Is this you? Claim this profile to correct it, add a bio and choose the work people see first.
Claim this profileWorks126 from public data
- LLaMA-Adapter: Efficient Fine-tuning of Language Models with Zero-init Attention1,068
A zero-initialized attention mechanism with zero gating is proposed, which adaptively injects the new instructional cues into LLaMA, while effectively preserves its pre-trained knowledge on traditional vision and language tasks, demonstrating the superior generalization capacity of the approach.
- 789
- LLaMA-Adapter V2: Parameter-Efficient Visual Instruction Model774
This work augments LLaMA-Adapter by unlocking more learnable parameters and proposes an early fusion strategy to feed visual tokens only into the early LLM layers, contributing to better visual knowledge incorporation and achieves strong multi-modal reasoning with only a small-scale image-text and instruction dataset.
- 725
- Personalize Segment Anything Model with One Shot400
A training-free Personalization approach for SAM, termed as PerSAM, which effectively adapt SAM for private use without any training, and can enhance DreamBooth to personalize Stable Diffusion for text-to-image generation, which discards the background disturbance for better target appearance learning.
- PointCLIP V2: Prompting CLIP and GPT for Powerful 3D Open-world Learning293
This paper first collaborate CLIP and GPT to be a unified 3D open-world learner, named as Point-CLIP V2, which fully unleashes their potential for zero-shot 3D classification, segmentation, and detection, demonstrating the generalization ability for unified 3D open-world learning.
- OneLLM: One Framework to Align All Modalities with Language286
This paper presents OneLLM, an MLLM that aligns eight modalities to language using a unified framework through a unified multimodal encoder and a progressive multimodal alignment pipeline.
- Prompt, Generate, Then Cache: Cascade of Foundation Models Makes Strong Few-Shot Learners255
CaFo, a Cascade of Foundation models that incorporates diverse prior knowledge of various pretraining paradigms for better few-shot learning is proposed that can fully unleash the potential of different pre-training methods and unify them to perform state-of-the-art for few- shot classification.
- Learning 3D Representations from 2D Pre-Trained Models via Image-to-Point Masked Autoencoders217
This paper proposes an alternative to obtain superior 3D representations from 2D pre-trained models via Image-to-Point Masked Autoencoders, named as I2P-MAE, which leverages the well learned 2D knowledge to guide 3D masked autoencoding, which reconstructs the masked point tokens with an encoder-decoder architecture.
- MonoDETR: Depth-guided Transformer for Monocular 3D Object Detection216
This paper introduces the first DETR framework for Monocular DEtection with a depth-guided TRansformer, named MonoDETR, and modify the vanilla transformer to be depth-aware and guide the whole detection process by contextual depth cues.
- Instruct2Act: Mapping Multi-modality Instructions to Robotic Actions with Large Language Model177
This paper presents Instruct2Act, a framework that utilizes Large Language Models to map multi-modal instructions to sequential actions for robotic manipulation tasks, employing the LLM model to generate Python programs that constitute a comprehensive perception, planning, and action loop for robotic tasks.
- ConvMAE: Masked Convolution Meets Masked Autoencoders164
The ConvMAE framework demonstrates that multi-scale hybrid convolution-transformer can learn more discriminative representations via the mask auto-encoding scheme, and adopts the masked convolution to prevent information leakage in the convolution blocks.
- 164
- Lumina-T2X: Transforming Text into Any Modality, Resolution, and Duration via Flow-based Large Diffusion Transformers147
This technical report introduces the Lumina-T2X family - a series of Flow-based Large Diffusion Transformers (Flag-DiT) equipped with zero-initialized attention, as a unified framework designed to transform noise into images, videos, multi-view 3D objects, and audio clips conditioned on text instructions.
- Not All Features Matter: Enhancing Few-shot CLIP with Adaptive Prior Refinement143
APE, an Adaptive Prior rEfinement method for CLIP’s pre-trained knowledge, which achieves superior accuracy with high computational efficiency and introduces two model variants, a training-free APE and a training-required APE-T.
- 122
- The Argo: a high channel count recording system for neural recording in vivo122
The Argo system is the highest channel count in vivo neural recording system, supporting simultaneous recording from 65 536 channels, sampled at 32 kHz and 12-bit resolution, and helps pave the way for a future high data rate intracortical implant.
- 104
- 99
- You Only Need 90K Parameters to Adapt Light: A Light Weight Transformer for Image Enhancement and Exposure Correction79
This work proposes a lightweight fast Illumination Adaptive Transformer (IAT) which comprises two transformer-style branches: local estimation branch and global ISP branch which could also conduct both object detection and semantic segmentation under various light conditions.
- 43
- 42
- Key technologies of human–computer interaction for immersive somatosensory interactive games using VR technology41
A naive Bayes gesture recognition algorithm based on joint vector, which has a high recognition rate and recognition speed, can effectively adapt to continuous data flow, has good compatibility with virtual reality applications, and significantly increases the user's sense of immersion in virtual scenes.
