Rui Huang

Haojun Jiang, Jianke Zhang, Rui Huang et al. · tsinghua

Vision-language retrieval is an important multi-modal learning topic, where the goal is to retrieve the most relevant visual candidate for a given text query. Recently, pre-trained models, e.g., CLIP, show great potential on retrieval tasks. However, as pre-trained models are scaling up, fully fine-tuning them on donwstream retrieval datasets has a high risk of overfitting. Moreover, in practice, it would be costly to train and store a large model for each task. To overcome the above issues, we present a novel Cross-Modal Adapter for parameter-efficient transfer learning. Inspired by adapter-based methods, we adjust the pre-trained model with a few parameterization layers. However, there are two notable differences. First, our method is designed for the multi-modal domain. Secondly, it allows encoder-level implicit cross-modal interactions between vision and language encoders. Although surprisingly simple, our approach has three notable benefits: (1) reduces the vast majority of fine-tuned parameters, (2) saves training time, and (3) allows all the pre-trained parameters to be fixed, enabling the pre-trained model to be shared across datasets. Extensive experiments demonstrate that, without bells and whistles, our approach outperforms adapter-based methods on image-text retrieval datasets (MSCOCO, Flickr30K) and video-text retrieval datasets (MSR-VTT, DiDeMo, and ActivityNet).

Rui Huang, Xuran Pan, Henry Zheng et al.

Recent advancements in vision-language pre-training (e.g. CLIP) have shown that vision models can benefit from language supervision. While many models using language modality have achieved great success on 2D vision tasks, the joint representation learning of 3D point cloud with text remains under-explored due to the difficulty of 3D-Text data pair acquisition and the irregularity of 3D data structure. In this paper, we propose a novel Text4Point framework to construct language-guided 3D point cloud models. The key idea is utilizing 2D images as a bridge to connect the point cloud and the language modalities. The proposed Text4Point follows the pre-training and fine-tuning paradigm. During the pre-training stage, we establish the correspondence of images and point clouds based on the readily available RGB-D data and use contrastive learning to align the image and point cloud representations. Together with the well-aligned image and text features achieved by CLIP, the point cloud features are implicitly aligned with the text embeddings. Further, we propose a Text Querying Module to integrate language information into 3D representation learning by querying text embeddings with point cloud features. For fine-tuning, the model learns task-specific 3D representations under informative language guidance from the label set without 2D images. Extensive experiments demonstrate that our model shows consistent improvement on various downstream tasks, such as point cloud semantic segmentation, instance segmentation, and object detection. The code will be available here: https://github.com/LeapLabTHU/Text4Point

Kangfu Mei, Vishal M. Patel, Rui Huang

The ultimate aim of image restoration like denoising is to find an exact correlation between the noisy and clear image domains. But the optimization of end-to-end denoising learning like pixel-wise losses is performed in a sample-to-sample manner, which ignores the intrinsic correlation of images, especially semantics. In this paper, we introduce the Deep Semantic Statistics Matching (D2SM) Denoising Network. It exploits semantic features of pretrained classification networks, then it implicitly matches the probabilistic distribution of clear images at the semantic feature space. By learning to preserve the semantic distribution of denoised images, we empirically find our method significantly improves the denoising capabilities of networks, and the denoised results can be better understood by high-level vision tasks. Comprehensive experiments conducted on the noisy Cityscapes dataset demonstrate the superiority of our method on both the denoising performance and semantic segmentation accuracy. Moreover, the performance improvement observed on our extended tasks including super-resolution and dehazing experiments shows its potentiality as a new general plug-and-play component.

Chaofei Wang, Qisen Yang, Rui Huang et al.

Knowledge distillation is an effective approach to learn compact models (students) with the supervision of large and strong models (teachers). As empirically there exists a strong correlation between the performance of teacher and student models, it is commonly believed that a high performing teacher is preferred. Consequently, practitioners tend to use a well trained network or an ensemble of them as the teacher. In this paper, we make an intriguing observation that an intermediate model, i.e., a checkpoint in the middle of the training procedure, often serves as a better teacher compared to the fully converged model, although the former has much lower accuracy. More surprisingly, a weak snapshot ensemble of several intermediate models from a same training trajectory can outperform a strong ensemble of independently trained and fully converged models, when they are used as teachers. We show that this phenomenon can be partially explained by the information bottleneck principle: the feature representations of intermediate models can have higher mutual information regarding the input, and thus contain more "dark knowledge" for effective distillation. We further propose an optimal intermediate teacher selection algorithm based on maximizing the total task-related mutual information. Experiments verify its effectiveness and applicability.

