Weakly supervised visual grounding aims to locate a region in an image based on an input query sentence, without access to the mapping between image regions and queries during training. Current methods treat spatial grounding as an object retrieval task, relying on cross-modal similarity scores for proposal selection. However, they fail to address model overfitting caused by unreliable cross-modal similarity scores. To overcome this, we first propose the Confidence-aware Pseudo-label Learning (CPL) framework. CPL first generates diverse pseudo queries for region proposals, and then establishes reliable associations for model training based on the uni-modal similarity score. Secondly, we propose a cross-modal verification module based on the pretrained vision-language model to verify associations. However, the verification module is isolated from the grounding model, so it can only assess associations in a static manner, but not correct the suspicious ones. Finally, we introduce CPL++ to make two-fold improvements. For one thing, we upgrade the verification process based on the model's grounding loss value to identify suspicious associations dynamically and selectively leverage them in the training. For another, we propose a self-supervised association correction module to rectify suspicious associations, thereby mitigating the risk of error propagation. Experimental results on five datasets demonstrate the superiority of our approach.

Video moment retrieval (VMR) aims to locate the most likely video moment(s) corresponding to a text query in untrimmed videos. Training of existing methods is limited by the lack of diverse and generalisable VMR datasets, hindering their ability to generalise moment-text associations to queries containing novel semantic concepts (unseen both visually and textually in a training source domain). For model generalisation to novel semantics, existing methods rely heavily on assuming to have access to both video and text sentence pairs from a target domain in addition to the source domain pair-wise training data. This is neither practical nor scalable. In this work, we introduce a more generalisable approach by assuming only text sentences describing new semantics are available in model training without having seen any videos from a target domain. To that end, we propose a Fine-grained Video Editing framework, termed FVE, that explores generative video diffusion to facilitate fine-grained video editing from the seen source concepts to the unseen target sentences consisting of new concepts. This enables generative hypotheses of unseen video moments corresponding to the novel concepts in the target domain. This fine-grained generative video diffusion retains the original video structure and subject specifics from the source domain while introducing semantic distinctions of unseen novel vocabularies in the target domain. A critical challenge is how to enable this generative fine-grained diffusion process to be meaningful in optimising VMR, more than just synthesising visually pleasing videos. We solve this problem by introducing a hybrid selection mechanism that integrates three quantitative metrics to selectively incorporate synthetic video moments (novel video hypotheses) as enlarged additions to the original source training data, whilst minimising potential detrimental noise or unnecessary repetitions in the novel synthetic videos harmful to VMR learning. Experiments on three datasets demonstrate the effectiveness of FVE to unseen novel semantic video moment retrieval tasks

Video procedure planning, i.e., planning a sequence of action steps given the video frames of start and goal states, is an essential ability for embodied AI. Recent works utilize Large Language Models (LLMs) to generate enriched action step description texts to guide action step decoding. Although LLMs are introduced these methods decode the action steps into a closed-set of one-hot vectors, limiting the model's capability of generalizing to new steps or tasks. Additionally, fixed action step descriptions based on world-level commonsense may contain noise in specific instances of visual states. In this paper, we propose PlanLLM, a cross-modal joint learning framework with LLMs for video procedure planning. We propose an LLM-Enhanced Planning module which fully uses the generalization ability of LLMs to produce free-form planning output and to enhance action step decoding. We also propose Mutual Information Maximization module to connect world-level commonsense of step descriptions and sample-specific information of visual states, enabling LLMs to employ the reasoning ability to generate step sequences. With the assistance of LLMs, our method can both closed-set and open vocabulary procedure planning tasks. Our PlanLLM achieves superior performance on three benchmarks, demonstrating the effectiveness of our designs.

The development of Text-to-Video (T2V) generation has made motion transfer possible, enabling the control of video motion based on existing footage. However, current methods have two limitations: 1) struggle to handle multi-subjects videos, failing to transfer specific subject motion; 2) struggle to preserve the diversity and accuracy of motion as transferring to subjects with varying shapes. To overcome these, we introduce ConMo, a zero-shot framework that disentangle and recompose the motions of subjects and camera movements. ConMo isolates individual subject and background motion cues from complex trajectories in source videos using only subject masks, and reassembles them for target video generation. This approach enables more accurate motion control across diverse subjects and improves performance in multi-subject scenarios. Additionally, we propose soft guidance in the recomposition stage which controls the retention of original motion to adjust shape constraints, aiding subject shape adaptation and semantic transformation. Unlike previous methods, ConMo unlocks a wide range of applications, including subject size and position editing, subject removal, semantic modifications, and camera motion simulation. Extensive experiments demonstrate that ConMo significantly outperforms state-of-the-art methods in motion fidelity and semantic consistency. The code is available at https://github.com/Andyplus1/ConMo.

The field of conditional Electrocardiogram(ECG) generation focuses on generating specified ECGs under given conditions for medical purposes. Existing methods are typically based on conditions of simple inputs like text or lead types. However, they struggle to handle the complexity of radar heart sound signals due to the lack of effective feature extraction, which hinders capturing the intricate waveform correlations between radar heart sounds and ECGs. Considering that radar-detected heart sound signals are contactless, the application is of essential value in a real-world deployment like sleep scenarios. Moreover, no prior approaches have addressed this specific task. To tackle this challenge, we propose a novel multi-scale feature fusion network framework, Radar2ECG. This model leverages pre-trained autoencoders for heart sound and ECG signals, aligning and integrating multi-layer features through a bottleneck structure to enhance receptive fields and reduce redundant features, thereby capturing the correlations between heart sounds and ECGs. Finally, we employ knowledge distillation to transfer knowledge from the ECG decoder to the heart sound decoder. We present three anomaly type datasets and extensive experiments conducted on both normal and abnormal datasets demonstrate that our method outperforms existing models in both accuracy and robustness. The multi-scale feature fusion significantly improves performance, showcasing strong potential in ECG generation and heart sound anomaly detection tasks.

Visual Robot Manipulation (VRM) aims to enable a robot to follow natural language instructions based on robot states and visual observations, and therefore requires costly multi-modal data. To compensate for the deficiency of robot data, existing approaches have employed vision-language pretraining with large-scale data. However, they either utilize web data that differs from robotic tasks, or train the model in an implicit way (e.g., predicting future frames at the pixel level), thus showing limited generalization ability under insufficient robot data. In this paper, we propose to learn from large-scale human action video datasets in an explicit way (i.e., imitating human actions from hand keypoints), introducing Visual Robot Manipulation with Analogical Reasoning (AR-VRM). To acquire action knowledge explicitly from human action videos, we propose a keypoint Vision-Language Model (VLM) pretraining scheme, enabling the VLM to learn human action knowledge and directly predict human hand keypoints. During fine-tuning on robot data, to facilitate the robotic arm in imitating the action patterns of human motions, we first retrieve human action videos that perform similar manipulation tasks and have similar historical observations , and then learn the Analogical Reasoning (AR) map between human hand keypoints and robot components. Taking advantage of focusing on action keypoints instead of irrelevant visual cues, our method achieves leading performance on the CALVIN benchmark {and real-world experiments}. In few-shot scenarios, our AR-VRM outperforms previous methods by large margins , underscoring the effectiveness of explicitly imitating human actions under data scarcity.

