instruction stringclasses 95
values | input stringlengths 141 265k | output stringlengths 15 1.61k |
|---|---|---|
Generate 4 bullet points for the "Core Method / Technical Approach" section of an academic poster. | ## Paper Content
# Paper Title
Symbolic Representation for Any-to-Any Generative Tasks
## Abstract
We propose a symbolic generative task description language and a corresponding inference engine capable of representing arbitrary multimodal tasks as structured symbolic flows. Unlike conventional generative models th... | • Task Interpretation: Pre-trained LM analyzes natural language to identify core task requirements.
• Topology Refinement: System automatically instantiates required functions (F) and parameters (Φ).
• Iterative validation optimizes workflow connections (Γ) for executability and conflict resolution.
• Enables natural l... |
Generate 3 bullet points for the "Experimental Results / Performance Analysis" section of an academic poster. | ## Paper Content
# Paper Title
ADIFF: EXPLAINING AUDIO DIFFERENCE USING NATURAL LANGUAGE
## Abstract
Understanding and explaining differences between audio recordings is crucial for fields like audio forensics, quality assessment, and audio generation. This involves identifying and describing audio events, acoustic... | • ADIFF outperforms baselines (QwenAC, Baseline) across BLEU4, METEOR, SPIDEr metrics.
• Performance improves across all three tiers (Tier 1 to Tier 3) for both ACD and CLD tasks.
• ADIFF achieves highest scores in Tier 2 and Tier 3, indicating better handling of complex explanations. |
Generate 4 bullet points for the "Core Method / Technical Approach" section of an academic poster. | ## Paper Content
# Paper Title
DIMAT: Decentralized Iterative Merging-And-Training for Deep Learning Models
## Abstract
Recent advances in decentralized deep learning algorithms have demonstrated cutting-edge performance on various tasks with large pre-trained models. However, a pivotal prerequisite for achieving t... | • Input includes mixing matrix Π, iterations K, initialization x_i^1, step size α, and merging frequency n.
• Output is the ensemble average x̄_K.
• Algorithm iterates over agents, calculates stochastic gradients, and merges models every n iterations using weighted averaging.
• Merging step uses permutation matrix P_k^... |
Generate 4 bullet points for the "Experimental Results / Performance Analysis" section of an academic poster. | ## Paper Content
# Paper Title
Fisher Calibration for Backdoor-Robust Heterogeneous Federated Learning
## Experiments
Datasets. Following [43, 54, 75], we experiment on three federated scenarios.
- MNIST [38] is 10 digits classes with 70,000 images.
- Fashion-MNIST [72] has 60k train and 10k test examples from 1... | • Compared SDFC with various methods on MNIST, Fashion-MNIST, and CIFAR-10 datasets.
• Results show SDFC outperforms baselines across different skew degrees and attack rates.
• Quantitative plots show convergence and performance over communication epochs.
• Qualitative results show successful defense against backdoor t... |
Generate 4 bullet points for the "Core Method / Technical Approach" section of an academic poster. | ## Paper Content
# Paper Title
ENHANCING THE SCALABILITY AND APPLICABILITY OF KOHN-SHAM HAMILTONIANS FOR MOLECULAR SYSTEMS
## Abstract
Density Functional Theory (DFT) is a pivotal method within quantum chemistry and materials science, with its core involving the construction and solution of the Kohn-Sham Hamiltonia... | • WANet uses SO(2) convolution, sparse mixture of experts, and many-body interactions.
• Architecture includes SO(2) Node Convolution Layer, Mixture of LSR Experts, and Expansion Block.
• Handles scalability challenges via efficient convolutions and long-range interactions.
• Incorporates density trick for implicit 3-b... |
Generate 2 bullet points for the "Ablation Study" section of an academic poster. | ## Paper Content
# Paper Title
MART: MultiscAle Relational Transformer Networks for Multi-agent Trajectory Prediction
## Experiments
We first transform the past input trajectories into agent embeddings to represent the initial node and edge features. Because edge relationships are often not explicitly defined in re... | • Ablation on Interaction Encoder: Comparing NMMP, EqMotion, Relational Transformer, and MART.
• Ablation on Group Estimation Module: Comparing different modules in GroupNet and AGE (ours). |
Generate 4 bullet points for the "Core Method / Technical Approach" section of an academic poster. | ## Paper Content
# Paper Title
FipTR: A Simple yet Effective Transformer Framework for Future Instance Prediction in Autonomous Driving
## Method
Future instance prediction takes as input a group of multi-view images $\mathbf{I} \in \mathbb{R}^{T_{in} \times N \times 3 \times H \times W}$ from $T_{in}$ consecut... | • Matching loss combines detection (class + box) and segmentation (mask) components.
• Detection loss uses classification and bounding box regression.
• Segmentation loss sums mask losses over future timesteps.
• Multi-frame mask cost enhances temporal coherence by linking the same instance query to identical ground tr... |
Generate 2 bullet points for the "Qualitative Results / Visualization" section of an academic poster. | ## Paper Content
# Paper Title
Dynamic Adapter Meets Prompt Tuning: Parameter-Efficient Transfer Learning for Point Cloud Analysis
## Abstract
Point cloud analysis has achieved outstanding performance by transferring point cloud pre-trained models. However, existing methods for model adaptation usually update all m... | • Visualizations show DAPT's effectiveness in part segmentation for objects like tables and chairs.
• Color-coded point clouds demonstrate accurate segmentation boundaries and object part delineation. |
Generate 2 bullet points for the "Background / Related Work" section of an academic poster. | ## Paper Content
# Paper Title
PeLK: Parameter-efficient Large Kernel ConvNets with Peripheral Convolution
## Abstract
Recently, some large kernel convnets strike back with appealing performance and efficiency. However, given the square complexity of convolution, scaling up kernels can bring about an enormous amoun... | • Current CNNs compromise scaling via stripe convolution (e.g., 51×5 + 5×51)
• Dense is better but square complexity brings proliferated parameter overhead |
Generate 4 bullet points for the "Core Method / Technical Approach" section of an academic poster. | ## Paper Content
# Paper Title
DeiT-LT: Distillation Strikes Back for Vision Transformer Training on Long-Tailed Datasets
## Abstract
Vision Transformer (ViT) has emerged as a prominent architecture for various computer vision tasks. In ViT, we divide the input image into patch tokens and process them through a sta... | • Head Expert Classifier uses CLS token as classifier, trained with CE loss against ground truth.
