Is this you? Claim this profile to correct it, add a bio and choose the work people see first.
Claim this profileAcademic lineage
Possible advisorsa guess from early papers, not confirmed
- Suggested from co-authorship
Is this you? Claim this profile to confirm or dismiss it.
Works90 from public data
- Deblurring by Realistic Blurring494
This paper proposes a new method which combines two GAN models, i.e., a learning-to-Blur GAN (BGAN) and learning- to-DeBlurGAN (DBGAN), in order to learn a better model for image deblurring by primarily learning how to blur images.
- Hierarchical Neural Architecture Search for Deep Stereo Matching438
This paper proposes the first end-to-end hierarchical NAS framework for deep stereo matching by incorporating task-specific human knowledge into the neural architecture search framework and optimizing the architectures of the entire pipeline jointly.
- MiniMax-M1: Scaling Test-Time Compute Efficiently with Lightning Attention236
Experiments on standard benchmarks show that the MiniMax-M1 models are comparable or superior to strong open-weight models such as the original DeepSeek-R1 and Qwen3-235B, with particular strengths in complex software engineering, tool utilization, and long-context tasks.
- Positive Sample Propagation along the Audio-Visual Event Line149
A new positive sample propagation (PSP) module is proposed to discover and exploit the closely related audio-visual pairs by evaluating the relationship within every possible pair by constructing an all-pair similarity map between each audio and visual segment.
- 131
- RGB-D Saliency Detection via Cascaded Mutual Information Minimization131
This paper introduces a novel multistage cascaded learning framework via mutual information minimization to explicitly model the multi-modal information between RGB image and depth data.
- 130
- Implicit Motion Handling for Video Camouflaged Object Detection111
A new video camouflaged object detection framework that can exploit both short-term dynamics and long-term temporal consistency to detect camouflaged objects from video frames is proposed and unifies motion estimation and object segmentation within a single optimization framework.
- Audio-Visual Segmentation with Semantics104
A new baseline method is proposed that uses a temporal pixel-wise audio-visual interaction module to inject audio semantics as guidance for the visual segmentation process, and a regularization loss is designed to encourage audio-visual mapping during training.
- ARVo: Learning All-Range Volumetric Correspondence for Video Deblurring92
Quantitative and qualitative experiments show that the proposed implicit method to learn spatial correspondence among blurry frames in the feature space performs favorably on both datasets against previous state-of-the-art methods, confirming the benefit of modeling all-range spatial correspondence for video deblurring.
- Open-World Stereo Video Matching with Deep RNN79
A novel Recurrent Neural Network (RNN) that takes a continuous stereo video as input, and directly predicts a depth-map at each frame without a pre-training process, and without the need of ground-truth depth-maps as supervision.
- Learning Audio-Visual Source Localization via False Negative Aware Contrastive Learning76
This work proposes a new learning strategy named False Negative Aware Contrastive (FNAC), which utilizes the intra-modal similarities to identify potentially similar samples and construct corresponding adjacency matrices to guide contrastive learning.
- Noise-Aware Unsupervised Deep Lidar-Stereo Fusion75
LidarStereoNet is presented, the first unsupervised Lidar-stereo fusion network, which can be trained in an end-to-end manner without the need of ground truth depth maps, and is proposed to incorporate the piecewise planar model into the network learning to further constrain depths to conform to the underlying 3D geometry.
- Unsupervised Deep Epipolar Flow for Stationary or Dynamic Scenes75
Experimental results on various benchmarking datasets show that the Deep Epipolar Flow method achieves competitive performance compared with supervised methods and outperforms state-of-the-art unsupervised deep-learning methods.
- Deep Two-View Structure-from-Motion Revisited69
This work proposes to revisit the problem of deep two-view SfM by leveraging the well-posedness of the classic pipeline by developing a scale-invariant depth estimation network that leverages epipolar geometry to reduce the search space, refine the dense correspondences, and estimate relative depth maps.