Instruction-based image editing, which aims to modify the image faithfully towards instruction while preserving irrelevant content unchanged, has made advanced progresses. However, there still lacks a comprehensive metric for assessing the editing quality. Existing metrics either require high costs concerning human evaluation, which hinders large-scale evaluation, or adapt from other tasks and lose specified concerns, failing to comprehensively evaluate the modification of instruction and the preservation of irrelevant regions, resulting in biased evaluation. To tackle it, we introduce a new metric Balancing Preservation Modification (BPM), that tailored for instruction-based image editing by explicitly disentangling the image into editing-relevant and irrelevant regions for specific consideration. We first identify and locate editing-relevant regions, followed by a two-tier process to assess editing quality: Region-Aware Judge evaluates whether the position and size of the edited region align with instruction, and Semantic-Aware Judge further assesses the instruction content compliance within editing-relevant regions as well as content preservation within irrelevant regions, yielding comprehensive and interpretable quality assessment. Moreover, the editing-relevant region localization in BPM can be integrated into image editing approaches to improve the editing quality, manifesting its wild application. We verify the effectiveness of BPM metric on comprehensive instruction-editing data, and the re- sults show that we yield the highest alignment with human evaluation compared to existing metrics, indicating efficacy. The code is available at https://joyli-x.github.io/BPM/.

The 3D weakly-supervised visual grounding task aims to localize oriented 3D boxes in point clouds based on natural language descriptions without requiring annotations to guide model learning. This setting presents two primary challenges: category-level ambiguity and instance-level complexity. Category-level ambiguity arises from representing objects of fine-grained categories in a highly sparse point cloud format, making category distinction challenging. Instance-level complexity stems from multiple instances of the same category coexisting in a scene, leading to distractions during grounding. To address these challenges, we propose a novel weaklysupervised grounding approach that explicitly differentiates between categories and instances. In the category-level branch, we utilize extensive category knowledge from a pre-trained external detector to align object proposal features with sentencelevel category features, thereby enhancing category awareness. In the instance-level branch, we utilize spatial relationship descriptions from language queries to refine object proposal features, ensuring clear differentiation among objects. These designs enable our model to accurately identify target-category objects while distinguishing instances within the same category. Compared to previous methods, our approach achieves state-of-the-art performance on three widely used benchmarks: Nr3D, Sr3D, and ScanRef.

Deep learning-based semantic segmentation models achieve impressive results yet remain limited in handling distribution shifts between training and test data. In this paper, we present SDGPA (Synthetic Data Generation and Progressive Adaptation), a novel method that tackles zero-shot domain adaptive semantic segmentation, in which no target images are available, but only a text description of the target domain’s style is provided. To compensate for the lack of target domain training data, we utilize a pretrained off-the-shelf text-to-image diffusion model, which generates training images by transferring source domain images to target style. Directly editing source domain images introduces noise that harms segmentation because the layout of source images cannot be precisely maintained. To address inaccurate layouts in synthetic data, we propose a method that crops the source image, edits small patches individually, and then merges them back together, which helps improve spatial precision. Recognizing the large domain gap, SDGPA constructs an augmented intermediate domain, leveraging easier adaptation subtasks to enable more stable model adaptation to the target domain. Additionally, to mitigate the impact of noise in synthetic data, we design a progressive adaptation strategy, ensuring robust learning throughout the training process. Extensive experiments demonstrate that our method achieves state-of-the-art performance in zero-shot semantic segmentation. The code is available at https://github.com/ROUJINN/SDGPA

As a fundamental task for indoor scene understanding, 3D object detection has been extensively studied, and the accuracy on indoor point cloud data has been substantially improved. However, existing researches have been conducted on limited datasets, where the training and testing sets share the same distribution. In this paper, we consider the task of adapting indoor 3D object detectors from one dataset to another, presenting a comprehensive benchmark with ScanNet, SUN RGB-D and 3D Front datasets, as well as our newly proposed large-scale datasets ProcTHOR-OD and ProcFront generated by a 3D simulator. Since indoor point cloud datasets are collected and constructed in different ways, the object detectors are likely to overfit to specific factors within each dataset, such as point cloud quality, bounding box layout and instance features. We conduct experiments across datasets on different adaptation scenarios including synthetic-to-real adaptation, point cloud quality adaptation, layout adaptation and instance feature adaptation, analyzing the impact of different domain gaps on 3D object detectors. We also introduce several approaches to improve adaptation performances, providing baselines for domain adaptive indoor 3D object detection, hoping that future works may propose detectors with stronger generalization ability across domains. Our project homepage can be found in https://jeremyzhao1998.github.io/DAVoteNet-release/.

It is inevitably crucial to classify the sleep stage for assessing sleep quality and diagnosing related diseases. However, existing automated diagnosis methods mostly adopt the goldstandard Electroencephalogram (EEG) or other sensing signals from the PolySomnoGraphy (PSG) machine in hospital, which are expensive, importable, and therefore unsuitable for point-ofcare monitoring at home. To enable the sleep stage monitoring at home, in this paper, we integrate the external visual information from infrared (IR) videos with internal bio-electrical knowledge from EEG signals, and propose a novel IR-based sleep monitoring system in IoT scenarios: which can non-contact monitor and predict body movement, respiration rate, and sleep staging tasks based solely on IR video. It is different from previous video classification and multi-modal analysis systems, mainly in that (1) the temporal duration of the IR video is relatively long, reaching 10 hours per night. (2) the semantic gap between the EEG signal and IR video is disparate and much larger than conventional cross-modal data in multimedia analysis such as video and audio. To establish a solid cross-modal benchmark in sleep monitoring, we develop a new dataset (S3IE), which is a large-scale dataset including synchronized IR videos and EEG signals. We also propose a novel cross-modal knowledge distillation method, namely the structure-aware contrastive distillation (SACD) to distill the EEG bio-electrical knowledge to IR visual features. Our SACD achieves the SOTA performances on both our S3IE and other cross-modal distillation benchmarks. We expect to raise more attention and promote more developments for the sleep and, more importantly, the cross-modal distillation from clinical signal/media to conventional media in IoT scenarios.

