Is this you? Claim this profile to correct it, add a bio and choose the work people see first.
Claim this profileWorks217 from public data
- Rethinking Semantic Segmentation from a Sequence-to-Sequence Perspective with Transformers3,852
This paper deploys a pure transformer to encode an image as a sequence of patches, termed SEgmentation TRansformer (SETR), and shows that SETR achieves new state of the art on ADE20K, Pascal Context, and competitive results on Cityscapes.
- Harmonious Attention Network for Person Re-identification1,463
A novel Harmonious Attention CNN (HA-CNN) model is formulated for joint learning of soft pixel attention and hard regional attention along with simultaneous optimisation of feature representations, dedicated to optimise person re-id in uncontrolled (misaligned) images.
- Multimodal Learning With Transformers: A Survey1,183
This paper presents a comprehensive survey of Transformer techniques oriented at multimodal data, and a summary of the common challenges and designs shared by the multi-modal Transformer models and applications.
- Person Re-identification by Video Ranking642
A novel model to automatically select the most discriminative video fragments from noisy image sequences of people where more reliable space-time features can be extracted, whilst simultaneously to learn a video ranking function for person re-id is presented.
- Transferable Joint Attribute-Identity Deep Learning for Unsupervised Person Re-identification637
This work introduces an Transferable Joint Attribute-Identity Deep Learning (TJ-AIDL) for simultaneously learning an attribute-semantic and identity-discriminative feature representation space transferrable to any new (unseen) target domain for re-id tasks without the need for collecting new labelled training data from the target domain.
- Distribution-Aware Coordinate Representation for Human Pose Estimation552
This work finds that the process of decoding the predicted heatmaps into the final joint coordinates in the original image space is surprisingly significant for the performance, and forms a novel Distribution-Aware coordinate Representation of Keypoints (DARK) method.
- Person Re-Identification by Deep Joint Learning of Multi-Loss Classification426
This work designs a novel CNN architecture for Jointly Learning Multi-Loss (JLML) of local and global discriminative feature optimisation subject concurrently to the same re-id labelled information.
- 422
- Imbalanced Deep Learning by Minority Class Incremental Rectification373
These experimental results demonstrate the performance advantages and model scalability of the proposed batch-wise incremental minority class rectification model over the existing state-of-the-art models for addressing the problem of imbalanced data learning.
- 339
- Person Re-Identification by Camera Correlation Aware Feature Augmentation338
This work forms a novel view-specific person re-identification framework from the feature augmentation point of view, called Camera coR relation Aware Feature augmenTation (CRAFT), and presents a domain-generic deep person appearance representation which is designed particularly to be towards view invariant for facilitating cross-view adaptation by CRAFT.
- Fast Human Pose Estimation299
This work investigates the under-studied but practically critical pose model efficiency problem, and presents a new Fast Pose Distillation (FPD) model learning strategy that trains a lightweight pose neural network architecture capable of executing rapidly with low computational cost.
- Incremental Few-Shot Object Detection269
OpeN-ended Centre nEt (ONCE), a detector designed for incrementally learning to detect novel class objects with few examples, is proposed by an elegant adaptation of the efficient CentreNet detector to the few-shot learning scenario, and meta-learning a class-wise code generator model for registering novel classes.
- 263
- Unsupervised Person Re-identification by Deep Learning Tracklet Association256
A Tracklet Association Unsupervised Deep Learning (TAUDL) framework characterised by jointly learning per-camera (within-camera) tracklet association (labelling) and cross-camera tracklet correlation by maximising the discovery of most likely tracklet relationships across camera views.
- Simpler is Better: Few-shot Semantic Segmentation with Classifier Weight Transformer243
This work proposes to simplify the meta-learning task by focusing solely on the simplest component – the classifier, whilst leaving the en-coder and decoder to pre-training, and introduces a Classifier Weight Transformer (CWT) designed to dynamically adapt the support-set trained classifier’s weights to each query image in an inductive way.
- Person Re-Identification by Discriminative Selection in Video Ranking231
This work presents a novel model to automatically select the most discriminative video fragments from noisy/incomplete image sequences of people from which reliable space-time and appearance features can be computed, whilst simultaneously learning a video ranking function for person ReID.