• Tail Expert Classifier uses DIST token via flat CNN teacher, distills from CNN using OOD images with strong augmentation and Mixup.
• Employs Deferred Re-weighting loss to improve DIST token’s focus on tail classes.
• Fi... |
Generate 3 bullet points for the "Core Method / Technical Approach" section of an academic poster. | ## Paper Content
# Paper Title
Vision-Language Models Do Not Understand Negation
## Abstract
Many practical vision-language applications require models that understand negation, e.g., when using natural language to retrieve images which contain certain objects but not others. Despite advancements in vision-language... | • Develop a large-scale, LLM-guided recaptioning approach generating 70M synthetic negation captions.
• Finetuning CLIP with this data improves negation understanding significantly.
• Results show gains in R@5, R-Neg@5, and MCQ scores across multiple models. |
Generate 3 bullet points for the "Research Motivation / Problem Background" section of an academic poster. | ## Paper Content
# Paper Title
Positive-Unlabeled Learning by Latent Group-Aware Meta Disambiguation
## Abstract
Positive-Unlabeled (PU) learning aims to train a binary classifier using minimal positive data supplemented by a substantially larger pool of unlabeled data, in the specific absence of explicitly annotat... | • Real-world PU scenarios aggregate binary classes from finer-grained categories.
• Primary bottleneck: developing discriminative representations under unreliable yet semantically poor binary labels.
• Key question: how to produce compact representations encapsulating intrinsic group semantics? |
Generate 4 bullet points for the "Research Motivation / Problem Background" section of an academic poster. | ## Paper Content
# Paper Title
Don't drop your samples! Coherence-aware training benefits Conditional diffusion
## Abstract
Conditional diffusion models are powerful generative models that can leverage various types of conditional information, such as class labels, segmentation masks, or text captions. However, in ... | • Goal: Learn diffusion models in the presence of conditioning noise
• Approach: Estimate conditioning alignment with a coherence score
• Prior work: Filtering the dataset into a high coherence subset (Rombach 2022)
• Our method: Keep every sample but condition the model by the coherence score |
Generate 3 bullet points for the "Core Method / Technical Approach" section of an academic poster. | ## Paper Content
# Paper Title
GaussCtrl: Multi-View Consistent Text-Driven 3D Gaussian Splatting Editing
## Method
We propose GaussCtrl, a novel approach to edit a 3D Gaussian Splatting (3DGS) model using textual prompts. Given a collection of images and their reconstructed 3D model, our method first re-renders ea... | • Utilizes ControlNet as editing backbone
• Follows DDIM Inversion approach for 2D editing
• Conditions editing on depth information to encourage multi-view consistency |
Generate 3 bullet points for the "Method Overview / Framework" section of an academic poster. | ## Paper Content
# Paper Title
Geometry Transfer for Stylizing Radiance Fields
## Abstract
Shape and geometric patterns are essential in defining stylistic identity. However, current 3D style transfer methods predominantly focus on transferring colors and textures, often overlooking geometric aspects. In this paper... | • Introduce Geometry Transfer: extracts geometry from depth map and stylizes radiance fields.
• Propose novel deformation field usage to ensure coherent stylization of shape and appearance.
• Introduce several techniques enhancing expressiveness by leveraging geometries. |
Generate 3 bullet points for the "Core Method / Technical Approach" section of an academic poster. | ## Paper Content
# Paper Title
Rethinking Vision-Language Model in Face Forensics: Multi-Modal Interpretable Forged Face Detector
## Abstract
Deepfake detection is a long-established research topic vital for mitigating the spread of malicious misinformation. Unlike prior methods that provide either binary classific... | • The M2F2-Det main architecture includes CLIP Image and Text Encoder, a deepfake encoder, and an LLM. The probability of generating answer tokens as (H and X are visual and text tokens, respectively).
• Forgery Prompt Learning (FPL) adapts CLIP to deepfake detection by optimizing UF-prompts and layer-wise forgery toke... |
Generate 3 bullet points for the "Experimental Results / Performance Analysis" section of an academic poster. | ## Paper Content
# Paper Title
MULDE: Multiscale Log-Density Estimation via Denoising Score Matching for Video Anomaly Detection
## Abstract
We propose a novel approach to video anomaly detection: we treat feature vectors extracted from videos as realizations of a random variable with a fixed distribution and model... | • MULDE achieves state-of-the-art performance on ShanghaiTech, UCF-Crime, and UBnormal datasets
• AUC scores show consistent improvement over baselines
• Anomaly score plots demonstrate accurate detection of ground truth anomaly frames |
Generate 3 bullet points for the "Method Overview / Framework" section of an academic poster. | ## Paper Content
# Paper Title
BPQP: A Differentiable Convex Optimization Framework for Efficient End-to-End Learning
## Abstract
Data-driven decision-making processes increasingly utilize end-to-end learnable deep neural networks to render final decisions. Sometimes, the output of the forward functions in certain ... | • Introduce Backward Pass as a Quadratic Programming (BPQP), enabling efficient large-scale gradients computation.
• Novel Backward Pass Reformulation: BPQP decouples the backward pass from the forward pass, reformulating it as a simple QP problem leveraging KKT matrix structure.
• Flexible Solver Choice: BPQP accommod... |
Generate 2 bullet points for the "Ablation Study" section of an academic poster. | ## Paper Content
# Paper Title
SocialCircle: Learning the Angle-based Social Interaction Representation for Pedestrian Trajectory Prediction
## Abstract
Analyzing and forecasting trajectories of agents like pedestrians and cars in complex scenes has become more and more significant in many intelligent systems and a... | • We add manual neighbors with different velocities v_m and distances d_m to the target agent to validate how different SocialCircle meta components modify predicted trajectories (Fig. 8).