Conventional Multi-Source Free Domain Adaptation (MSFDA) assumes that each source domain provides a single source model, and all source models adopt a uniform architecture. This paper introduces Zoo-MSFDA, a more general setting that allows each source domain to offer a zoo of multiple source models with different architectures. While it enriches the source knowledge, Zoo-MSFDA risks being dominated by suboptimal/harmful models. To address this issue, we theoretically analyze the model selection problem in Zoo-MSFDA, and introduce two principles: transferability principle and diversity principle. Recognizing the challenge of measuring transferability, we subsequently propose a novel Source-Free Unsupervised Transferability Estimation (SUTE). It enables assessing and comparing transferability across multiple source models with different architectures under domain shift, without requiring target labels and source data. Based on above, we introduce a Selection, Ensemble, and Adaptation (SEA) framework to address Zoo-MSFDA, which consists of: 1) source models selection based on the proposed principles and SUTE; 2) ensemble construction based on SUTE-estimated transferability; 3) target-domain adaptation of the ensemble model. Evaluations demonstrate that our SEA framework, with the introduced Zoo-MSFDA setting, significantly improves adaptation performance in 2D image classification tasks. Additionally, our SUTE achieves state-of-the-art performance in transferability estimation.

In deep learning, initializing models with pre-trained weights has become the de facto practice for various downstream tasks. Many unsupervised domain adaptation (UDA) methods typically adopt a backbone pre-trained on ImageNet, and focus on reducing the source-target domain discrepancy. However, the impact of pre-training on adaptation received little attention. In this study, we delve into UDA from the novel perspective of pre-training. We first demonstrate the impact of pre-training by analyzing the dynamic distribution discrepancies between pre-training data domain and the source/ target domain during adaptation. Then, we reveal that the target error also stems from the pre-training in the following two factors: 1) empirically, target error arises from the gradually degenerative pre-trained knowledge during adaptation; 2) theoretically, the error bound depends on difference between the gradient of loss function, i.e., on the target domain and pre-training data domain. To address these two issues, we redefine UDA as a three-domain problem, i.e., source domain, target domain, and pre-training data domain; then we propose a novel framework, named TriDA. We maintain the pre-trained knowledge and improve the error bound by incorporating pre-training data into adaptation for both vanilla UDA and source-free UDA scenarios. For efficiency, we introduce a selection strategy for pre-training data, and offer a solution with synthesized images when pre-training data is unavailable during adaptation. Notably, TriDA is effective even with a small amount of pre-training or synthesized images, and seamlessly complements the two scenario UDA methods, demonstrating state-of-the-art performance across multiple benchmarks. We hope our work provides new insights for better understanding and application of domain adaptation.

In 3D Visual Question Answering (3D VQA), the scarcity of fully annotated data and limited visual content diversity hampers the generalization to novel scenes and 3D concepts (e.g., only around 800 scenes are utilized in ScanQA and SQA dataset). Current approaches resort supplement 3D reasoning with 2D information. However, these methods face challenges: either they use top-down 2D views that introduce overly complex and sometimes question-irrelevant visual clues, or they rely on globally aggregated scene/image-level representations from 2D VLMs, losing the fine-grained vision-language correlations. To overcome these limitations, our approach utilizes question-conditional 2D view selection procedure, pinpointing semantically relevant 2D inputs for crucial visual clues. We then integrate this 2D knowledge into the 3D-VQA system via a two-branch Transformer structure. This structure, featuring a Twin-Transformer design, compactly combines 2D and 3D modalities and captures fine-grained correlations between modalities, allowing them mutually augmenting each other. Integrating proposed mechanisms above, we present BridgeQA, that offers a fresh perspective on multi-modal transformer-based architectures for 3D-VQA. Experiments validate that BridgeQA achieves state-of-the-art on 3D-VQA datasets and significantly outperforms existing solutions. Code is available at https://github.com/matthewdm0816/BridgeQA.

The Novel Category Discovery problem aims to cluster an unlabeled set with the help of a labeled set consisting of disjoint but related classes. However, existing models treat class names as discrete one-hot labels and ignore the semantic understanding of these classes. In this paper, we propose a new setting named Semantic-guided Novel Category Discovery (SNCD), which requires the model to not only cluster the unlabeled images but also semantically recognize these images based on a set of their class names. The first challenge we confront pertains to effectively leveraging the class names of unlabeled images, given the inherent gap between the visual and linguistic domains. To address this issue, we incorporate a semantic-aware recognition mechanism. This is achieved by constructing dynamic class-wise visual prototypes as well as a semantic similarity matrix that enables the projection of visual features into the semantic space. The second challenge originates from the granularity disparity between the classification and clustering tasks. To deal with this, we develop a semantic-aware clustering process to facilitate the exchange of knowledge between the two tasks. Through extensive experiments, we demonstrate the mutual benefits of the recognition and clustering tasks, which can be jointly optimized. Experimental results on multiple datasets confirm the effectiveness of our proposed method. Our code is available at https://github.com/wang-weishuai/Semantic-guided-NCD.

Novel class discover(NCD) aims to identify new classes undefined during model training phase with the help of knowledge of known classes. Many methods have been proposed and notably boosted performance of NCD in natural images. However, there has been no work done in discovering new classes based on medical images and disease categories, which is crucial for understanding and diagnosing specific diseases. Moreover, most of the existing methods only utilize information from image modality and use labels as the only supervisory information. In this paper, we propose a multi-modal novel class discovery method based on paired images and text, inspired by the low classification accuracy of chest X-ray images and the relatively higher accuracy of the paired text. Specifically, we first pretrain the image encoder and text encoder with multi-modal contrastive learning on the entire dataset and then we generate pseudo-labels separately on the image branch and text branch. We utilize intra-modal consistency to assess the quality of pseudo-labels and adjust the weights of the pseudo-labels from both branches to generate the ultimate pseudo-labels for training. Experiments on eight subset splits of MIMIC-CXR-JPG dataset show that our method improves the clustering performance of unlabeled classes by about 10% on average compared to state-of-the-art methods. Code is available at: https://github.com/zzzzzzzzjy/MMNCD-main.

Open-vocabulary human-object interaction (HOI) detection, which is concerned with the problem of detecting novel HOIs guided by natural language, is crucial for understanding human-centric scenes. However, prior zero-shot HOI detectors often employ the same levels of feature maps to model HOIs with varying distances, leading to suboptimal performance in scenes containing human-object pairs with a wide range of distances. In addition, these detectors primarily rely on category names and over-look the rich contextual information that language can provide, which is essential for capturing open vocabulary concepts that are typically rare and not well-represented by category names alone. In this paper, we introduce a novel end-to-end open vocabulary HOI detection framework with conditional multi-level decoding and fine-grained semantic enhancement (CMD-SE), harnessing the potential of Visual-Language Models (VLMs). Specifically, we propose to model human-object pairs with different distances with different levels of feature maps by incorporating a soft constraint during the bipartite matching process. Furthermore, by leveraging large language models (LLMs) such as GPT models, we exploit their extensive world knowledge to generate descriptions of human body part states for various interactions. Then we integrate the generalizable and fine- grained semantics of human body parts to improve interaction recognition. Experimental results on two datasets, SWIG-HOI and HICO-DET, demonstrate that our proposed method achieves state-of-the-art results in open vocabulary HOI detection. The code and models are available at https://github.com/ltttpku/CMD-SE-release.