- Lightning Attention-2: A Free Lunch for Handling Unlimited Sequence Lengths in Large Language Models64
Lightning Attention-2 is presented, the first linear attention implementation that enables linear attention to realize its theoretical computational benefits and retains consistent training and inference speed regardless of input sequence length and is significantly faster than other attention mechanisms.
- Multimodal Variational Auto-encoder based Audio-Visual Segmentation64
An Explicit Conditional Multimodal Variational Auto-Encoder for audio-visual segmentation (AVS), aiming to segment sound sources in the video sequence, with a new state-of-the-art performance leap on the challenging MS3 subset for multiple sound source segmentation.
- 64
- Vicinity Vision Transformer58
Vicinity Attention, which is a type of linear attention that integrates 2D locality that achieves state-of-the-art image classification accuracy with 50% fewer parameters than previous approaches, is proposed.
- Contrastive Conditional Latent Diffusion for Audio-Visual Segmentation55
A contrastive conditional latent diffusion model for audio-visual segmentation (AVS) is proposed to thoroughly investigate the impact of audio, where the correlation between audio and the final segmentation map is modeled to guarantee the strong correlation between them.
- Toeplitz Neural Network for Sequence Modeling55
A lightweight sub-network called relative position encoder is proposed to generate relative position coefficients with a fixed budget of parameters, enabling the proposed Toeplitz neural network to deal with varying sequence lengths.
- TAVGBench: Benchmarking Text to Audible-Video Generation53
A baseline model for TAVG is presented called TAVDiffusion, which uses a two-stream latent diffusion model to provide a fundamental starting point for further research in this area and proposes an automatic annotation pipeline to ensure each audible video has detailed descriptions for both its audio and video contents.
- Advancing Weakly-Supervised Audio-Visual Video Parsing via Segment-Wise Pseudo Labeling49
A new pseudo label generation strategy that can explicitly assign labels to each video segment by utilizing prior knowledge learned from the open world is proposed and achieves state-of-the-art video parsing performance on all types of event parsing, i.e., audio event, visual event, and audio-visual event.
- Towards Open-Vocabulary Audio-Visual Event Localization46
This paper introduces the Open-Vocabulary Audio-Visual Event Localization (OV-AVEL) problem, which requires localizing audio-visual events and predicting explicit categories for both seen and unseen test data at inference, and investigates two baseline approaches, one training-free and one using a further fine-tuning paradigm.
- 44
- An Alternative to WSSS? An Empirical Study of the Segment Anything Model (SAM) on Weakly-Supervised Semantic Segmentation Problems38
This report adapts SAM as the pseudo-label generation pipeline given only the image-level class labels in Weakly-Supervised Semantic Segmentation (WSSS), and observes impressive results in most cases.
- Various Lengths, Constant Speed: Efficient Language Modeling with Lightning Attention37
Lightning Attention is presented, the first linear attention implementation that maintains a constant training speed for various sequence lengths under fixed memory consumption and TransNormerLLM (TNL) is introduced, a new architecture that is tailored to the authors' lightning attention.
- TransNormerLLM: A Faster and Better Large Language Model with Improved TransNormer35
A robust inference algorithm is developed that ensures numerical stability and consistent inference speed, regardless of the sequence length, showcasing superior efficiency during both training and inference stages.
- 33
- MUNet: Motion uncertainty-aware semi-supervised video object segmentation29
A motion uncertainty-aware framework (MUNet) for semi-supervised VOS is proposed and an implicit method to learn the spatial correspondences between neighboring frames, building upon a correlation cost volume is proposed.
- 29
- Fine-grained Audible Video Description28
An audio-visual-language transformer that extends existing video captioning model with an additional audio branch is proposed, and the masked language modeling and auto-regressive language modeling losses are combined to optimize the model so that it can produce paragraph-level descriptions.
- Identification of lncRNA MEG3 Binding Protein Using MS2-Tagged RNA Affinity Purification and Mass Spectrometry27
Findings indicate that the newly developed method can effectively enrich lncRNA binding proteins and provides a strong basis for studying MEG3 functions.