• We also validate the contribution of each circle partition by visualizing the square-summed representation f^i(θ_n) (Fig. 7). |
Generate 2 bullet points for the "Other Content" section of an academic poster. | ## Paper Content
# Paper Title
Can Large Vision-Language Models Correct Semantic Grounding Errors By Themselves?
## Abstract
Improving semantic grounding in Vision-Language Models (VLMs) often involves collecting domain-specific training data, refining the network architectures, or modifying the training recipes. I... | • EMNLP'24: Enhancing quantitative spatial reasoning with no extra training; training-free, spatial-reasoning.
• arXiv'25: Synthesizing System-2 reasoning traces for System-1 Perception; synthetic-data-generation, cognitive-behaviors. |
Generate 2 bullet points for the "Core Method / Technical Approach" section of an academic poster. | ## Paper Content
# Paper Title
HumanGaussian: Text-Driven 3D Human Generation with Gaussian Splatting
## Abstract
Realistic 3D human generation from text prompts is a desirable yet challenging task. Existing methods optimize 3D representations like mesh or neural fields via score distillation sampling (SDS), which ... | • Structure-Aware SDS: Initialize 3DGS center positions based on SMPL-X mesh shape; train an SDS model capturing joint distribution of texture and structure; guide 3DGS optimization from both structural and textural aspects.
• Annealed Negative Prompt Guidance: Regularize gradient with negative prompts using NFSD and C... |
Generate 3 bullet points for the "Qualitative Results / Visualization" section of an academic poster. | ## Paper Content
# Paper Title
4Deform: Neural Surface Deformation for Robust Shape Interpolation
## Abstract
Generating realistic intermediate shapes between non-rigidly deformed shapes is a challenging task in computer vision, especially with unstructured data (e.g., point clouds) where temporal consistency acros... | • Handles partial input shapes.
• Shows interpolation results for partial point clouds.
• Outperforms baselines in handling incomplete data. |
Generate 3 bullet points for the "Other Content" section of an academic poster. | ## Paper Content
# Paper Title
BrepGiff: Lightweight Generation of Complex B-rep with 3D GAT Diffusion
## Abstract
Despite advancements in Computer-Aided-Design (CAD) generation, direct generation of complex Boundary Representation (B-rep) CAD models remains challenging. The difficulty arises from the parametric na... | • Fundamental Research Funds for the Central Universities.
• Natural Science Basic Research Program of Shaanxi Province (Grant 2024JC-YBQN-0702).
• NUS Research Scholarship and NUS ORIA. |
Generate 3 bullet points for the "Experimental Results / Performance Analysis" section of an academic poster. | ## Paper Content
# Paper Title
CAN NEURAL NETWORKS ACHIEVE OPTIMAL COMPUTATIONAL-STATISTICAL TRADE-OFF? AN ANALYSIS ON SINGLE-INDEX MODEL
## Abstract
In this work, we tackle the following question: Can neural networks trained with gradient-based methods achieve the optimal statistical-computational tradeoff in lear... | • Introduces a unified gradient-based method that achieves the statistical lower bound
• Works across nearly all sparsity levels
• Demonstrates theoretical optimality under broad conditions |
Generate 2 bullet points for the "Core Method / Technical Approach" section of an academic poster. | ## Paper Content
# Paper Title
STING-BEE : Towards Vision-Language Model for Real-World X-ray Baggage Security Inspection
## Abstract
Advancements in Computer-Aided Screening (CAS) systems are essential for improving the detection of security threats in X-ray baggage scans. However, current datasets are limited in ... | • STCray includes 57,218 threat instances with captions, bounding boxes, and pixel-level labels.
• Dataset covers 21 threat categories including knives, guns, explosives, and 3D-printed threats. |
Generate 3 bullet points for the "Experimental Results / Performance Analysis" section of an academic poster. | ## Paper Content
# Paper Title
PROBLEM-PARAMETER-FREE FEDERATED LEARNING
## Abstract
Federated learning (FL) has garnered significant attention from academia and industry in recent years due to its advantages in data privacy, scalability, and communication efficiency. However, current FL algorithms face a critical ... | • Compares PAdaMFed with SCAFFOLD, Mime, FedSPS, SCAFFOLD-M, FAFED, and SCAFFOLD-M-VR across assumptions, stepsize restrictions, problem parameters, and communication complexity.
• Highlights that PAdaMFed requires no additional assumptions and has favorable communication complexity O(1/SKε²).
• Shows PAdaMFed-VR varia... |
Generate 2 bullet points for the "Core Method / Technical Approach" section of an academic poster. | ## Paper Content
# Paper Title
COMPOSING UNBALANCED FLOWS FOR FLEXIBLE DOCKING AND RELAXATION
## Abstract
Diffusion models have emerged as a successful approach for molecular docking, but they often cannot model protein flexibility or generate nonphysical poses. We argue that both these challenges can be tackled by... | • Tradeoff between approximation error and sampling efficiency in the flow matching framework.
• Energy loss and filter terms are introduced to improve pose quality. |
Generate 2 bullet points for the "Research Motivation / Problem Background" section of an academic poster. | ## Paper Content
# Paper Title
RoDUS: Robust Decomposition of Static and Dynamic Elements in Urban Scenes
## Introduction
There have been numerous works that achieved success in various 3D representations including both implicit [23] and explicit [36,37] scene modeling. Therefore, reconstructing and decomposing lar... | • Problem statement: Without dynamic masks or bounding boxes as guidance, static/dynamic NeRF baselines often get stuck in local minima and struggle with separating dynamic objects in challenging urban setups.
• Contributions: A self-supervised NeRF pipeline for decomposing static and dynamic elements, leveraging both ... |
Generate 3 bullet points for the "Core Method / Technical Approach" section of an academic poster. | ## Paper Content
# Paper Title
Masked Spatial Propagation Network for Sparsity-Adaptive Depth Refinement
## Abstract
The main function of depth completion is to compensate for an insufficient and unpredictable number of sparse depth measurements of hardware sensors. However, existing research on depth completion as... | • Uses an off-the-shelf monocular depth estimator (MDE) to generate initial depth.
• Guidance network extracts features from RGB image.