Dynamic Scene Graph Generation (DSGG) focuses on identifying visual relationships within the spatial-temporal domain of videos. Conventional approaches often employ multi-stage pipelines, which typically consist of object detection, temporal association, and multi-relation classification. However, these methods exhibit inherent limitations due to the separation of multiple stages, and independent optimization of these sub-problems may yield sub-optimal solutions. To remedy these limitations, we propose a one-stage end-to-end framework, termed OED, which streamlines the DSGG pipeline. This framework reformulates the task as a set prediction problem and leverages pairwise features to represent each subject-object pair within the scene graph. Moreover, another challenge of DSGG is capturing temporal dependencies, we introduce a Progressively Refined Module (PRM) for aggregating temporal context without the constraints of additional trackers or handcrafted trajectories, enabling end-to-end optimization of the network. Extensive experiments conducted on the Action Genome benchmark demonstrate the effectiveness of our design. The code and models are available at https://github.com/guanw-pku/OED.

Accurately detecting active objects undergoing state changes is essential for comprehending human interactions and facilitating decision-making. The existing methods for active object detection (AOD) primarily rely on visual appearance of the objects within input, such as changes in size, shape and relationship with hands. However, these visual changes can be subtle, posing challenges, particularly in scenarios with multiple distracting no-change instances of the same category. We observe that the state changes are often the result of an interaction being performed upon the object, thus propose to use informed priors about object related plausible interactions (including semantics and visual appearance) to provide more reliable cues for AOD. Specifically, we propose a knowledge aggregation procedure to integrate the aforementioned informed priors into oracle queries within the teacher decoder, offering more object affordance commonsense to locate the active object. To streamline the inference process and reduce extra knowledge inputs, we propose a knowledge distillation approach that encourages the student decoder to mimic the detection capabilities of the teacher decoder using the oracle query by replicating its predictions and attention. Our proposed framework achieves state-of-the-art performance on four datasets, namely Ego4D, Epic-Kitchens, MECCANO, and 100DOH, which demonstrates the effectiveness of our approach in improving AOD. The code and models are available at https://github.com/idejie/KAD.git.

Recent text-to-image generative models have demonstrated remarkable abilities in generating realistic images. Despite their great success, these models struggle to generate high-fidelity images with prompts oriented toward human-object interaction (HOI). The difficulty in HOI generation arises from two aspects. Firstly, the complexity and diversity of human poses challenge plausible human generation. Furthermore, untrustworthy generation of interaction boundary regions may lead to deficiency in HOI semantics. To tackle the problems, we propose a Semantic-Aware HOI generation framework SA-HOI . It utilizes human pose quality and interaction boundary region information as guidance for denoising process, thereby encouraging refinement in these regions to produce more reasonable HOI images. Based on it, we establish an iterative inversion and image refinement pipeline to continually enhance generation quality. Further, we introduce a comprehensive benchmark for HOI generation, which comprises a dataset involving diverse and fine-grained HOI categories, along with multiple custom-tailored evaluation metrics for HOI generation. Experiments demonstrate that our method significantly improves generation quality under both HOI-specific and conventional image evaluation metrics. The code is available at https://github.com/XZPKU/SA-HOI.git

Text-driven 3D indoor scene generation aims to automatically generate and arrange the objects, which form a 3D scene that accurately captures the semantics detailed in the given text description. Recent works have shown the potential to generate 3D scenes guided by specific object categories and room layouts but lack a robust mechanism to maintain consistent spatial relationships in alignment with the provided text description during the 3D scene generation. Besides, the annotations of the object and relationships of the 3D scenes are usually time- and cost-consuming, which are not easily obtained for the model training. Thus, in this paper, we conduct a dataset and benchmark for assessing spatial relations in text-driven 3D scene generation, which contains a comprehensive collection of 3D scenes, including textual descriptions, annotating object spatial relations, and providing both template and free-form natural language descriptions. We also provide a pseudo description feature generation method to address the 3D scenes without language annotations. We design an aligned latent space for spatial relation in 3D scenes and text description, in which we can sample the features according to the spatial relation for the few-shot learning. We also propose new metrics to investigate the ability of the approach to generate correct spatial relationships among objects.

Visual grounding aims to localize the object referred to in an image based on a natural language query. Although progress has been made recently, accurately localizing target objects within multiple-instance distractions (multiple objects of the same category as the target) remains a significant challenge. Existing methods demonstrate a significant performance drop when there are multiple distractions in an image, indicating an insufficient understanding of the fine-grained semantics and spatial relationships between objects. In this paper, we propose a novel approach, the Relation and Semantic-sensitive Visual Grounding (ResVG) model, to address this issue. Firstly, we enhance the model's understanding of fine-grained semantics by injecting semantic prior information derived from text queries into the model. This is achieved by leveraging text-to-image generation models to produce images representing the semantic attributes of target objects described in queries. Secondly, we tackle the lack of training samples with multiple distractions by introducing a relation-sensitive data augmentation method. This method generates additional training data by synthesizing images containing multiple objects of the same category and pseudo queries based on their spatial relationships. The proposed ReSVG model significantly improves the model's ability to comprehend both object semantics and spatial relations, leading to enhanced performance in visual grounding tasks, particularly in scenarios with multiple-instance distractions. We conduct extensive experiments to validate the effectiveness of our methods on five datasets. Code is available at https://github.com/minghangz/ResVG.

Accurate video moment retrieval (VMR) requires universal visual-textual correlations that can handle unknown vocabulary and unseen scenes. However, the learned correlations are likely either biased when derived from a limited amount of moment-text data which is hard to scale up because of the prohibitive annotation cost (fully-supervised), or unreliable when only the video-text pairwise relationships are available without fine-grained temporal annotations (weakly-supervised). Recently, the vision-language models (VLM) demonstrate a new transfer learning paradigm to benefit different vision tasks through the universal visual-textual correlations derived from large-scale vision-language pairwise web data, which has also shown benefits to VMR by fine-tuning in the target domains.In this work, we propose a zero-shot method for adapting generalisable visual-textual priors from arbitrary VLM to facilitate moment-text alignment, without the need for accessing the VMR data. To this end, we devise a conditional feature refinement module to generate boundary-aware visual features conditioned on text queries to enable better moment boundary understanding. Additionally, we design a bottom-up proposal generation strategy that mitigates the impact of domain discrepancies and breaks down complex-query retrieval tasks into individual action retrievals, thereby maximizing the benefits of VLM. Extensive experiments conducted on three VMR benchmark datasets demonstrate the notable performance advantages of our zero-shot algorithm, especially in the novel-word and novel-location out-of-distribution setups.