Jincheng Huang, Ping Li, Rui Huang et al.

Graph representation learning aim at integrating node contents with graph structure to learn nodes/graph representations. Nevertheless, it is found that many existing graph learning methods do not work well on data with high heterophily level that accounts for a large proportion of edges between different class labels. Recent efforts to this problem focus on improving the message passing mechanism. However, it remains unclear whether heterophily truly does harm to the performance of graph neural networks (GNNs). The key is to unfold the relationship between a node and its immediate neighbors, e.g., are they heterophilous or homophilious? From this perspective, here we study the role of heterophily in graph representation learning before/after the relationships between connected nodes are disclosed. In particular, we propose an end-to-end framework that both learns the type of edges (i.e., heterophilous/homophilious) and leverage edge type information to improve the expressiveness of graph neural networks. We implement this framework in two different ways. Specifically, to avoid messages passing through heterophilous edges, we can optimize the graph structure to be homophilious by dropping heterophilous edges identified by an edge classifier. Alternatively, it is possible to exploit the information about the presence of heterophilous neighbors for feature learning, so a hybrid message passing approach is devised to aggregate homophilious neighbors and diversify heterophilous neighbors based on edge classification. Extensive experiments demonstrate the remarkable performance improvement of GNNs with the proposed framework on multiple datasets across the full spectrum of homophily level.

Jianwei Zhao, Qiang Zhai, Pengbo Zhao et al.

Geolocation is a fundamental component of route planning and navigation for unmanned vehicles, but GNSS-based geolocation fails under denial-of-service conditions. Cross-view geo-localization (CVGL), which aims to estimate the geographical location of the ground-level camera by matching against enormous geo-tagged aerial (\emph{e.g.}, satellite) images, has received lots of attention but remains extremely challenging due to the drastic appearance differences across aerial-ground views. In existing methods, global representations of different views are extracted primarily using Siamese-like architectures, but their interactive benefits are seldom taken into account. In this paper, we present a novel approach using cross-view knowledge generative techniques in combination with transformers, namely mutual generative transformer learning (MGTL), for CVGL. Specifically, by taking the initial representations produced by the backbone network, MGTL develops two separate generative sub-modules -- one for aerial-aware knowledge generation from ground-view semantics and vice versa -- and fully exploits the entirely mutual benefits through the attention mechanism. Moreover, to better capture the co-visual relationships between aerial and ground views, we introduce a cascaded attention masking algorithm to further boost accuracy. Extensive experiments on challenging public benchmarks, \emph{i.e.}, {CVACT} and {CVUSA}, demonstrate the effectiveness of the proposed method which sets new records compared with the existing state-of-the-art models.

Jiahui Zhang, Shitao Tang, Kejie Qiu et al.

Visual relocalization has been a widely discussed problem in 3D vision: given a pre-constructed 3D visual map, the 6 DoF (Degrees-of-Freedom) pose of a query image is estimated. Relocalization in large-scale indoor environments enables attractive applications such as augmented reality and robot navigation. However, appearance changes fast in such environments when the camera moves, which is challenging for the relocalization system. To address this problem, we propose a virtual view synthesis-based approach, RenderNet, to enrich the database and refine poses regarding this particular scenario. Instead of rendering real images which requires high-quality 3D models, we opt to directly render the needed global and local features of virtual viewpoints and apply them in the subsequent image retrieval and feature matching operations respectively. The proposed method can largely improve the performance in large-scale indoor environments, e.g., achieving an improvement of 7.1\% and 12.2\% on the Inloc dataset.