- 37
- 32
- 25
- 24
- 23
- POS-BERT: Point Cloud One-Stage BERT Pre-Training23
POS-BERT is proposed, a one-stage BERT pre-training method for point clouds that achieves the state-of-the-art classification accuracy and has significantly improved many downstream tasks, such as fine-tuned classification, few-shot classification, part segmentation.
- The Argo: A 65,536 channel recording system for high density neural recording in vivo21
The Argo System, a massively parallel neural recording system based on platinum-iridium microwire electrode arrays bonded to a CMOS voltage amplifier array, validated by recording spiking activity from 791 neurons in rats and cortical surface Local Field Potential activity from over 30,000 channels in sheep.
- 21
- 18
- Milk fermented by combined starter cultures comprising three Lactobacillus strains exerts an alleviating effect on loperamide-induced constipation in BALB/c mice18
It is indicated that the combined starter culture fermented milk can effectively alleviate Lop-induced-constipation in BALB/c mice and exerted a regulatory effect on the gut microbiota of the mice by up-regulating the abundance of Lactobacillus and Bacteroides, and decreasing the abundanceof Helicobacter, Pseudomonas and Porphyromonas.
- Fundus Changes Evaluated by OCTA in Patients With Cerebral Small Vessel Disease and Their Correlations: A Cross-Sectional Study18
OCTA parameters were correlated with CSVD, indicating that OCTA is a potential method for CSVD screening, and the correlations between CSVD and fundus changes were investigated.
- 17
- 14
- 13
- Vision-Language Navigation with Embodied Intelligence: A Survey12
This survey systematically reviews the research progress of VLN and details the research direction of VLN with embodied intelligence, aiming to provide a practical reference for researchers.
- Cytokine and Chemokine Profile Changes in Patients with Neovascular Age-Related Macular Degeneration After Intravitreal Ranibizumab Injection for Choroidal Neovascularization12
Angiogenic, growth factors and inflammatory are involved in the formation of neovascularization of AMD patients and IVR did not cause significant differences in any growth factors or inflammatory cytokines in nAMD patients with the exception of VEGF.
- 11
- X-ray Photon Counting and Two-Color X-ray Imaging Using Indirect Detection11
The design and performance of a 1 cm2, 90 × 92-pixel image sensor made X-ray sensitive by the use of a scintillator, which has a charge packet counting circuit topology with two channels realizing a different charge packet size threshold and analog domain event counting is reported.
- 10
- 10
- 10
- A 40MHz-BW two-step open-loop VCO-based ADC with 42fJ/step FoM in 40nm CMOS10
A two-step open-loop VCO-based ADC with 1st-order noise shaping and intrinsic nonlinearity mitigation is presented and a robust performance, high bandwidth and high efficiency is achieved.
- Coordinated Multi-Robot Navigation with Formation Adaptation8
This work develops AFOR under the theoretical framework of hierarchical learning and integrates a spring-damper model with hierarchical learning to enable both team coordination and individual robot control and demonstrates that AFOR enables smooth navigation with formation adaptation in complex scenarios and outperforms previous methods.
- Collaboration of Pre-trained Models Makes Better Few-shot Learner8
This paper proposes CoMo, a Collaboration of pre-trained Models that incorporates diverse prior knowledge from various pre-training paradigms for better few-shot learning and introduces a learnable Multi-Knowledge Adapter (MK-Adapter) to adaptively blend the predictions from CLIP and DINO.
- 7
- Learning Decoupling Features Through Orthogonality Regularization6
A two-branch deep network (KWS branch and SV branch) with the same network structure is developed and a novel decoupling feature learning method is proposed to push up the performance of KWS and SV simultaneously where speaker-invariant keyword representations and keyword- Invariant speaker representations are expected respectively.
- 6
- 6
- 5
- 5
- 5
- Application Research on Multimedia Information Technology in the Universities Physical Teaching5
Analysis of the application of the multimedia information technology education in colleges and universities physical teaching mode, advantages and existing problems of teaching, makes judgments for the role of network teaching in the universitiesphysical teaching and provides theoretical and practical significance.
- IMAGINE-E: Image Generation Intelligence Evaluation of State-of-the-art Text-to-Image Models4
This comprehensive assessment highlights each model's strengths and limitations, particularly the outstanding performance of FLUX.1 and Ideogram2.0 in structured and specific domain tasks, underscoring the expanding applications and potential of T2I models as foundational AI tools.
- Subteaming and Adaptive Formation Control for Coordinated Multi-Robot Navigation4
Experimental results show that STAF enables the novel capability for subteaming and adaptive formation control, and achieves promising performance in coordinated multi-robot navigation through challenging scenarios.