Estimating and monitoring the sleep states at home using ubiquitous infrared (IR) visual camera sensors is an essential healthcare problem. Currently, the common challenge of using IoT sensors to predict sleep stages is the “semantic gap” between the IoT sensory signals and the medical signals, where fewer correlations between IoT sensory signals and the sleep stage labels are observed. To bridge this gap, we propose a novel systematic and methodological IoT design (IoT-V2E) to retrieve the most similar electroencephalogram signal representations in a database given an IR visual query for sleep-related analysis. Specifically, we make the following specific contributions: 1) we collect a cross-modal retrieval data set, including the IR sensory signals and the synchronized Polysomnography signals with sleep stage ground-truth annotations; 2) we propose a novel uncertainty-aware hashing retrieval method, presenting superior performances, sufficient interpretability, and high memory efficiency; 3) our method achieves the state-of-the-art sleep stage retrieval results and provides the uncertainty for each query in the inference; and 4) most importantly, our system is evaluated to be able to assist the physicians not only in diagnosing sleep-related diseases but also finding the subjects with the most similar sleep patterns. Our project is available athttps://github.com/SPIresearch/IoT-V2E.

Multi-Source-Free Unsupervised Domain Adaptation (MSFUDA) requires aggregating knowledge from multiple source models and adapting it to the target domain. Two challenges remain: 1) suboptimal coarse-grained (domain-level) aggregation of multiple source models, and 2) risky semantics propagation based on local structures. In this article, we propose an evidential learning method for MSFUDA, where we formulate two uncertainties, i.e. Evidential Prediction Uncertainty (EPU) and Evidential Adjacency-Consistent Uncertainty (EAU), respectively for addressing the two challenges. The former, EPU, captures the uncertainty of a sample fitted to a source model, which can suggest the preferences of target samples for different source models. Based on this, we develop an EPU-Based Multi-Source Aggregation module to achieve fine-grained, instance-level source knowledge aggregation. The latter, EAU, provides a robust measure of consistency among adjacent samples in the target domain. Utilizing this, we develop an EAU-Guided Local Structure Mining module to ensure the trustworthy propagation of semantics. The two modules are integrated into the Evidential Aggregation and Adaptation Framework (EAAF), and we demonstrated that this framework achieves state-of-the-art performances on three MSFUDA benchmarks.

The objective of text-driven 3D indoor scene generation is to automatically generate and arrange the objects to form a 3D scene that accurately captures the semantics detailed in the given text description. Existing approaches are mainly guided by specific object categories and room layout to generate and position objects like furniture within 3D indoor scenes. However, few methods harness the potential of the text description to precisely control bothspatial relationshipsandobject combinations. Consequently, these methods lack a robust mechanism for determining accurate object attributes necessary to craft a plausible 3D scene that maintains consistent spatial relationships in alignment with the provided text description. To tackle these issues, we propose the Memory-Augmented Auto-regressive Network (MAAN), which is a text-driven method for synthesizing 3D indoor scenes with controllable spatial relationships and object compositions. Firstly, we propose a memory-augmented network to help the model decide the attributes of the objects, such as 3D coordinates, rotation and size, which improves the consistency of the object spatial relations with text descriptions. Our approach constructs a memory context to select relevant objects within the scene, which provides spatial information that aids in generating the new object with the correct attributes. Secondly, we develop a prior attribute prediction network to learn how to generate a complete scene with suitable and reasonable object compositions. This prior attribute prediction network adopts a pre-training strategy to extract composition priors from existing scenes, which enables the organization of multiple objects to form a reasonable scene and keeps the object relations according to the text descriptions. We conduct experiments on three different room types (bedroom, living room, and dining room) on the 3D-FRONT dataset. The results of these experiments underscore the accuracy of our method in governing spatial relationships among objects, showcasing its superior flexibility compared to existing techniques.

In this paper, we address the problem of video temporal sentence localization, which aims to localize a target moment from videos according to a given language query. We observe that existing models suffer from a sheer performance drop when dealing with simple phrases contained in the sentence. It reveals the limitation that existing models only capture the annotation bias of the datasets but lack sufficient understanding of the semantic phrases in the query. To address this problem, we propose a phrase-level Temporal Relationship Mining (TRM) framework employing the temporal relationship relevant to the phrase and the whole sentence to have a better understanding of each semantic entity in the sentence. Specifically, we use phrase-level predictions to refine the sentence-level prediction, and use Multiple Instance Learning to improve the quality of phrase-level predictions. We also exploit the consistency and exclusiveness constraints of phrase-level and sentence-level predictions to regularize the training process, thus alleviating the ambiguity of each phrase prediction. The proposed approach sheds light on how machines can understand detailed phrases in a sentence and their compositions in their generality rather than learning the annotation biases. Experiments on the ActivityNet Captions and Charades-STA datasets show the effectiveness of our method on both phrase and sentence temporal localization and enable better model interpretability and generalization when dealing with unseen compositions of seen concepts. Code can be found at https://github.com/minghangz/TRM.

Video sentence localization aims to locate moments in an unstructured video according to a given natural language query.A main challenge is the expensive annotation costs and the annotation bias.In this work, we study video sentence localization in a zero-shot setting, which learns with only video data without any annotation.Existing zero-shot pipelines usually generate event proposals and then generate a pseudo query for each event proposal.However, their event proposals are obtained via visual feature clustering, which is query-independent and inaccurate; and the pseudo-queries are short or less interpretable.Moreover, existing approaches ignores the risk of pseudo-label noise when leveraging them in training.To address the above problems, we propose a Structurebased Pseudo Label generation (SPL), which first generate free-form interpretable pseudo queries before constructing query-dependent event proposals by modeling the event temporal structure.To mitigate the effect of pseudolabel noise, we propose a noise-resistant iterative method that repeatedly re-weight the training sample based on noise estimation to train a grounding model and correct pseudo labels.Experiments on the ActivityNet Captions and Charades-STA datasets demonstrate the advantages of our approach.Code can be found at https://github.com/minghangz/SPL.

The correlation between the vision and text is essential for video moment retrieval (VMR), however, existing methods heavily rely on separate pre-training feature extractors for visual and textual understanding. Without sufficient temporal boundary annotations, it is non-trivial to learn universal video-text alignments. In this work, we explore multi-modal correlations derived from large-scale image-text data to facilitate generalisable VMR. To address the limitations of image-text pre-training models on capturing the video changes, we propose a generic method, referred to as Visual-Dynamic Injection (VDI), to empower the model's understanding of video moments. Whilst existing VMR methods are focusing on building temporalaware video features, being aware of the text descriptions about the temporal changes is also critical but originally overlooked in pre-training by matching static images with sentences. Therefore, we extract visual context and spatial dynamic information from video frames and explicitly enforce their alignments with the phrases describing video changes (e.g. verb). By doing so, the potentially relevant visual and motion patterns in videos are encoded in the corresponding text embeddings (injected) so to enable more accurate video-text alignments. We conduct extensive experiments on two VMR benchmark datasets (Charades-STA and ActivityNet-Captions) and achieve state-of-the-art performances. Especially, VDI yields notable advantages when being tested on the out-of-distribution splits where the testing samples involve novel scenes and vocabulary.