Rui Huang, Ze Huang, Songzhi Su

Visual Place Recognition is an essential component of systems for camera localization and loop closure detection, and it has attracted widespread interest in multiple domains such as computer vision, robotics and AR/VR. In this work, we propose a faster, lighter and stronger approach that can generate models with fewer parameters and can spend less time in the inference stage. We designed RepVGG-lite as the backbone network in our architecture, it is more discriminative than other general networks in the Place Recognition task. RepVGG-lite has more speed advantages while achieving higher performance. We extract only one scale patch-level descriptors from global descriptors in the feature extraction stage. Then we design a trainable feature matcher to exploit both spatial relationships of the features and their visual appearance, which is based on the attention mechanism. Comprehensive experiments on challenging benchmark datasets demonstrate the proposed method outperforming recent other state-of-the-art learned approaches, and achieving even higher inference speed. Our system has 14 times less params than Patch-NetVLAD, 6.8 times lower theoretical FLOPs, and run faster 21 and 33 times in feature extraction and feature matching. Moreover, the performance of our approach is 0.5\% better than Patch-NetVLAD in Recall@1. We used subsets of Mapillary Street Level Sequences dataset to conduct experiments for all other challenging conditions.

Abhijit Mahabal, Jiyun Luo, Rui Huang et al.

Creating a taxonomy of interests is expensive and human-effort intensive: not only do we need to identify nodes and interconnect them, in order to use the taxonomy, we must also connect the nodes to relevant entities such as users, pins, and queries. Connecting to entities is challenging because of ambiguities inherent to language but also because individual interests are dynamic and evolve. Here, we offer an alternative approach that begins with bottom-up discovery of $μ$-topics called pincepts. The discovery process itself connects these $μ$-topics dynamically with relevant queries, pins, and users at high precision, automatically adapting to shifting interests. Pincepts cover all areas of user interest and automatically adjust to the specificity of user interests and are thus suitable for the creation of various kinds of taxonomies. Human experts associate taxonomy nodes with $μ$-topics (on average, 3 $μ$-topics per node), and the $μ$-topics offer a high-level data layer that allows quick definition, immediate inspection, and easy modification. Even more powerfully, $μ$-topics allow easy exploration of nearby semantic space, enabling curators to spot and fill gaps. Curators' domain knowledge is heavily leveraged and we thus don't need untrained mechanical Turks, allowing further cost reduction. These $μ$-topics thus offer a satisfactory "symbolic" stratum over which to define taxonomies. We have successfully applied this technique for very rapidly iterating on and launching the home decor and fashion styles taxonomy for style-based personalization, prominently featured at the top of Pinterest search results, at 94% precision, improving search success rate by 34.8% as well as boosting long clicks and pin saves.

Long Xu, Yongquan Chen, Rui Huang et al.

Click-point-based interactive segmentation has received widespread attention due to its efficiency. However, it's hard for existing algorithms to obtain precise and robust responses after multiple clicks. In this case, the segmentation results tend to have little change or are even worse than before. To improve the robustness of the response, we propose a structured click intent model based on graph neural networks, which adaptively obtains graph nodes via the global similarity of user-clicked Transformer tokens. Then the graph nodes will be aggregated to obtain structured interaction features. Finally, the dual cross-attention will be used to inject structured interaction features into vision Transformer features, thereby enhancing the control of clicks over segmentation results. Extensive experiments demonstrated the proposed algorithm can serve as a general structure in improving Transformer-based interactive segmenta?tion performance. The code and data will be released at https://github.com/hahamyt/scc.

Hualie Jiang, Laiyan Ding, Junjie Hu et al.

Unsupervised learning of depth from indoor monocular videos is challenging as the artificial environment contains many textureless regions. Fortunately, the indoor scenes are full of specific structures, such as planes and lines, which should help guide unsupervised depth learning. This paper proposes PLNet that leverages the plane and line priors to enhance the depth estimation. We first represent the scene geometry using local planar coefficients and impose the smoothness constraint on the representation. Moreover, we enforce the planar and linear consistency by randomly selecting some sets of points that are probably coplanar or collinear to construct simple and effective consistency losses. To verify the proposed method's effectiveness, we further propose to evaluate the flatness and straightness of the predicted point cloud on the reliable planar and linear regions. The regularity of these regions indicates quality indoor reconstruction. Experiments on NYU Depth V2 and ScanNet show that PLNet outperforms existing methods. The code is available at \url{https://github.com/HalleyJiang/PLNet}.

Rui Huang, Songyou Peng, Ayca Takmaz et al.