- 4
- Bilateral Femtosecond Laser-Assisted Clear Lens Extraction with Trifocal Intraocular Lens Implantation for Presbyopia Correction: A Short-Term Observational Study in Chinese Adults4
Satisfactory visual outcomes and safe surgical procedures were observed in this study, which demonstrate that bilateral femtosecond laser-assisted clear lens extraction with PanOptix IOL implantation could be a feasible approach for presbyopia correction in working-age Chinese patients.
- Exploring representation learning for small-footprint keyword spotting4
Experiments on speech commands dataset show that the self-training WVC module and theSelf-supervised LGCSiam module significantly improve accuracy, especially in the case of training on a small labeled dataset.
- Knockout of Shelterin subunit genes in zebrafish results in distinct outcomes4
This study generated knockout zebrafish of the Shelterin subunit genes acd, pot1, tinf2, terf1 and pinx1 using the CRISPR/Cas9 technology and provided novel animal models for further exploring the relationship between telomeres and aging as well as the pathogenesis of human diseases associated with telomere deficiency.
- 4
- 4
- 4
- ISCLEs: Importance Sampled Circuit Learning Ensembles for Trustworthy Analog Circuit Topology Synthesis4
Importance Sampled Circuit Learning Ensembles (ISCLEs) is a novel analog circuit topology synthesis method that returns designer-trustworthy circuits yet can apply to a broad range of circuit design problems including novel functionality.
- A Multiple Approximate String Matching Algorithm of Network Information Audit System4
This paper shows a simple, efficient, and practical algorithm for locating all occurrences of a finite number of keywords in a char/Chinese character string allowing k chars inserting errors.
- Filter Pruning Via Filters Similarity in Consecutive Layers3
This paper intuitively proposes a novel pruning method by explicitly leveraging the Filters Similarity in Consecutive Layers (FSCL), which compresses models by pruning filters whose corresponding features are more worthless in the model.
- 3
- 3
- 2
- 2
- 2
- Cryogenic and radiation-hard asic for interfacing large format NIR/SWIR detector arrays2
This “ASIC” prototype is realized in 0.18um CMOS technology and is designed for radiation hardness, which will significantly reduce the number of wire goes in and out the cryogenic chamber.
- 2
- Model Architecture Analysis and Implementation of TENET for Cell–Cell Interaction Network Reconstruction Using Spatial Transcriptomics Data1
A novel method called triple-enhancement-based graph neural network (TENET) is proposed, which considers global and local cellular and genetic features rather than solely leveraging the graph neural network (GNN) to model such information.
- 1
- 1
- 1
- 1
- 1
- 1
- 1
- PASH at TREC 2021 Deep Learning Track: Generative Enhanced Model for Multi-stage Ranking1
The PASH participation in TREC 2021 Deep Learning Track is described, with a scheme combining sparse and dense retrieval method and the generative model T5 to further enhance the performance.
- The Relationship Between Simplified Chinese Character Height and Cognition Research in Signage Design1
The research outcome shows the relationship between simplified Chinese character height and cognition distance of on public signage system, which provides a theoretical basis for the related research and design and revealed that the minimum of character height in the current design standard can be further increased.
- 1
- 1
- 1
- 1
- 1
- 1
- –
- –
- –
- –
- –
- –
- –
- –
- –
- –
- Optimization of the Emergency Evacuation Sign's Color Cognition for Users with Color Vision Deficiency–
The objective of the research is to build a more inclusive practical color model for improving EES and other safety sign design and conforms to the design thinking of universal design, inclusive design and human-centred design.
- –
- –
- –
- –
- –
- –
- Synchronous Colorectal Carcinoma: Three Cases Report and Literature Review–
Compared with isolated colorectal cancer, SC usually has a low stage，grade and incidence, but the age, sex and location is still a controversial issue.
- –
- –
- –
- –
- –
- –
- –
- –
- –
- –
- –
- [Clinical analysis of lymphangioleiomyomatosis].–
Clinicians should be familiar with the characteristics of LAM so that diagnosis and treatment can be made earlier to improve the prognosis.
- Feedback and Memory–
The concept of Logic Loop is brought forward from the angle of logic feedback, and it is used to analyze basic memory circuit-basic trigger.
- Design of Full-Function Control System of Counting Based on VHDL–
A full-function control system of counting is designed on a PLD device using VHDL language and shows the standardization, high-efficiency of V HDL, advantages of flexible programming, and easy updating.
- Application of High Power Operation Amplifier in Underwater Sound Transmitter–
Advantages of power operational amplifier in underwater sound transmitter and problems of power opertional amplifier with reliability and stability in application are described.
- Design of Programmable Frequency Counter Based on ispLSI 1032E Devices–
This paper provides a design means of programmable digital frequency counter by applying an ispLSI 1032E-70LJ84 device made in American Lattice company that provides the design with convenience, high-efficiency and none-risk together with the improvement of reliability and flexibility of the system.
- –
- –
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 2026-10-15. 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.