• Iterative refinement via MSPN updates depth and mask over N iterations using attention and multiplication operations. |
Generate 3 bullet points for the "Experimental Results / Performance Analysis" section of an academic poster. | ## Paper Content
# Paper Title
REGENESIS: LLMs CAN GROW INTO REASONING GENERALISTS VIA SELF-IMPROVEMENT
## Abstract
Post-training Large Language Models (LLMs) with explicit reasoning trajectories can enhance their reasoning abilities. However, acquiring such high-quality trajectory data typically demands meticulous... | • ReGenesis shows 6.1% improvement compared to original model without finetuning.
• Fine-tuning language model solely on ground-truth solutions (FT w/ GT) significantly decreases performance on all OOD tasks.
• Other baselines like CoT prompting task-specific reasoning paths (LMSI w/ GT, StaR) do NOT improve performanc... |
Generate 3 bullet points for the "Core Method / Technical Approach" section of an academic poster. | ## Paper Content
# Paper Title
Beyond Matryoshka: Revisiting Sparse Coding for Adaptive Representation
## Abstract
Many large-scale systems rely on high-quality deep representations (embeddings) to facilitate tasks like retrieval, search, and generative modeling. Matryoshka Representation Learning (MRL) recently em... | • Uses Torch.sparse.mm() for sparse matrix multiplication to enable faster similarity search.
• Shows that larger embedding dimensions R^h can achieve faster retrieval with more information.
• Demonstrates CSR handles larger datasets with accelerated efficiency gains at scale. |
Generate 3 bullet points for the "Core Method / Technical Approach" section of an academic poster. | ## Paper Content
# Paper Title
WeConvene: Learned Image Compression with Wavelet-Domain Convolution and Entropy Model
## Method
Fig. 3 shows the details of the proposed WeConv and inverse WeConv modules. In this paper, we use WeConv when the size of the latent representation is changed (this might not be necessary ... | • Core encoder ga and hyper encoder ha.
• Core decoder gs and hyper decoder hs.
• Integrates wavelet-domain convolution and entropy modeling for improved compression. |
Generate 2 bullet points for the "Core Method / Technical Approach" section of an academic poster. | ## Paper Content
# Paper Title
LayeredFlow: A Real-World Benchmark for Non-Lambertian Multi-Layer Optical Flow
## Method
Hongyu Wen, Erich Liang, and Jia Deng
Department of Computer Science, Princeton University {hongyu.wen, erliang, jiadeng}@princeton.edu
Abstract. Achieving 3D understanding of non-Lambertian ob... | • We utilize AprilTag Markers to collect ground-truth.
• We manually annotate each marker with its material property (transparent, reflective or diffuse) and layer index to provide multi-layer ground-truth. |
Generate 3 bullet points for the "Research Motivation / Problem Background" section of an academic poster. | ## Paper Content
# Paper Title
Choose What You Need: Disentangled Representation Learning for Scene Text Recognition, Removal and Editing
## Abstract
Scene text images contain not only style information (font, background) but also content information (character, texture). Different scene text tasks need different i... | • Scene text images contain both style (font, background) and content (character, texture) information, requiring different features.
• In STR, content features are essential; style is noise.
• In STE, removal needs content features; rendering needs style features. |
Generate 2 bullet points for the "Core Method / Technical Approach" section of an academic poster. | ## Paper Content
# Paper Title
Medusa: A Multi-Scale High-order Contrastive Dual-Diffusion Approach for Multi-View Clustering
## Abstract
Deep multi-view clustering methods utilize information from multiple views to achieve enhanced clustering results and have gained increasing popularity in recent years. Most exis... | • We construct a hypergraph for each view to capture rich high-order relationships.
• We use contrastive learning to better harness high-order information by treating samples within the same hyperedge as positive pairs. |
Generate 2 bullet points for the "Core Method / Technical Approach" section of an academic poster. | ## Paper Content
# Paper Title
PromptIQA: Boosting the Performance and Generalization for No-Reference Image Quality Assessment via Prompts
## Method
In this part, we firstly formalize paradigms of typical IQA models and the PromptIQA. Then, we provide a detailed information to each part of PromptIQA.
Given an ima... | • (A) Typical IQA Framework: Takes image as input, outputs score without prompt guidance.
• (B) Prompt-Based IQA Framework: Uses prompts from Koniq and Livec datasets to guide score prediction, enabling adaptation to different assessment requirements. |
Generate 2 bullet points for the "Experimental Results / Performance Analysis" section of an academic poster. | ## Paper Content
# Paper Title
Hybrid Functional Maps for Crease-Aware Non-Isometric Shape Matching
## Abstract
Non-isometric shape correspondence remains a fundamental challenge in computer vision. Traditional methods using Laplace-Beltrami operator (LBO) eigenmodes face limitations in characterizing high-frequenc... | • Quantitative evaluation across multiple datasets (FAUST, SCAPE, SHREC'19, SMAL, DT4D-H, TOPKIDS) showing our Hybrid GeomFMaps outperforming baselines in most categories.
• Ablation study in Table 1 shows the effect of adapting to non-orthogonal basis, with our approach achieving 3.83 ± 0.74 geometric error, outperfor... |
Generate 2 bullet points for the "Experimental Results / Performance Analysis" section of an academic poster. | ## Paper Content
# Paper Title
REVISITING TEXT-TO-IMAGE EVALUATION WITH GECKO: ON METRICS, PROMPTS AND HUMAN RAT-ING
## Abstract
While text-to-image (T2I) generative models have become ubiquitous, they do not necessarily generate images that align with a given prompt. While many metrics and benchmarks have been pro... | • Compares Model 1 and Model 2 using Likert, Word Annotation, QA-based, and Side-by-Side methods
• Shows inconsistent metric rankings across annotation types (e.g., Metric A < Metric B in some cases, Metric A > Metric B in others) |
Generate 4 bullet points for the "Core Method / Technical Approach" section of an academic poster. | ## Paper Content
## Abstract
The development of vision-language models (VLMs) is driven by large-scale and diverse multi-modal datasets. However, progress toward generalist biomedical VLMs is limited by the lack of annotated, publicly accessible datasets across biology and medicine. Existing efforts are limited to na... | • Extract: Process PMC-OA articles to extract metadata, figures, captions, and full-text.