Tutorial videos play an increasingly important role in professional development and self-directed education. For users to realise the full benefits of this medium, tutorial videos must be efficiently searchable. In this work, we focus on the task of moment detection, in which the goal is to localise the temporal window where a given event occurs within a given tutorial video. Prior work on moment detection has focused primarily on short videos (typically on videos shorter than three minutes). However, many tutorial videos are substantially longer (stretching to hours in duration), presenting significant challenges for existing moment detection approaches.To study this problem, we propose the first dataset of untrimmed, long-form tutorial videos for the task of Moment Detection called the Behance Moment Detection (BMD) dataset. BMD videos have an average duration of over one hour and are characterised by slowly evolving visual content and wide-ranging dialogue. To meet the unique challenges of this dataset, we propose a new framework, LongMoment-Detr, and demonstrate that it outperforms strong baselines. Additionally, we introduce a variation of the dataset that contains YouTube Chapter annotations and show that the features obtained by our framework can be successfully used to boost the performance on the task of chapter detection. Code and data can be found at https://github.com/ioanacroi/longmoment-detr.

Human Object Interaction (HOI) detection aims to localize and infer the relationships between a human and an object. Arguably, training supervised models for this task from scratch presents challenges due to the performance drop over rare classes and the high computational cost and time required to handle long-tailed distributions of HOIs in complex HOI scenes in realistic settings. This observation motivates us to design an HOI detector that can be trained even with long-tailed labeled data and can leverage existing knowledge from pre-trained models. Inspired by the powerful generalization ability of the large Vision-Language Models (VLM) on classification and retrieval tasks, we propose an efficient Adaptive HOI Detector with Concept-guided Memory (ADA-CM). ADA-CM has two operating modes. The first mode makes it tunable without learning new parameters in a training-free paradigm. Its second mode incorporates an instance-aware adapter mechanism that can further efficiently boost performance if updating a lightweight set of parameters can be afforded. Our proposed method achieves competitive results with state-of-the-art on the HICO-DET and V-COCO datasets with much less training time. Code can be found at https://github.com/ltttpku/ADA-CM.

Visual grounding aims at localizing the target object in image which is most related to the given free-form natural language query. As labeling the position of target object is labor-intensive, the weakly supervised methods, where only image-sentence annotations are required during model training have recently received increasing attention. Most of the existing weakly-supervised methods first generate region proposals via pre-trained object detectors and then employ either cross-modal similarity score or reconstruction loss as the criteria to select proposal from them. However, due to the cross-modal heterogeneous gap, these method often suffer from high confidence spurious association and model prone to error propagation. In this paper, we propose Confidence-aware Pseudo-label Learning (CPL) to overcome the above limitations. Specifically, we first adopt both the uni-modal and cross-modal pre-trained models and propose conditional prompt engineering to automatically generate multiple ‘descriptive, realistic and diverse’ pseudo language queries for each region proposal, and then establish reliable cross-modal association for model training based on the uni-modal similarity score (between pseudo and real text queries). Secondly, we propose a confidence-aware pseudo label verification module which reduces the amount of noise encountered in the training process and the risk of error propagation. Experiments on five widely used datasets validate the efficacy of our proposed components and demonstrate state-of-the-art performance. Code can be found at https://github.com/zjh31/CPL.git

Domain adaptive Object Detection (DAOD) leverages a labeled domain (source) to learn an object detector generalizing to a novel domain without annotation (target). Recent advances use a teacher-student framework, i.e., a student model is supervised by the pseudo labels from a teacher model. Though great success, they suffer from the limited number of pseudo boxes with incorrect predictions caused by the domain shift, misleading the student model to get sub-optimal results. To mitigate this problem, we propose Masked Retraining Teacher-student framework (MRT) which leverages masked autoencoder and selective retraining mechanism on detection transformer. Specifically, we present a customized design of masked autoencoder branch, masking the multi-scale feature maps of target images and reconstructing features by the encoder of the student model and an auxiliary decoder. This helps the student model capture target domain characteristics and become a more data-efficient learner to gain knowledge from the limited number of pseudo boxes. Furthermore, we adopt selective retraining mechanism, periodically re-initializing certain parts of the student parameters with masked autoencoder refined weights to allow the model to jump out of the local optimum biased to the incorrect pseudo labels. Experimental results on three DAOD benchmarks demonstrate the effectiveness of our method. Code can be found at https://github.com/JeremyZhao1998/MRT-release.

Source-free unsupervised domain adaptation (SFUDA) aims to learn a target domain model using unlabeled target data and the knowledge of a well-trained source domain model. Most previous SFUDA works focus on inferring semantics of target data based on the source knowledge. Without measuring the transferability of the source knowledge, these methods insufficiently exploit the source knowledge, and fail to identify the reliability of the inferred target semantics. However, existing transferability measurements require either source data or target labels, which are infeasible in SFUDA. To this end, firstly, we propose a novel Uncertainty-induced Transferability Representation (UTR), which leverages uncertainty as the tool to analyse the channel-wise transferability of the source encoder in the absence of the source data and target labels. The domain-level UTR unravels how transferable the encoder channels are to the target domain and the instance-level UTR characterizes the reliability of the inferred target semantics. Secondly, based on the UTR, we propose a novel Calibrated Adaption Framework (CAF) for SFUDA, including i) the source knowledge calibration module that guides the target model to learn the transferable source knowledge and discard the non-transferable one, and ii) the target semantics calibration module that calibrates the unreliable semantics. With the help of the calibrated source knowledge and the target semantics, the model adapts to the target domain safely and ultimately better. We verified the effectiveness of our method using experimental results and demonstrated that the proposed method achieves state-of-the-art performances on the three SFUDA benchmarks. Code is available at https://github.com/SPIresearch/UTR.

Video moment localization aims at localizing the video segments which are most related to the given free-form natural language query. The weakly supervised setting, where only video level description is available during training, is getting more and more attention due to its lower annotation cost. Prior weakly supervised methods mainly use sliding windows to generate temporal proposals, which are independent of video content and low quality, and train the model to distinguish matched video-query pairs and unmatched ones collected from different videos, while neglecting what the model needs is to distinguish the unaligned segments within the video. In this work, we propose a novel weakly supervised solution by introducing Contrastive Negative sample Mining (CNM). Specifically, we use a learnable Gaussian mask to generate positive samples, highlighting the video frames most related to the query, and consider other frames of the video and the whole video as easy and hard negative samples respectively. We then train our network with the Intra-Video Contrastive loss to make our positive and negative samples more discriminative. Our method has two advantages: (1) Our proposal generation process with a learnable Gaussian mask is more efficient and makes our positive sample higher quality. (2) The more difficult intra-video negative samples enable our model to distinguish highly confusing scenes. Experiments on two datasets show the effectiveness of our method. Code can be found at https://github.com/minghangz/cnm.