- Semi-Supervised and Unsupervised Deep Visual Learning: A Survey228
This survey reviews the recent advanced deep learning algorithms on semi- supervised learning and unsupervised learning for visual recognition from a unified perspective and proposes a unified taxonomy to offer a holistic understanding of the state-of-the-art in these areas.
- Class Rectification Hard Mining for Imbalanced Deep Learning227
This work develops an end-to-end deep learning framework capable of avoiding the dominant effect of majority classes by discovering sparsely sampled boundaries of minority classes by introducing a Class Rectification Loss (CRL) regularising algorithm.
- 221
- 194
- 193
- Unsupervised Tracklet Person Re-Identification188
This work forms an Unsupervised Tracklet Association Learning (UTAL) framework that is capable of incrementally discovering and exploiting the underlying re-id discriminative information from automatically generated person tracklet data end-to-end.
- Semi-Supervised Learning under Class Distribution Mismatch182
This work addresses under-studied and realistic SSL problem by a novel algorithm named Uncertainty-Aware Self-Distillation (UASD), which produces soft targets that avoid catastrophic error propagation, and empower learning effectively from unconstrained unlabelled data with out-of-distribution (OOD) samples.
- Stochastic Classifiers for Unsupervised Domain Adaptation179
This paper introduces a novel method called STochastic clAssifieRs (STAR) for addressing the problem of misaligned local regions between source and target domain, which finds that using more classifiers leads to better performance, but also introduces more model parameters, therefore risking overfitting.
- Person Search by Multi-Scale Matching179
This work proposes a Cross-Level Semantic Alignment (CLSA) deep learning approach capable of learning more discriminative identity feature representations in a unified end-to-end model that favourably eliminates the need for constructing a computationally expensive image pyramid and a complex multi-branch network architecture.
- Attribute Recognition by Joint Recurrent Learning of Context and Correlation162
The JRL model learns jointly pedestrian attribute correlations in a pedestrian image and in particular their sequential ordering dependencies (latent high-order correlation) in an end-to-end encoder/ decoder recurrent network.
- Low-Resolution Face Recognition157
There is a significant gap between the reported FR performances on popular benchmarks and the results on TinyFace, and the advantages of the proposed CSRI over a variety of state-of-the-art FR and super-resolution deep models on solving this largely ignored FR scenario.
- 125
- Deep Low-Resolution Person Re-Identification124
This work develops a novel approach called Super-resolution and Identity joiNt learninG (SING) to simultaneously optimise image super- resolution and person re-id matching and introduces an adaptive fusion algorithm for accommodating multi-resolution LR images.
- Person re-identification by unsupervised video matching120
This work proposes an unsupervised approach to person re-identification based on typical surveillance image-sequences, and presents a new video representation particularly tailored for person ReID, built up on existing action space-time features.
- Human-in-the-Loop Person Re-identification115
A novel human-in-the-loop re-id model based on Human Verification Incremental Learning (HVIL) is formulated which does not require any pre-labelled training data to learn a model, therefore readily scalable to new camera pairs, and is formulated to combine a series of incrementally learned HVIL models into a single ensemble model to be used when human feedback becomes unavailable.
- 105
- Source-Free Domain Adaptation with Frozen Multimodal Foundation Model103
This work for the first time explores the potentials of off-the-shelf vision-language (ViL) multimodal models with rich whilst heterogeneous knowledge and proposes a novel Distilling multImodal Foundation mOdel (DIFO) approach, which significantly outperforms the state-of-the-art alternatives.
- Source-Free Domain Adaptation via Target Prediction Distribution Searching98
This paper proves that in case of sufficient small distribution shift, the domain transfer error could be well bounded, and proposes a novel Target Prediction Distribution Searching (TPDS) paradigm, which achieves new state-of-the-art performance.
- Deep Learning Logo Detection with Data Expansion by Synthesising Context98
This work designs a novel algorithm for generating Synthetic Context Logo (SCL) training images to increase model robustness against unknown background clutters, resulting in superior logo detection performance.
