the source domain and target domain models, or require side-features. However, jointly training and side features would affect the prediction on the target domain as the learned embedding is dominated by the source domain containing bias information. Inspired by the contemporary arts in pre-training from graph representation learning, we propose a pre-training and fine-tuning diagram for cross-domain recommendation. We devise a novel Pre-training Graph Neural Network for Cross-Domain Recommendation (PCRec), which adopts the contrastive self-supervised pre-training of a graph encoder. Then, we transfer the pre-trained graph encoder to initialize the node embeddings on the target domain, which benefits the fine-tuning of the single domain recommender system on the target domain. The experimental results demonstrate the superiority of PCRec. Detailed analyses verify the superiority of PCRec in transferring information while avoiding biases from source domains.
With the widespread accumulation of observational data, researchers obtain a new direction to learn counterfactual effects in many domains (e.g., health care and computational advertising) without Randomized Controlled Trials(RCTs). However, observational data suffer from inherent missing counterfactual outcomes, and distribution discrepancy between treatment and control groups due to behaviour preference. Motivated by recent advances of representation learning in the field of domain adaptation, we propose a novel framework based on Cycle-Balanced REpresentation learning for counterfactual inference (CBRE), to solve above problems. Specifically, we realize a robust balanced representation for different groups using adversarial training, and meanwhile construct an information loop, such that preserve original data properties cyclically, which reduces information loss when transforming data into latent representation space.Experimental results on three real-world datasets demonstrate that CBRE matches/outperforms the state-of-the-art methods, and it has a great potential to be applied to counterfactual inference.
Despite recent improvement of supervised monocular depth estimation, the lack of high quality pixel-wise ground truth annotations has become a major hurdle for further progress. In this work, we propose a new unsupervised depth estimation method based on pseudo supervision mechanism by training a teacher-student network with knowledge distillation. It strategically integrates the advantages of supervised and unsupervised monocular depth estimation, as well as unsupervised binocular depth estimation. Specifically, the teacher network takes advantage of the effectiveness of binocular depth estimation to produce accurate disparity maps, which are then used as the pseudo ground truth to train the student network for monocular depth estimation. This effectively converts the problem of unsupervised learning to supervised learning. Our extensive experimental results demonstrate that the proposed method outperforms the state-of-the-art on the KITTI benchmark.
Fraud detection problems are usually formulated as a machine learning problem on a graph. Recently, Graph Neural Networks (GNNs) have shown solid performance on fraud detection. The successes of most previous methods heavily rely on rich node features and high-fidelity labels. However, labeled data is scarce in large-scale industrial problems, especially for fraud detection where new patterns emerge from time to time. Meanwhile, node features are also limited due to privacy and other constraints. In this paper, two improvements are proposed: 1) We design a graph transformation method capturing the structural information to facilitate GNNs on non-attributed fraud graphs. 2) We propose a novel graph pre-training strategy to leverage more unlabeled data via contrastive learning. Experiments on a large-scale industrial dataset demonstrate the effectiveness of the proposed framework for fraud detection.
We present a new approach for solving (minimum disagreement) correlation clustering that results in sublinear algorithms with highly efficient time and space complexity for this problem. In particular, we obtain the following algorithms for $n$-vertex $(+/-)$-labeled graphs $G$: -- A sublinear-time algorithm that with high probability returns a constant approximation clustering of $G$ in $O(n\log^2{n})$ time assuming access to the adjacency list of the $(+)$-labeled edges of $G$ (this is almost quadratically faster than even reading the input once). Previously, no sublinear-time algorithm was known for this problem with any multiplicative approximation guarantee. -- A semi-streaming algorithm that with high probability returns a constant approximation clustering of $G$ in $O(n\log{n})$ space and a single pass over the edges of the graph $G$ (this memory is almost quadratically smaller than input size). Previously, no single-pass algorithm with $o(n^2)$ space was known for this problem with any approximation guarantee. The main ingredient of our approach is a novel connection to sparse-dense graph decompositions that are used extensively in the graph coloring literature. To our knowledge, this connection is the first application of these decompositions beyond graph coloring, and in particular for the correlation clustering problem, and can be of independent interest.
Human pose estimation in unconstrained images and videos is a fundamental computer vision task. To illustrate the evolutionary path in technique, in this survey we summarize representative human pose methods in a structured taxonomy, with a particular focus on deep learning models and single-person image setting. Specifically, we examine and survey all the components of a typical human pose estimation pipeline, including data augmentation, model architecture and backbone, supervision representation, post-processing, standard datasets, evaluation metrics. To envisage the future directions, we finally discuss the key unsolved problems and potential trends for human pose estimation.