• Transform: Apply concept labeling with human supervision, PCA + k-means clustering, and label propagation to generate 170 annotated concepts.
• Load: Serialize data into 24M image-caption pairs, 30M figure references, 6M full-ar... |
Generate 3 bullet points for the "Conclusion / Future Work" section of an academic poster. | ## Paper Content
# Paper Title
Human-centered Interactive Learning via MLLMs for Text-to-Image Person Re-identification
## Abstract
Despite remarkable advancements in text-to-image person re-identification (TIREID) facilitated by the breakthrough of cross-modal embedding models, existing methods often struggle to d... | • Extensive experiments on four TIRelD benchmarks demonstrate remarkable performance with substantial improvement.
• Code is available at https://github.com/QinYang79/ICL.
• Future work includes extending ICL to other cross-modal retrieval tasks. |
Generate 3 bullet points for the "Core Method / Technical Approach" section of an academic poster. | ## Paper Content
# Paper Title
Unknown Prompt, the only Lacuna: Unveiling CLIP's Potential for Open Domain Generalization
## Abstract
We delve into Open Domain Generalization (ODG), marked by domain and category shifts between training's labeled source and testing's unlabeled target domains. Existing solutions to O... | • Incorporates class information into visual space using learnable context and stable diffusion.
• Uses a contrastive loss (L_sem) to align textual and visual features for known and unknown classes.
• Applies a domain-specific loss (L_dom) to enhance domain-agnostic generalization. |
Generate 3 bullet points for the "Core Method / Technical Approach" section of an academic poster. | ## Paper Content
# Paper Title
Language Model Guided Interpretable Video Action Reasoning
## Abstract
While neural networks have excelled in video action recognition tasks, their "black-box" nature often obscures the understanding of their decision-making processes. Recent approaches used inherently interpretable m... | • Method exploits an interpretable language model to guide the video model in capturing relationship transitions during training.
• During inference, the model processes videos and directly recognizes actions, providing supportive evidence.
• Two-phase design: training with language guidance, inference with direct reco... |
Generate 4 bullet points for the "Method Overview / Framework" section of an academic poster. | ## Paper Content
# Paper Title
Meta-Point Learning and Refining for Category-Agnostic Pose Estimation
## Abstract
Category-agnostic pose estimation (CAPE) aims to predict keypoints for arbitrary classes given a few support images annotated with keypoints. Existing methods only rely on the features extracted at supp... | • First to learn class-agnostic potential keypoints for CAPE.
• Propose a novel framework based on meta-point learning and refining.
• Introduce progressive deformable point decoder and slacked regression loss.
• Our method produces meaningful meta-points and predicts more precise keypoints than existing methods. |
Generate 2 bullet points for the "Experimental Results / Performance Analysis" section of an academic poster. | ## Paper Content
# Paper Title
Deterministic Image-to-Image Translation via Denoising Brownian Bridge Models with Dual Approximators
## Abstract
Image-to-Image (I2I) translation involves converting an image from one domain to another. Deterministic I2I translation, such as in image super-resolution, extends this co... | • Our method achieves 48.70 FID, 15.70 PSNR, and 53.26% SSIM on Cityscape, outperforming Pix2pix, CycleGAN, and UNIT.
• Lower FID and higher PSNR/SSIM indicate better image quality and structural similarity. |
Generate 3 bullet points for the "Qualitative Results / Visualization" section of an academic poster. | ## Paper Content
# Paper Title
PoseCrafter: One-Shot Personalized Video Synthesis Following Flexible Pose Control
## Experiments
Our goal with a human-centric reference video is to create a high-quality video that follows a specific target pose sequence while preserving the identity from the reference. This task is... | • Figure 2 presents qualitative comparisons of all methods with M=100 frames.
• Left side shows results on TED dataset; right side on TikTok dataset.
• PoseCrafter yields the highest quality videos, with smoother motion and better pose alignment. |
Generate 3 bullet points for the "Experimental Results / Performance Analysis" section of an academic poster. | ## Paper Content
# Paper Title
Generalized Predictive Model for Autonomous Driving
## Abstract
In this paper, we introduce the first large-scale video prediction model in the autonomous driving discipline. To eliminate the restriction of high-cost data collection and empower the generalization ability of our model,... | • ST-P3*: ADE 2.65, FDE 3.73
• UniAD*: ADE 1.03, FDE 1.65
• GenAD (Ours): ADE 1.23, FDE 2.31 |
Generate 3 bullet points for the "Core Method / Technical Approach" section of an academic poster. | ## Paper Content
# Paper Title
HDQMF: Holographic Feature Decomposition Using Quantum Algorithms
## Abstract
This paper addresses the decomposition of holographic feature vectors in Hyperdimensional Computing (HDC) aka Vector Symbolic Architectures (VSA). HDC uses high-dimensional vectors with brain-like properties... | • Memorizing: Stores object representations.
• Querying: Asks questions like "Which object is on the left?", "Who’s the protagonist?", "What’s are the active clusters?"
• Reasoning: Uses Raven’s Progressive Matrix, Theme Analysis, Cognitive Anomaly Detection. |
Generate 4 bullet points for the "Research Motivation / Problem Background" section of an academic poster. | ## Paper Content
# Paper Title
MVBoost: Boost 3D Reconstruction with Multi-View Refinement
## Abstract
Recent advancements in 3D object reconstruction have been remarkable, yet most current 3D models rely heavily
on existing 3D datasets. The scarcity of diverse 3D datasets results in limited generalization capabil... | • Single-view 3D generation models suffer from insufficient 3D data, resulting in poor model generalization and inability to generate high-fidelity 3D assets.
• 3D synthetic data has issues such as low quality and multi-view inconsistency.