- Tracklet Self-Supervised Learning for Unsupervised Person Re-Identification97
A novel tracklet self-supervised learning (TSSL) method, which is capable of capitalising directly from abundant unlabelled tracklet data, to optimise a feature embedding space for both video and image unsupervised re-id.
- 92
- Multi-task Curriculum Transfer Deep Learning of Clothing Attributes92
A novel Multi-Task Curriculum Transfer (MTCT) deep learning method to explore multiple sources of different types of web annotations with multi-labelled fine-grained attributes for model transfer learning from well-controlled shop clothing images collected from web retailers to in-the-wild images from the street.
- FAME-ViL: Multi-Tasking Vision-Language Model for Heterogeneous Fashion Tasks86
This work proposes a novel FAshion-focused Multi-task Efficient learning method for Vision-and-Language tasks (FAME-ViL), which applies a single model for multiple heterogeneous fashion tasks, therefore being much more parameter-efficient.
- Generative Semantic Segmentation77
Generative Semantic Segmentation (GSS), a generative learning approach for semantic segmentation, is presented by replacing the conventional per-pixel discriminative learning with a latent prior learning process that model the variational posterior distribution of latent vari-ables given the segmentation mask.
- 74
- 73
- Semi-supervised Deep Learning with Memory70
This work proposes a novel Memory-Assessment Deep Neural Network capable of exploiting the memory of model learning to enable semi-supervised learning and introduces a memory mechanism into the network training process as an assimilation-accommodation interaction between the network and an external memory module.
- 68
- Vehicle Re-identification in Context67
Extensive evaluations show that the proposed multi-scale method outperforms the state-of-the-art vehicle re-id methods on three benchmark datasets: VehicleID, VeRi-776, and VRIC (Available at http://qmul-vric.io).
- Boundary-sensitive Pre-training for Temporal Localization in Videos66
This paper proposes to synthesize temporal boundaries in existing video action classification datasets and introduces a novel boundary-sensitive pretext (BSP) task, which is superior and complementary to the existing action classification-based pre-training counterpart and achieves new state-of-the-art performance on several temporal localization tasks.
- Deep Reinforcement Learning Attention Selection For Person Re-Identification66
This work develops a joint learning deep model that optimises person re-id attention selection within any auto-detected person bounding boxes by reinforcement learning of background clutter minimisation subject to re-ID label pairwise constraints.
- Proposal-Free Temporal Action Detection via Global Segmentation Mask Learning65
The TAGS model differs significantly from the conventional proposal-based methods by focusing on global temporal representation learning to directly detect local start and end points of action instances without proposals, and needs a much simpler model architecture with lower computational cost.
- Few-shot Action Recognition with Prototype-centered Attentive Learning64
A prototype-centered contrastive learning loss is introduced to complement the conventional query-centered learning objective, in order to make full use of the limited training samples in each episode of few-shot action recognition.
- 64
- 63
- 60
- 60
- 59
- 59
- 55
- 54
- 49
- 47
- Low-resolution human pose estimation47
A novel Confidence-Aware Learning (CAL) method is proposed which selectively weighs the learning of heatmap and offset with respect to ground-truth and most confident prediction, whilst capturing the statistical importance of model output in mini-batch learning manner.
- Few-shot website fingerprinting attack47
A model-agnostic, efficient, and Harmonious Data Augmentation (HDA) method that can improve deep WF attacking methods significantly and effectively addressing the intrinsic data scarcity problem is introduced.
- Intra-Camera Supervised Person Re-Identification46
MATE is designed for self-discovering the cross-camera identity correspondence in a per-camera multi-task inference framework and significantly outperforming unsupervised learning models and closely approaching conventional fully supervised learning competitors.
- 45
- 44
- 44
- Single-Label Multi-Class Image Classification by Deep Logistic Regression42
This work develops two novel LR based objective functions that not only generalise the conventional LR but importantly turn out to be competitive alternatives to SR in single label classification.
- 40
- Face re-identification challenge: Are face recognition models good enough?39
A true large scale Surveillance Face Re-ID benchmark (SurvFace) is introduced, characterised by natively low-resolution, motion blur, uncontrolled poses, varying occlusion, poor illumination, and background clutters, where facial images are captured and detected under realistic surveillance scenarios.
