Is this you? Claim this profile to correct it, add a bio and choose the work people see first.
Claim this profileAcademic lineage
Traced back 39 generations →Advisors
- Wang XinchaoFrom a thesis record ↗
Works54 from public data
- AdaAttN: Revisit Attention Mechanism in Arbitrary Neural Style Transfer456
A novel attention and normalization module, named Adaptive Attention Normalization (AdaAttN), to adaptively perform attentive normalization on per-point basis is proposed, which achieves state-of-the-art arbitrary image/video style transfer.
- OminiControl: Minimal and Universal Control for Diffusion Transformer386
OminiControl, a novel approach that rethinks how image conditions are integrated into Diffusion Transformer (DiT) architectures, demonstrates that effective image control can be achieved without architectural complexity, opening new possibilities for efficient and versatile image generation systems.
- Dataset Distillation: A Comprehensive Review213
A comprehensive review and summary of recent advances in dataset distillation and its application is given and an overall algorithmic framework followed by all existing DD methods is proposed.
- Dataset Distillation via Factorization190
A novel approach todataset factorization is introduced, termed HaBa, which is a plug-and-play strategy portable to any existing DD baseline and can yield significant improvement on downstream classification tasks compared with previous state of the arts.
- Deep Model Reassembly162
It is demonstrated that on ImageNet, the best reassemble model achieves 78.6% top-1 accuracy without fine-tuning, which could be further elevated to 83.2% with end-to-end training.
- Paint Transformer: Feed Forward Neural Painting with Stroke Prediction112
This paper formulate the task as a set prediction problem and proposes a novel Transformer-based framework, dubbed Paint Transformer, to predict the parameters of a stroke set with a feed forward network, which achieves better painting performance than previous ones with cheaper training and inference costs.
- SG-Former: Self-guided Transformer with Evolving Token Reallocation78
A novel model, termed as Self-guided Transformer (SG-Former), towards effective global self-attention with adaptive fine granularity, which assigns more tokens to the salient regions for achieving fine-grained attention, while allocating fewer tokens to the minor regions in exchange for efficiency and global receptive fields.
- 52
- LinFusion: 1 GPU, 1 Minute, 16K Image47
A generalized linear attention paradigm is introduced, which serves as a low-rank approximation of a wide spectrum of popular linear token mixers and achieves performance on par with or superior to the original SD after only modest training, while significantly reducing time and memory complexity.
- 47
- 39
- Distribution Shift Inversion for Out-of-Distribution Prediction34
This paper proposes a portable Distribution Shift Inversion (DSI) algorithm, in which, before being fed into the prediction model, the OoD testing samples are first linearly combined with additional Gaussian noise and then transferred back towards the training distribution using a diffusion model trained only on the source distribution.
- Master: Meta Style Transformer for Controllable Zero-Shot and Few-Shot Artistic Style Transfer29
A novel Transformer model termed as Master specifically for style transfer is devised so that it can not only work in the typical setting of arbitrary style transfer, but also adaptable to the few-shot setting, by only fine-tuning the Transformer encoder layer in the many-shot stage for one specific style.
- 29
- 29
- 27
- 26
- Any-to-Any Style Transfer: Making Picasso and Da Vinci Collaborate22
This work proposes a novel strategy termed Any-to-Any Style Transfer, which enables users to interactively select styles of regions in the style image and apply them to the prescribed content regions, and performs in a plug-and-play manner portable to any style transfer method and enhance the controllablity.
- Teddy: Efficient Large-Scale Dataset Distillation via Taylor-Approximated Matching17
This paper introduces Teddy, a Taylor-approximated dataset distillation framework designed to handle large-scale dataset and enhance efficiency, and proposes a memory-efficient approximation derived from Taylor expansion, which transforms the original form dependent on multi-step gradients to a first-order one.
- Partial Network Cloning17
An innovative learning scheme is introduced that allows us to identify simultaneously the component to be cloned from the source and the position to be inserted within the target network, so as to ensure the optimal performance.
- 15
- StyDeSty: Min-Max Stylization and Destylization for Single Domain Generalization11
This paper proposes a simple yet effective scheme, termed as StyDeSty, to explicitly account for the alignment of the source and pseudo domains in the process of data augmentation, enabling them to interact with each other in a self-consistent manner and further giving rise to a latent domain with strong generalization power.
- 11
- 9
- Enhanced social emotional optimisation algorithm with generalised opposition-based learning9
A novel social emotional optimisation algorithm, called GOSEOA, which performs the generalised opposition-based learning GOBL strategy with a certain probability during the evolution process, and can obtain promising performance on the majority of the test functions.