• We present a novel framework for constructing pseudo ground truth using high-a... |
Generate 3 bullet points for the "Experimental Results / Performance Analysis" section of an academic poster. | ## Paper Content
# Paper Title
VL-RewardBench: A Challenging Benchmark for Vision-Language Generative Reward Models
## Abstract
Vision-language generative reward models (VL-GenRMs) play a crucial role in aligning and evaluating multimodal AI systems, yet their own evaluation remains under-explored. Current assessme... | • Compares performance of various open-source and proprietary models across categories.
• Highlights top performers like GPT-4o and Claude-3.5-Sonnet.
• Shows metrics such as Overall Accuracy and Macro Average Accuracy for each model. |
Generate 4 bullet points for the "Research Motivation / Problem Background" section of an academic poster. | ## Paper Content
# Paper Title
T2ISafety: Benchmark for Assessing Fairness, Toxicity, and Privacy in Image Generation
## Abstract
Text-to-image (T2I) models have rapidly advanced, enabling the generation of high-quality images from text prompts across various domains. However, these models present notable safety co... | • Risks of generating harmful, biased, or private content.
• External safety filters are fragile and easily bypassed.
• Lack of large-scale, labeled safety datasets.
• Incomplete and unreliable risk evaluation. |
Generate 2 bullet points for the "Method Overview / Framework" section of an academic poster. | ## Paper Content
# Paper Title
Question-Aware Gaussian Experts for Audio-Visual Question Answering
## Abstract
Audio-Visual Question Answering (AVQA) requires not only question-based multimodal reasoning but also precise temporal grounding to capture subtle dynamics for accurate prediction. However, existing method... | • QA-TIGER progressively refines attention using question-aware fusion and Gaussian experts to capture continuous, question-aligned temporal dynamics.
• Question-aware attention and Gaussian weights are traceable, making the model’s predictions interpretable. |
Generate 2 bullet points for the "Background / Related Work" section of an academic poster. | ## Paper Content
# Paper Title
Data-Efficient Multimodal Fusion on a Single GPU
## Abstract
The goal of multimodal alignment is to learn a single latent space that is shared between multimodal inputs. The most powerful models in this space have been trained using massive datasets of paired inputs and large-scale co... | • Given any two data modalities X and Y (e.g., image and text): Aims to learn two networks, fX: X → S and fY: Y → S, that embed each respective modality into a shared latent space S.
• Commonly trained with a symmetric InfoNCE loss [3] (contrastive learning), defined as: L = -E[log(exp(fX(x)·fY(y)/τ) / Σ exp(fX(x)·fY(y... |
Generate 3 bullet points for the "Core Method / Technical Approach" section of an academic poster. | ## Paper Content
# Paper Title
ProMerge: Prompt and Merge for Unsupervised Instance Segmentation
## Method
In this section, we first describe the problem scenario (Sec. 3.1) and introduce ProMerge, a simple yet effective prompt-and-merge method to tackle unsupervised instance segmentation (Sec. 3.2). We then descri... | • Not all mask proposals are foreground masks.
• To filter background masks among the prompted masks, we take a two-step strategy: (i) Background aggregation and (ii) Cascade filtering.
• Background aggregation combines candidate masks; cascade filtering iteratively refines masks using IoA thresholds. |
Generate 2 bullet points for the "Experimental Results / Performance Analysis" section of an academic poster. | ## Paper Content
# Paper Title
Consistent 3D Line Mapping
## Experiments
We compare our system with two state-of-the-art methods: L3D++ [11] and LIMAP [24], using the traditional LSD line detector [42] and the learning-based DeepLSD line detector [32]. We consider the synthetic Hypersim [35] and the real Tanks and ... | • Comparison with SOTA Methods: Evaluate on DeepLSD, Line, L3D++, LIMAP, and Ours across multiple datasets (ai_001_001 to Caterpillar).
• Ablation Studies: Evaluate contributions of G (proposal generation), S (best proposal selection), and B (line track building) on Hypersim dataset. |
Generate 2 bullet points for the "Research Motivation / Problem Background" section of an academic poster. | ## Paper Content
# Paper Title
RNG: Relightable Neural Gaussians
## Abstract
3D Gaussian Splatting (3DGS) has shown impressive results for the novel view synthesis task, where lighting is assumed to be fixed. However, creating relightable 3D assets, especially for objects with ill-defined shapes (fur, fabric, etc.)... | • Creating relightable 3D assets, especially for objects with ill-defined shapes (fur, fabric, etc.) is a challenging task.
• The decomposition between light, geometry, and material is ambiguous, especially if either smooth surface assumptions or surface-based analytical shading models do not apply. |
Generate 2 bullet points for the "Ablation Study" section of an academic poster. | ## Paper Content
# Paper Title
Steering Away from Harm: An Adaptive Approach to Defending Vision Language Model Against Jailbreaks
## Abstract
Vision Language Models (VLMs) can produce unintended and harmful content when exposed to adversarial attacks, particularly because their vision capabilities create new vulne... | • Ablation study confirms both image attribution and activation calibration are necessary.
• Shows toxicity score decreases as steering vector or calibration activation is applied, with best performance when both are used. |
Generate 4 bullet points for the "Core Method / Technical Approach" section of an academic poster. | ## Paper Content
# Paper Title
Rethinking Temporal Fusion with a Unified Gradient Descent View for 3D Semantic Occupancy Prediction
## Abstract
We present GDFusion, a temporal fusion method for vision-based 3D semantic occupancy prediction (VisionOcc). GDFusion opens up the underexplored aspects of temporal fusion ... | • Heterogeneous representations of each temporal cue make fusion non-trivial.
• The framework reinterprets standard RNN updates as gradient descent steps to minimize discrepancies.
• For each temporal cue, a loss function is defined to measure discrepancies between current and historical states.
• Historical residuals ... |
Generate 3 bullet points for the "Ablation Study" section of an academic poster. | ## Paper Content
# Paper Title
CNC-Net: Self-Supervised Learning for CNC Machining Operations
## Abstract
CNC manufacturing is a process that employs computer numerical control (CNC) machines to govern the movements of various industrial tools and machinery, encompassing equipment ranging from grinders and lathes t... | • Ablation table shows effect of removing operations (O^M, O^D, O^R) and users (U^M, U^D, U^R).