Profiting from large-scale training datasets, advances in neural architecture design and efficient inference, joint embeddings have become the dominant approach for tackling cross-modal retrieval. In this work we first show that, despite their effectiveness, state-of-the-art joint embeddings suffer significantly from the longstanding “hubness problem” in which a small number of gallery embeddings form the nearest neighbours of many queries. Drawing inspiration from the NLP literature, we formulate a simple but effective framework called Querybank Normalisation (QB-NORM) that re-normalises query similarities to account for hubs in the embedding space. QB-NORM improves retrieval performance without requiring retraining. Differently from prior work, we show that QB-NORM works effectively without concurrent access to any test set queries. Within the QB-NORM framework, we also propose a novel similarity normalisation method, the Dynamic Inverted Softmax, that is significantly more robust than existing approaches. We showcase QB-NORM across a range of cross modal retrieval models and benchmarks where it consistently enhances strong baselines beyond the state of the art. Code is available at https://vladbogo.github.io/QB-Norm/.

Temporal sentence grounding aims to detect the most salient moment corresponding to the natural language query from untrimmed videos. As labeling the temporal boundaries is labor-intensive and subjective, the weakly- supervised methods have recently received increasing attention. Most of the existing weakly-supervised methods gen-erate the proposals by sliding windows, which are content- independent and of low quality. Moreover, they train their model to distinguish positive visual-language pairs from negative ones randomly collected from other videos, ignoring the highly confusing video segments within the same video. In this paper, we propose Contrastive Proposal Learning(CPL) to overcome the above limitations. Specifi-cally, we use multiple learnable Gaussian functions to gen-erate both positive and negative proposals within the same video that can characterize the multiple events in a long video. Then, we propose a controllable easy to hard neg-ative proposal mining strategy to collect negative samples within the same video, which can ease the model opti-mization and enables CPL to distinguish highly confusing scenes. The experiments show that our method achieves state-of-the-art performance on Charades-STA and Activi-tyNet Captions datasets. The code and models are available at https://github.com/minghangz/cpl.

Video temporal localization aims to locate a period that semantically matches a natural language query in a given untrimmed video. We empirically observe that although existing approaches gain steady progress on sentence localization, the performance of phrase localization is far from satisfactory. In principle, the phrase should be easier to localize as fewer combinations of visual concepts need to be considered; such incapability indicates that the existing models only capture the sentence annotation bias in the benchmark but lack sufficient understanding of the intrinsic relationship between simple visual and language concepts, thus the model generalization and interpretability is questioned. This paper proposes a unified framework that can deal with both sentence and phrase-level localization, namely Phrase Level Prediction Net (PLPNet). Specifically, based on the hypothesis that similar phrases tend to focus on similar video cues, while dissimilar ones should not, we build a contrastive mechanism to restrain phrase-level localization without fine-grained phrase boundary annotation required in training. Moreover, considering the sentence's flexibility and wide discrepancy among phrases, we propose a clustering-based batch sampler to ensure that contrastive learning can be conducted efficiently. Extensive experiments demonstrate that our method surpasses state-of-the-art methods of phrase-level temporal localization while maintaining high performance in sentence localization and boosting the model's interpretability and generalization capability. Our code is available at https://github.com/sizhelee/PLPNet.

Existing domain adaptation methods assume that domain discrepancies are caused by a few discrete attributes and variations, e.g., art, real, painting, quickdraw, etc. We argue that this is not realistic as it is implausible to define the real-world datasets using a few discrete attributes. Therefore, we propose to investigate a new problem namely the Continuous Domain Adaptation (CDA) through the lens where infinite domains are formed by continuously varying attributes. Leveraging knowledge of two labeled source domains and several observed unlabeled target domains data, the objective of CDA is to learn a generalized model for whole data distribution with the continuous attribute. Besides the contributions of formulating a new problem, we also propose a novel approach as a strong CDA baseline. To be specific, firstly we propose a novel alternating training strategy to reduce discrepancies among multiple domains meanwhile generalize to unseen target domains. Secondly, we propose a continuity constraint when estimating the cross-domain divergence measurement. Finally, to decouple the discrepancy from the mini-batch size, we design a domain-specific queue to maintain the global view of the source domain that further boosts the adaptation performances. Our method is proven to achieve the state-of-the-art in CDA problem using extensive experiments. The code is available at https://github.com/SPIresearch/CDA.

When can we expect a text-video retrieval system to work effectively on datasets that differ from its training domain? In this work, we investigate this question through the lens of unsupervised domain adaptation in which the objective is to match natural language queries and video content in the presence of domain shift at query-time. Such systems have significant practical applications since they are capable generalising to new data sources without requiring corresponding text annotations. We make the following contributions: (1) We propose the UDAVR (Unsupervised Domain Adaptation for Video Retrieval) benchmark and employ it to study the performance of text-video retrieval in the presence of domain shift. (2) We propose Concept-Aware-Pseudo-Query (CAPQ), a method for learning discriminative and transferable features that bridge these cross-domain discrepancies to enable effective target domain retrieval using source domain supervision. (3) We show that CAPQ outperforms alternative domain adaptation strategies on UDAVR.

In this paper, we study the task of visual-text retrieval in the highly practical setting in which labelled visual data with paired text descriptions are available in one domain (the "source"), but only unlabelled visual data (without text descriptions) are available in the domain of interest (the "target"). We propose the Adaptive Cross-Modal Prototypes framework which seeks to enable target domain retrieval by learning cross-modal visual-text representations while minimising both uni-modal and cross-modal distribution shift across the source and target domains. Our approach is built upon two key ideas: first, we encode the inductive bias that the learned cross-modal representations should be compositional with respect to concepts in each modality—this is achieved through clustering pretrained uni-modal features across each domain and designing a careful regularisation scheme to preserve the resulting structure. Second, we employ mutual information maximisation between cross-modal representations in the source and target domains during learning—this provides a mechanism that preserves commonalities between the domains while discarding signal in each that cannot be inferred from the other. We showcase our approach for the task of cross-domain visual-text retrieval, outperforming existing approaches for both images and videos.