Current 3D scene segmentation methods are heavily dependent on manually annotated 3D training datasets. Such manual annotations are labor-intensive, and often lack fine-grained details. Importantly, models trained on this data typically struggle to recognize object classes beyond the annotated classes, i.e., they do not generalize well to unseen domains and require additional domain-specific annotations. In contrast, 2D foundation models demonstrate strong generalization and impressive zero-shot abilities, inspiring us to incorporate these characteristics from 2D models into 3D models. Therefore, we explore the use of image segmentation foundation models to automatically generate training labels for 3D segmentation. We propose Segment3D, a method for class-agnostic 3D scene segmentation that produces high-quality 3D segmentation masks. It improves over existing 3D segmentation models (especially on fine-grained masks), and enables easily adding new training data to further boost the segmentation performance -- all without the need for manual training labels.

Yung-Hsu Yang, Luigi Piccinelli, Mattia Segu et al.

Monocular 3D object detection is valuable for various applications such as robotics and AR/VR. Existing methods are confined to closed-set settings, where the training and testing sets consist of the same scenes and/or object categories. However, real-world applications often introduce new environments and novel object categories, posing a challenge to these methods. In this paper, we address monocular 3D object detection in an open-set setting and introduce the first end-to-end 3D Monocular Open-set Object Detector (3D-MOOD). We propose to lift the open-set 2D detection into 3D space through our designed 3D bounding box head, enabling end-to-end joint training for both 2D and 3D tasks to yield better overall performance. We condition the object queries with geometry prior and overcome the generalization for 3D estimation across diverse scenes. To further improve performance, we design the canonical image space for more efficient cross-dataset training. We evaluate 3D-MOOD on both closed-set settings (Omni3D) and open-set settings (Omni3D to Argoverse 2, ScanNet), and achieve new state-of-the-art results. Code and models are available at royyang0714.github.io/3D-MOOD.

Chunjiang Ge, Rui Huang, Mixue Xie et al.

Unsupervised domain adaption (UDA) aims to adapt models learned from a well-annotated source domain to a target domain, where only unlabeled samples are given. Current UDA approaches learn domain-invariant features by aligning source and target feature spaces. Such alignments are imposed by constraints such as statistical discrepancy minimization or adversarial training. However, these constraints could lead to the distortion of semantic feature structures and loss of class discriminability. In this paper, we introduce a novel prompt learning paradigm for UDA, named Domain Adaptation via Prompt Learning (DAPL). In contrast to prior works, our approach makes use of pre-trained vision-language models and optimizes only very few parameters. The main idea is to embed domain information into prompts, a form of representations generated from natural language, which is then used to perform classification. This domain information is shared only by images from the same domain, thereby dynamically adapting the classifier according to each domain. By adopting this paradigm, we show that our model not only outperforms previous methods on several cross-domain benchmarks but also is very efficient to train and easy to implement.

Jingxuan Li, Rui Huang, Wei Li et al.

Code completion is widely used by software developers to provide coding suggestions given a partially written code snippet. Apart from the traditional code completion methods, which only support single token completion at minimal positions, recent studies show the ability to provide longer code completion at more flexible positions. However, such frequently triggered and longer completion results reduce the overall precision as they generate more invalid results. Moreover, different studies are mostly incompatible with each other. Thus, it is vital to develop an ensemble framework that can combine results from multiple models to draw merits and offset defects of each model. This paper conducts a coding simulation to collect data from code context and different code completion models and then apply the data in two tasks. First, we introduce an acceptance model which can dynamically control whether to display completion results to the developer. It uses simulation features to predict whether correct results exist in the output of these models. Our best model reduces the percentage of false-positive completion from 55.09% to 17.44%. Second, we design a fusion ranking scheme that can automatically identify the priority of the completion results and reorder the candidates from multiple code completion models. This scheme is flexible in dealing with various models, regardless of the type or the length of their completion results. We integrate this ranking scheme with two frequency models and a GPT-2 styled language model, along with the acceptance model to yield 27.80% and 37.64% increase in TOP1 and TOP5 accuracy, respectively. In addition, we propose a new code completion evaluation metric, Benefit-Cost Ratio(BCR), taking into account the benefit of keystrokes saving and hidden cost of completion list browsing, which is closer to real coder experience scenario.