• Removing milling operations (O^M) reduces performance significantly.
• User selection (U^M, U^D) improves results moderately. |
Generate 2 bullet points for the "Qualitative Results / Visualization" section of an academic poster. | ## Paper Content
# Paper Title
CMA: A Chromaticity Map Adapter for Robust Detection of Screen-Recapture Document Images
## Abstract
The rebroadcasting of screen-recaptured document images introduces a significant risk to the confidential documents processed in government departments and commercial companies. Howeve... | • Visual comparison of document low-quality images and chromaticity maps under various distortions.
• Shows how chromaticity maps preserve forensic cues even under severe compression and blurring. |
Generate 4 bullet points for the "Core Method / Technical Approach" section of an academic poster. | ## Paper Content
# Paper Title
AdaGlimpse: Active Visual Exploration with Arbitrary Glimpse Position and Scale
## Method
Glimpses. Let $X$ be an unobserved scene to explore. We assume $X$ to be a rectangle within the Cartesian coordinate system. A glimpse $G$ is a square region within $X$ that can be observ... | • AdaGlimpse combines a vision transformer-based encoder with a task-specific head and Soft Actor-Critic RL agent.
• Encoder fuses information from all previous observations into a single environment representation.
• Task head is a linear layer for classification and a transformer for reconstruction/segmentation.
• RL... |
Generate 3 bullet points for the "Research Motivation / Problem Background" section of an academic poster. | ## Paper Content
# Paper Title
VP3D: Unleashing 2D Visual Prompt for Text-to-3D Generation
## Abstract
Recent innovations on text-to-3D generation have featured Score Distillation Sampling (SDS), which enables the zero-shot learning of implicit 3D models (NeRF) by directly distilling prior knowledge from 2D diffusi... | • We introduce a Visual Prompt-guided text-to-3D generation model (VP3D) that explicitly unleashes visual appearance knowledge in 2D visual prompt to boost text-to-3D generation.
• VP3D first capitalizes on 2D diffusion model to generate an image from input text, which subsequently acts as visual prompt to strengthen S... |
Generate 3 bullet points for the "Core Method / Technical Approach" section of an academic poster. | ## Paper Content
# Paper Title
Sampling Innovation-Based Adaptive Compressive Sensing
## Abstract
Scene-aware Adaptive Compressive Sensing (ACS) has attracted significant interest due to its promising capability for efficient and high-fidelity acquisition of scene images. ACS typically prescribes adaptive sampling ... | • Defines Innovation Criterion as the L2 norm difference between estimated and actual signals.
• Implements Innovation-guided adaptive sampling across multiple stages.
• Uses lightweight networks and sum pooling for sampling resource allocation. |
Generate 3 bullet points for the "Core Method / Technical Approach" section of an academic poster. | ## Paper Content
# Paper Title
Dynamic Motion Blending for Versatile Motion Editing
## Abstract
Text-guided motion editing enables high-level semantic control and iterative modifications beyond traditional keyframe animation. Existing methods rely on limited precollected training triplets (original motion, edited m... | • Diffusion-based motion editing model with Transformer Encoder and MLP components.
• Includes body part coordinator, instruction encoder, and auto-regressive diffusion sampler.
• Online data augmentation via motion composition and mean-pooling. |
Generate 4 bullet points for the "Research Motivation / Problem Background" section of an academic poster. | ## Paper Content
# Paper Title
A Unified, Resilient, and Explainable Adversarial Patch Detector
## Abstract
Deep Neural Networks (DNNs), backbone architecture in 'almost' every computer vision task, are vulnerable to adversarial attacks, particularly physical out-of-distribution (OOD) adversarial patches. Existing ... | • Deep Neural Networks (DNNs) are highly vulnerable to physical adversarial attacks.
• Physical adversarial patches are small, imperceptible patterned sub-images used by attackers to deceive AI systems.
• Limitations include lack of robustness, generalization, and explainability, especially in out-of-distribution scena... |
Generate 1 bullet points for the "Research Motivation / Problem Background" section of an academic poster. | ## Paper Content
# Paper Title
AM-RADIO: Agglomerative Vision Foundation Model Reduce All Domains Into One
## Abstract
A handful of visual foundation models (VFMs) have recently emerged as the backbones for numerous downstream tasks. VFMs like CLIP, DINOv2, SAM are trained with distinct objectives, exhibiting uniqu... | • Create a unified visual foundation model (VFM) by distilling from multiple distinct VFM teachers |
Generate 3 bullet points for the "Research Motivation / Problem Background" section of an academic poster. | ## Paper Content
# Paper Title
VLog: Video-Language Models by Generative Retrieval of Narration Vocabulary
## Abstract
Human daily activities can be concisely narrated as sequences of routine events (e.g., turning off an alarm) in video streams, forming an event vocabulary. Motivated by this, we introduce VLog, a n... | • LLM vocab (subword tokens like 'pot') may lack visual meaning
• Token-wise generation is slow in inference
• Proposes narration-level daily vocab (e.g., 'wake up', 'take a bath') |
Generate 3 bullet points for the "Experimental Results / Performance Analysis" section of an academic poster. | ## Paper Content
# Paper Title
ROOT CAUSE ANALYSIS OF ANOMALIES IN MULTIVARIATE TIME SERIES THROUGH GRANGER CAUSAL DISCOVERY
## Abstract
Identifying the root causes of anomalies in multivariate time series is challenging due to the complex dependencies among the series. In this paper, we propose a comprehensive app... | • Datasets: Linear, Nonlinear, Lorenz 96, SWaT, MSDS with training/test splits and root variable counts.
• Causal Discovery: AERCA achieves superior performance across all datasets.
• Root Cause Analysis: AERCA shows exceptional performance even at AC@1 metric; performance declines for SWaT due to violations of assumpt... |
Generate 2 bullet points for the "Implementation Details" section of an academic poster. | ## Paper Content
# Paper Title
Efficient Test-time Adaptive Object Detection via Sensitivity-Guided Pruning
## Abstract
Continual test-time adaptive object detection (CTTA-OD) aims to online adapt a source pre-trained detector to ever-changing environments during inference under continuous domain shifts. Most exist... | • Pruning channels at BN layers in ResNet backbone reduces both the output channels of the preceding convolution and the input channels of the following convolution.