We introduce QuerYD, a new large-scale dataset for retrieval and event localisation in video. A unique feature of our dataset is the availability of two audio tracks for each video: the original audio, and a high-quality spoken description of the visual content. The dataset is based on YouDescribe [1], a volunteer project that assists visually-impaired people by attaching voiced narrations to existing YouTube videos. This ever-growing collection of videos contains highly detailed, temporally aligned audio and text annotations. The content descriptions are more relevant than dialogue, and more detailed than previous description attempts, which can be observed to contain many superficial or uninformative descriptions. To demonstrate the utility of the QuerYD dataset, we show that it can be used to train and benchmark strong models for retrieval and event localisation. Data, code and models are made publicly available, and we hope that QuerYD inspires further research on video understanding with written and spoken natural language.

In recent years, considerable progress on the task of text-video retrieval has been achieved by leveraging large-scale pretraining on visual and audio datasets to construct powerful video encoders. By contrast, despite the natural symmetry, the design of effective algorithms for exploiting large-scale language pretraining remains under-explored. In this work, we are the first to investigate the design of such algorithms and propose a novel generalized distillation method, TeachText, which leverages complementary cues from multiple text encoders to provide an enhanced supervisory signal to the retrieval model. Moreover, we extend our method to video side modalities and show that we can effectively reduce the number of used modalities at test time without compromising performance. Our approach advances the state of the art on several video retrieval benchmarks by a significant margin and adds no computational overhead at test time. Last but not least, we show an effective application of our method for eliminating noise from retrieval datasets. Code and data can be found at https://www.robots.ox.ac.uk/˜vgg/research/teachtext/.

Video activity localisation has recently attained increasing attention due to its practical values in automatically localising the most salient visual segments corresponding to their language descriptions (sentences) from untrimmed and unstructured videos. For supervised model training, a temporal annotation of both the start and end time index of each video segment for a sentence (a video moment) must be given. This is not only very expensive but also sensitive to ambiguity and subjective annotation bias, a much harder task than image labelling. In this work, we develop a more accurate weakly-supervised solution by introducing Cross-Sentence Relations Mining (CRM) in video moment proposal generation and matching when only a paragraph description of activities without per-sentence temporal annotation is available. Specifically, we explore two cross-sentence relational constraints: (1) Temporal ordering and (2) semantic consistency among sentences in a paragraph description of video activities. Existing weakly-supervised techniques only consider within-sentence video segment correlations in training without considering cross-sentence paragraph context. This can mislead due to ambiguous expressions of individual sentences with visually indiscriminate video moment proposals in isolation. Experiments on two publicly available activity localisation datasets show the advantages of our approach over the state-of-the-art weakly supervised methods, especially so when the video activity descriptions become more complex.

Unsupervised domain Adaptation (UDA) aims to learn and transfer generalized features from a labelled source domain to a target domain without any annotations. Existing methods only aligning high-level representation but without exploiting the complex multi-class structure and local spatial structure. This is problematic as 1) the model is prone to negative transfer when the features from different classes are misaligned; 2) missing the local spatial structure poses a major obstacle in performing the fine-grained feature alignment. In this paper, we integrate the valuable information conveyed in classifier prediction and local feature maps into global feature representation and then perform a single mini-max game to make it domain invariant. In this way, the domain-invariant feature not only describes the holistic representation of the original image but also preserves mode-structure and fine-grained spatial structural information. The feature integration is achieved by estimating and maximizing the mutual information (MI) among the global feature, local feature and classifier prediction simultaneously. As the MI is hard to measure directly in high-dimension spaces, we adopt a new objective function that implicitly maximizes the MI via an effective sampling strategy and a discriminator design. Our STructure-Aware Feature Fusion (STAFF) network achieves the state-of-the-art performances in various UDA datasets.

Scene recognition remains one of the most challenging problems in image understanding. With the help of fully connected layers (FCL) and rectified linear units (ReLu), deep networks can extract the moderately sparse and discriminative feature representation required for scene recognition. However, few methods consider exploiting a sparsity model for learning the feature representation in order to provide enhanced discriminative capability. In this paper, we replace the conventional FCL and ReLu with a new dictionary learning layer, that is composed of a finite number of recurrent units to simultaneously enhance the sparse representation and discriminative abilities of features via the determination of optimal dictionaries. In addition, with the help of the structure of the dictionary, we propose a new label discriminative regressor to boost the discrimination ability. We also propose new constraints to prevent overfitting by incorporating the advantage of the Mahalanobis and Euclidean distances to balance the recognition accuracy and generalization performance. Our proposed approach is evaluated using various scene datasets and shows superior performance to many state-of-the-art approaches.

Unsupervised Domain Adaptation (UDA) aims to transfer domain knowledge from existing well-defined tasks to new ones where labels are unavailable. In the real-world applications, as the domain (task) discrepancies are usually uncontrollable, it is significantly motivated to match the feature distributions even if the domain discrepancies are disparate. Additionally, as no label is available in the target domain, how to successfully adapt the classifier from the source to the target domain still remains an open question. In this paper, we propose the Re-weighted Adversarial Adaptation Network (RAAN) to reduce the feature distribution divergence and adapt the classifier when domain discrepancies are disparate. Specifically, to alleviate the need of common supports in matching the feature distribution, we choose to minimize optimal transport (OT) based Earth-Mover (EM) distance and reformulate it to a minimax objective function. Utilizing this, RAAN can be trained in an end-to-end and adversarial manner. To further adapt the classifier, we propose to match the label distribution and embed it into the adversarial training. Finally, after extensive evaluation of our method using UDA datasets of varying difficulty, RAAN achieved the state-of-the-art results and outperformed other methods by a large margin when the domain shifts are disparate.

We address the problem of image feature learning for the applications where multiple factors exist in the image generation process and only some factors are of our interest. We present a novel multi-task adversarial network based on an encoder-discriminator-generator architecture. The encoder extracts a disentangled feature representation for the factors of interest. The discriminators classify each of the factors as individual tasks. The encoder and the discriminators are trained cooperatively on factors of interest, but in an adversarial way on factors of distraction. The generator provides further regularization on the learned feature by reconstructing images with shared factors as the input image. We design a new optimization scheme to stabilize the adversarial optimization process when multiple distributions need to be aligned. The experiments on face recognition and font recognition tasks show that our method outperforms the state-of-the-art methods in terms of both recognizing the factors of interest and generalization to images with unseen variations.

The aim of single image super-resolution (SR) is to generate a high-resolution (HR) image from a low-resolution (LR) observable image. In this paper, we address this task by integrating sparse coding and dictionary learning schemes into an end-to-end deep architecture. More specifically, we propose a new non-linear dictionary learning layer composed of a finite number of recurrent units to solve the sparse codes and also to yield the relevant gradients to update the dictionary. In addition, we present a new deep network architecture using the proposed non-linear layers, where two separate parallel dictionaries are adopted to represent the LR and HR images respectively. The whole network is optimized by back propagation, constraining not only reconstruction errors between the restored and the ground truth HR images but also between the sparse codes of the LR and HR image pairs. Various datasets are used to evaluate the performance of the proposed approach and it is shown to outperform many state-of-the-art single image super-resolution algorithms.