- CO2: Efficient Distributed Training with Full Communication-Computation Overlap25
This work proposes a new approach called CO2 that introduces local-updating and asynchronous communication to the distributed data-parallel training, thereby facilitating the full overlap of COmunication with COmputation and provides a mathematical proof of convergence, accompanied by the establishment of a stringent upper bound.
- Neural Architecture Search on Efficient Transformers and Beyond24
This paper argues that the handcrafted vanilla Transformer architectures for Softmax attention may not be suitable for efficient Transformers, and proposes a new framework to find optimal architectures for efficientTransformers with the neural architecture search (NAS) technique.
- Improving Audio-Visual Video Parsing with Pseudo Visual Labels23
A new pseudo label generation strategy that can explicitly assign labels to each video segment by utilizing prior knowledge learned from the open world is proposed and can generate high-quality segment-level pseudo labels with the help of a newly proposed loss and the label denoising strategy.
- Linearized Relative Positional Encoding20
This work puts together a variety of existing linear relative positional encoding approaches under a canonical form and further proposes a family of linearrelative positional encoding algorithms via unitary transformation, leading to a principled framework that can be used to develop new relative positional encode methods that preserve linear space-time complexity.
- Rethinking RL Scaling for Vision Language Models: A Transparent, From-Scratch Framework and Comprehensive Evaluation Scheme18
A transparent, from-scratch framework for RL in VLMs is introduced, offering a minimal yet functional four-step pipeline validated across multiple models and datasets, and a standardized evaluation scheme is proposed to assess training dynamics and reflective behaviors.
- Deep Non-Rigid Structure-From-Motion: A Sequence-to-Sequence Translation Perspective17
The Context Layer is proposed, which enables the deep learning framework to effectively impose overall constraints on sequences based on the structural characteristics of non-rigid sequences based on the structural characteristics of non-rigid sequences.
- All-pairs Consistency Learning for Weakly Supervised Semantic Segmentation17
A new transformer-based regularization to better localize objects for Weakly supervised semantic segmentation (WSSS) and introduces a novel class-wise localization method that leverages the gradients of the class token.
- 17
- Stereo Computation for a Single Mixture Image16
This work gives a novel deep-learning based solution to the original problem of stereo computation from a single mixture image, by jointly solving the two subtasks of image layer separation as well as stereo matching.
- Robust Multi-Body Feature Tracker: A Segmentation-Free Approach16
A segmentationfree solution to multi-body feature tracking that bypasses the motion assignment step and reduces to solving a series of subproblems with closed-form solutions is introduced.
- 13
- Transcribing Natural Languages for the Deaf via Neural Editing Programs13
This work designs a new neural agent that learns to synthesize and execute editing programs, conditioned on sentence contexts and partial editing results, that outperforms previous glossification models by a large margin.
- 3D Geometry-Aware Semantic Labeling of Outdoor Street Scenes13
This paper proposes a 3D semantic labeling method to label outdoor street scenes whenever a dense depth map is available, and demonstrates that directly performing 3D convolution in the framework of a residual connected 3D voxel top-down modulation network can lead to superior results.
- LASP-2: Rethinking Sequence Parallelism for Linear Attention and Its Hybrid12
This paper introduces LASP-2, a new SP method to enhance both communication and computation parallelism when training linear attention transformer models with very-long input sequences, and extends LASP-2 to LASP-2H by applying similar communication redesign to standard attention modules, offering an efficient SP solution for hybrid models that blend linear and standard attention layers.
- 12
- Deep robust image deblurring via blur distilling and information comparison in latent space12
A latent contrastive model, Blur Distilling and Information Reconstruction Networks (BDIRNet), to learn image prior and improve the robustness of deep deblurring and achieve improved and robust results compared to recent state-of-the-art methods.
- 12
- Deep-Learning-Based Segmentation of Keyhole in In-Situ X-ray Imaging of Laser Powder Bed Fusion11
A deep-learning-based computer vision tool that can automatically segment keyhole shapes out of X-ray images is presented and shows promising average accuracy and may be applied to automatically label keyhole shapes in large image sets.