- Intra-Camera Supervised Person Re-Identification: A New Benchmark38
The Multi-Task Multi-Label (MTML) deep learning method is specially designed for self-discovering the inter-camera identity correspondence and can also efficiently learn the discriminative re-id feature representations by fully using the available identity labels within each camera-view.
- DiffTAD: Temporal Action Detection with Proposal Denoising Diffusion37
A new formulation of temporal action detection with denoising diffusion with denoising diffusion, DiffTAD, which can yield action proposals accurately given an untrimmed long video, presents a generative modeling perspective, against previous discriminative learning manners.
- Scalable Person Re-Identification by Harmonious Attention36
This work forms a novel harmonious attention network (HAN) framework to jointly learn soft pixel attention and hard region attention alongside simultaneous deep feature representation learning, particularly enabling more discriminative re-id matching by efficient networks with more scalable model inference and feature matching.
- Self-supervised Video Representation Learning with Motion-Aware Masked Autoencoders35
Extensive experiments show that the MotionMAE outperforms significantly both supervised learning baseline and state-of-the-art MAE alternatives, under both domain-specific and domain-generic pretraining-then-finetuning settings.
- Towards Uncovering the Intrinsic Data Structures for Unsupervised Domain Adaptation Using Structurally Regularized Deep Clustering35
This work proposes a hybrid model of Structurally Regularized Deep Clustering, which integrates the regularized discriminative clustering of target data with a generative one, and term the method as H-SRDC, which outperforms all the existing methods under both the inductive and transductive settings.
- Person Re-identification in Identity Regression Space33
This work proposes to construct an identity regression space (IRS) based on embedding different training person identities (classes) and formulate re-id as a regression problem solved by identity regression in the IRS, characterised by a closed-form solution with high learning efficiency and an inherent incremental learning capability with human-in theloop.
- DeepInteraction++: Multi-Modality Interaction for Autonomous Driving31
This work designs DeepInteraction++, a multi-modal interaction framework characterized by a multi-modal representational interaction encoder and a multi-modal predictive interaction decoder that demonstrates the effectiveness of the proposed modality interaction strategy.
- Uncertainty-aware pseudo-label filtering for source-free unsupervised domain adaptation31
This work introduces a sample selection module named Adaptive Pseudo-label Selection (APS), which is responsible for filtering noisy pseudo labels, and incorporates Class-Aware Contrastive Learning (CACL) to mitigate the memorization of pseudo-label noise by learning robust pair-wise representation supervised by pseudo labels.
- 31
- 29
- Vision Transformers: From Semantic Segmentation to Dense Prediction29
This work explores the global context learning potentials of ViTs for dense visual prediction and forms a family of Hierarchical Local-Global (HLG) Transformers, characterized by local attention within windows and global-attention across windows in a pyramidal architecture.
- 29
- 28
- 27
- 27
- Softmax-Free Linear Transformers26
A family of SOftmax-Free Transformers (SOFT) is proposed, with a Gaussian kernel function adopted to replace the dot-product similarity, enabling a full self-attention matrix to be approximated under low-rank matrix decomposition.
- 24
- 24
- 4D Gaussian Splatting: Modeling Dynamic Scenes with Native 4D Primitives23
This paper reformulates the reconstruction of a time-varying 3D scene as approximating its underlying spatiotemporal 4D volume by optimizing a collection of native 4D primitives, i.e., 4D Gaussians, with explicit geometry and appearance modeling.
- 23
- Global Aggregation Then Local Distribution for Scene Parsing23
A novel local distribution module is designed which models the affinity map between global and local relationship for each pixel adaptively and can be modularized as an end-to-end trainable block and easily plugged into existing semantic segmentation networks, giving rise to the GALD networks.
- 22
- Few-Shot Temporal Action Localization with Query Adaptive Transformer22
A novel FS-TAL model is proposed which maximizes the knowledge transfer from training classes whilst enabling the model to be dynamically adapted to both the new class and each video of that class simultaneously.