- Risk Minimization Against Transmission Failures of Federated Learning in Mobile Edge Networks8
A randomized algorithm is designed to choose suitable participants by using a series of delicately calculated probabilities, and it is proved that the result is concentrated on its optimum with high probability, showing that through delicate participant selection, the maximal error rate of model updates is decreased.
- 7
- 7
- Control and Realism: Best of Both Worlds in Layout-to-Image without Training6
A Langevin dynamics-based adaptive update scheme is introduced as a remedy that promotes in-domain updating while respecting layout constraints, outperforming the current state-of-the-art methods.
- Stable Video Style Transfer Based on Partial Convolution with Depth-Aware Supervision6
This paper focuses on the ghosting problem existing in most previous works and uses partial convolution-based strategy to utilize inter-frame context and correlation, together with additional depth loss as a constrain to the generated frames to suppress ghosting artifacts and preserve stability at the same time.
- Heavy Labels Out! Dataset Distillation with Label Space Lightening4
A novel label-lightening framework aimed at effective image-to-label projectors with which synthetic labels can be directly generated online from synthetic images is proposed, and an effective image optimization method is proposed to further mitigate the potential error between the original and distilled label generators.
- Understanding Dataset Distillation via Spectral Filtering3
UniDD is introduced, a spectral filtering framework that unifies diverse DD objectives and reveals that the essence of DD fundamentally lies in matching frequency-specific features.
- Top–Down Compression: Revisit Efficient Vision Token Projection for Visual Instruction Tuning2
This paper presents LLaVA-Meteor, a novel approach designed to achieve a favorable accuracy-efficiency trade-off, equipped with a new Top-Down Compression paradigm that strategically compresses visual tokens while preserving a substantial amount of critical visual information.
- ViFeEdit: A Video-Free Tuner of Your Video Diffusion Transformer2
A video-free tuning framework termed ViFeEdit for video diffusion transformers that enables visually faithful editing while maintaining temporal consistency with only minimal additional parameters, and delivers promising results of controllable video generation and editing with only minimal training on 2D image data.
- 2
- 2
- 1
- 1
- Graph-Based Locality-Sensitive Circuit Sketch Recognizer1
A novel sketch recognition algorithm that uses graph to model the input strokes and their relationships, and leverages cycles by local strokes to detect some circuit components is proposed, which outperforms previous state-of-the-art methods numerically.
- –
- –
- FreLay: Frequency-aware Energy Function for Training-free Layout-to-Image Generation–
This paper introduces FreLay, a novel training-free approach equipped with a frequency-aware energy function that consistently outperforms existing state-of-the-art training-free methods both qualitatively and quantitatively across multiple datasets.
- –
- CoDA: From Text-to-Image Diffusion Models to Training-Free Dataset Distillation–
Core Distribution Alignment (CoDA), a framework that enables effective DD using only an off-the-shelf text-to-image model, and achieves performance on par with or even superior to previous methods with such reliance across all benchmarks, including ImageNet-1K and its subsets.
- –
- –
- –
- [Application of Digital Galvanometer Scanner System for CO2 Fractional Laser Safety Improvement].–
An adaptive limit protection method, a coil open circuit fault diagnosis, a communication timeout protection based on two handshakes, and a galvanometer control timeout protection are proposed, respectively, based on a digital driver platform, to supplement the deficiencies in the original fault diagnosis and protection system.
- Kernel-based Informative Feature Extraction via Gradient Learning–
This paper derives an informative energy model to quantification of feature difference and moves the features in the same class closer and push away those belong to different classes according to the model and derivate its objective function.
- INFORMATIVE ENERGY METRIC FOR SIMILARITY MEASURE IN REPRODUCING KERNEL HILBERT SPACES–
Information energy metric is obtained by similarity computing for high-dimensional samples in a reproducing kernel Hilbert space (RKHS) and IEM is proposed for similarity measure of those subsets, which converts the non-metric distances into metric ones.
- A Novel Fusion Method for Dynamic Updating in Inverted Index Establishment–
A novel fusion method using dynamic updating for hash lists based index that can save index merging time as well as spare some memory space is proposed.
- The Technology of the Fractal's Encrpt and Inencrpt–
The Fractal picture is created with a method of the Newton, associate it with Encrpt and Decrypt, construct the Encrypt and the Decrypt to the nonlinear model and analyze the theories of the Fractal in the information security technique.
- –
- Approaching the Design of Information System of Financial Management Based on Excel Context–
The paper interpret mainly the basic contents of FMIS and enthrone EXCEL software as the most convenient and ideal design platform and tool to design and maintain FMIS directly for users.
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 2026-10-10. 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. You will be asked to sign in.