- 11
- You Only Scan Once: Efficient Multi-dimension Sequential Modeling with LightNet10
This paper identifies the inefficiency caused by a multiplicative linear recurrence and proposes an efficient alternative additive linear recurrence to avoid the issue, as it can handle multi-dimensional data within a single scan.
- Linear Attention Sequence Parallelism9
This paper designs an efficient point-to-point ring-style communication mechanism to leverage the right-product kernel trick of linear attention, which sharply decreases the communication overhead, comparing with existing SP methods.
- Displacement-Invariant Cost Computation for Stereo Matching9
This work proposes a displacement-invariant cost computation module to compute the matching costs without needing a 4D feature volume, and proposes an entropy-based refinement strategy to refine the computed disparity map, which further improves the speed by avoiding the need to compute a second disparity map on the right image.
- 8
- 8
- 8
- Local enhanced toeplitz neural network for in-vehicle network intrusion detection7
A Locally Enhanced Toeplitz Neural Network that takes advantage of the periodic nature of in-vehicle network messages and outperforms competitors DCNN, LSTM, Transformer, and TNN in terms of Accuracy, Precision, Recall, FPR, F1, and MCC indicators.
- Joint Salient Object Detection and Camouflaged Object Detection via Uncertainty-aware Learning7
An uncertainty-aware learning pipeline is introduced to extensively explore the contradictory information of salient object detection and camouflaged object detection via data-level and task-wise contradiction modeling, and introduces a contrastive module with a joint-task contrastive learning framework to explicitly model the contradictory attributes of these two tasks.
- Linear Video Transformer with Feature Fixation7
This work proposes a feature fixation module to reweight the feature importance of the query and key before computing linear attention, which significantly improves the linear attention baseline and achieves state-of-the-art performance among linear video Transformers on three popular video classification benchmarks.
- Locality Matters: A Locality-Biased Linear Attention for Automatic Speech Recognition7
By introducing a locality-biased linear attention mechanism to the Conformer, the method achieves a lower word error rate with more than 22% inference speed and enjoys linear space-time computational complexity.
- 6
- 6
- Tri-Ergon: Fine-Grained Video-to-Audio Generation with Multi-Modal Conditions and LUFS Control5
Tri-Ergon is introduced, a diffusion-based V2A model that incorporates textual, auditory, and pixel-level visual prompts to enable detailed and semantically rich audio synthesis, and introduces Loudness Units relative to Full Scale (LUFS) embedding, which allows for precise manual control of the loudness changes over time for individual audio channels.
- Autoregressive Image Generation with Linear Complexity: A Spatial-Aware Decay Perspective5
This work proposes Linear Attention with Spatial-Aware Decay (LASAD), a novel attention mechanism that explicitly preserves genuine 2D spatial relationships within the flattened image sequences by computing position-dependent decay factors based on true 2D spatial location rather than 1D sequence positions.
- 5
- Image-based Geolocalization by Ground-to-2.5D Map Matching5
This paper presents a new approach to learning representative embeddings from multi-model data that achieves significantly higher localization accuracy and faster convergence than previous 2D map-based approaches.
- 5
- 3
- 3
- 3
- AMF: Adaptable Weighting Fusion with Multiple Fine-tuning for Image Classification3
The Adaptable Multi-tuning method is proposed, which adaptively determines each data sample’s fine- Tuning strategy, and outperforms the standard approach by 1.69 %, 2.79 % on the datasets FGVC-Aircraft, and Describ- able Texture.
- 3
- 3
- Geolocation on Cartographic Maps with Multi-Modal Fusion2
This paper proposes a new approach to learning representative embeddings from multi-modal data, and establishes a projection relationship between 2D and 2.5D space, which achieves significantly higher geolocation accuracy and faster convergence than previous 2D map-based approaches.
- 2
- 2
- Object-oriented Implementation of Chess Game in C++2
This paper implements the standard chess game using C++, a popular object-oriented programming language, developed and fully tested on Mac OS X system and allows two players to compete together.
- 1
- 1
- 1
- 1
- 1
- –
- –
- –
- –
- –
- –
Publication data from OpenAlex; citation counts are the higher of OpenAlex and Semantic Scholar, last synced 2026-10-14. 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.