- 22
- 21
- 21
- 21
- 19
- 18
- 18
- 18
- 18
- 18
- 18
- 17
- 17
- 16
- 16
- 16
- 15
- 13
- 13
- 12
- 12
- 11
- 11
- 11
- 11
- 10
- 10
- 10
- Post-Processing Temporal Action Detection10
This work introduces a novel model-agnostic post-processing method without model redesign and retraining, dubbed as Gaussian Approximated Post-processing (GAP), which enables lower temporal resolutions for more efficient inference, facilitating low-resource application.
- 10
- 9
- 9
- FrozenSeg: Harmonizing Frozen Foundation Models for Open-Vocabulary Segmentation9
A novel approach, FrozenSeg, designed to integrate spatial knowledge from a localization foundation model and semantic knowledge extracted from a ViL model, in a synergistic framework, which advances state-of-the-art results across various segmentation benchmarks.
- DreamCreature: Crafting Photorealistic Virtual Creatures from Imagination9
This work proposes a new method called DreamCreature, which identifies and extracts the underlying sub-concepts in an unsupervised manner, and adapts to generate novel concepts with faithful structures and photorealistic appearance by seamlessly and flexibly composing learned sub- Concepts.
- 9
- 8
- Multi-Modal Few-Shot Temporal Action Detection8
A novel MUlti-modality PromPt mETa-learning (MUPPET) method that bridges pretrained vision and language models whilst maximally reusing already learned capacity to tackle few-shot object detection problem and achieves the state-of-the-art performance on MS-COCO dataset.
- 8
- 8
- 7
- 7
- 6
- 6
- Open-Set Semi-Supervised Learning for 3D Point Cloud Understanding6
This work proposes to selectively utilize unlabeled data through sample weighting, so that only conducive unlabeling data would be prioritized in semi-supervised learning for 3D point cloud.
- 5
- 5
- 5
- 4
- ConceptHash: Interpretable Fine-Grained Hashing via Concept Discovery4
Existing fine-grained hashing methods typically lack code interpretability as they compute hash code bits holistically using both global and local features, so this approach allows us to develop hash codes that exhibit similarity within families of species while remaining distinct from species in other families.
- 4
- 4
- 4
- 4
- 4
- 4
- 3
- Unified Unsupervised Anomaly Detection via Matching Cost Filtering3
The UCF framework, a generic post-hoc refinement framework for refining anomaly cost volume of any UAD model, is presented, consistently achieving new state-of-the-art results in both unimodal (RGB) and multimodal (RGB--3D, RGB--Text) UAD scenarios.
- 3
- 3
- 3
- 3
- 3
- 3
- 3
- 3
- 3
- 3
- 2
- 2
- Unsupervised Hashing with Similarity Distribution Calibration2
A novel Similarity Distribution Calibration (SDC) method is introduced, which aligns the hash code similarity distribution towards a calibration distribution with sufficient spread across the entire similarity capacity/range, to alleviate the similarity collapse problem.
- Knowing What, Where and When to Look: Video Action modelling with Attention2
A novel What-Where-When (W3) video attention module is proposed, which is extremely efficient by factorising the high-dimensional video feature data into low-dimensional meaningful spaces (1D channel vector for ‘what’ and 2D spatial tensors for ‘where’), followed by a lightweight temporal attention reasoning.
- 2
- 2
- 1
- 1
- 1
- 1
- 1
- 1
- 1
- 1
- 1
- 1
- 1
- 1
- 1
- 1
- 1
- 1
- 1
- –
- –
- –
- –
- –
- –
- –
- –
- –
- –
- –
- –
- –
- –
- –
- –
- –
- –
- –
- –
- –
- –
- –
- –
- –
- –
- –
- –
- –
- –
- –
- –
- –
- –
- –
Publication data from OpenAlex, with missing venues and authors filled in from Crossref; citation counts are the higher of OpenAlex and Semantic Scholar, last synced 2026-10-13. One-sentence summaries under some papers are written by Semantic Scholar’s model. Citation counts may be lower than on Google Scholar, which indexes more sources.
Report an error
Wrong papers, two people merged into one, or a profile that should not be here? Tell us and we will fix or hide it.