• Removing their corresponding parameters and physically altering the model structure and computational graph at test time. |
Generate 3 bullet points for the "Research Motivation / Problem Background" section of an academic poster. | ## Paper Content
# Paper Title
Memory-Efficient Fine-Tuning for Quantized Diffusion Model
## Introduction
Diffusion models have been a de facto standard in generative models, especially in image synthesis [6,16,34,40,44]. They are widely used in various applications, such as image super-resolution [28,45], inpainti... | • Fine-tuning large-scale diffusion models (SDXL, SD3, FLUX, etc.) requires huge amounts of memory.
• Quantized diffusion models are memory-efficient and suitable for end-users without high-performance GPUs.
• We propose the first method for fine-tuning quantized diffusion models on downstream tasks. |
Generate 2 bullet points for the "Core Method / Technical Approach" section of an academic poster. | ## Paper Content
# Paper Title
Unleashing the Potential of Consistency Learning for Detecting and Grounding Multi-Modal Media Manipulation
## Abstract
To tackle the threat of fake news, the task of detecting and grounding multi-modal media manipulation $(DGM^4)$ has received increasing attention. However, most st... | • The framework includes a multi-modal encoder, consistency processor, and dual decoders (contextual and semantic) with forgery-aware reasoning and aggregating modules.
• Outputs are fed into classifiers and a threshold filter for final detection and grounding. |
Generate 4 bullet points for the "Core Method / Technical Approach" section of an academic poster. | ## Paper Content
# Paper Title
TokenMotion: Decoupled Motion Control via Token Disentanglement for Human-centric Video Generation
## Abstract
Human-centric motion control in video generation remains a critical challenge, particularly when jointly controlling camera movements and human poses in scenarios like the ic... | • Decoupled encoding of camera and human-motion conditions using separate representations.
• Fusing motion representation with human-pose guided dynamic mask to learn localized human motion and global camera conditions.
• Restricting model to access only fused motion to force motion decoupling by minimizing distance fr... |
Generate 2 bullet points for the "Method Overview / Framework" section of an academic poster. | ## Paper Content
# Paper Title
Adapting Visual-Language Models for Generalizable Anomaly Detection in Medical Images
## Abstract
Recent advancements in large-scale visual-language pre-trained models have led to significant progress in zero/few-shot anomaly detection within natural image domains. However, the substa... | • A novel multi-level feature adaptation framework, the first attempt to adapt pre-trained visual-language models for medical AD in zero-/few-shot scenarios.
• Extensive experiments on a challenging benchmark for AD in medical images have demonstrated its exceptional generalizability across diverse data modalities and ... |
Generate 3 bullet points for the "Core Method / Technical Approach" section of an academic poster. | ## Paper Content
# Paper Title
OpticalDR: A Deep Optical Imaging Model for Privacy-Protective Depression Recognition
## Abstract
Depression Recognition (DR) poses a considerable challenge, especially in the context of the growing concerns surrounding privacy. Traditional automatic diagnosis of DR technology necessi... | • OpticalIDR combines an optical lens and deep learning model, jointly optimized with auxiliary tasks.
• Integrates facial structure, emotions, and depression knowledge.
• Strips away facial identification info to ensure unrecognizable outputs. |
Generate 2 bullet points for the "Research Motivation / Problem Background" section of an academic poster. | ## Paper Content
# Paper Title
REWIND: Real-Time Egocentric Whole-Body Motion Diffusion with Exemplar-Based Identity Conditioning
## Abstract
We present REWIND (Real-Time Egocentric Whole-Body Motion Diffusion), a one-step diffusion model for real-time, high-fidelity human motion estimation from egocentric image in... | • Most existing works focus on egocentric body-only motion estimation, overlooking the importance of hand motions in enhancing realism in VR/AR.
• Recent work on egocentric whole-body motion estimation is non-real-time and acausal, limiting its practical applicability. |
Generate 3 bullet points for the "Core Method / Technical Approach" section of an academic poster. | ## Paper Content
# Paper Title
Pathology-knowledge Enhanced Multi-instance Prompt Learning for Few-shot Whole Slide Image Classification
## Method
We provide a brief introduction to the problem definition of FSWC and the pre-trained VLM used in our study. Given a dataset $W = \{W_{1}, W_{2}, \ldots, W_{N}\}$ cont... | • Visual Prompt Learning: We integrate prior visual pathology knowledge into prompts at both patch and slide levels, guiding the model to generate meaningful activations for key pathology patterns, even with limited training samples.
• Textual Prompt Learning: Textual prompts are modeled with simple language descriptio... |
Generate 3 bullet points for the "Experimental Results / Performance Analysis" section of an academic poster. | ## Paper Content
# Paper Title
LSK3DNet: Towards Effective and Efficient 3D Perception with Large Sparse Kernels
## Abstract
Autonomous systems need to process large-scale, sparse, and irregular point clouds with limited compute resources. Consequently, it is essential to develop LiDAR perception methods that are b... | • Table compares Dense and Ours (1.8×D) kernels across Kernel Size, mIoU, Param, FLOPs, and Speed.
• LSK3DNet achieves 70.2 mIoU with 28.8 Params and 763.6 FLOPs, outperforming Dense variants.
• Scatter plot shows LSK3DNet outperforming SphereFormer, 2DPASS, RPVNet, and others on mIoU vs FPS trade-off. |
Generate 4 bullet points for the "Research Motivation / Problem Background" section of an academic poster. | ## Paper Content
# Paper Title
Atlantis: Enabling Underwater Depth Estimation with Stable Diffusion
## Abstract
Monocular depth estimation has experienced significant progress on terrestrial images in recent years thanks to deep learning advancements. But it remains inadequate for underwater scenes primarily due to... | • Obtaining precise terrestrial depth is easy
• High-quality terrestrial training data is enormous
• We have Stable Diffusion & ControlNet now
• Given the depth, realistic underwater imagery can be generated, paired, diverse, unlimited amount |
Generate 4 bullet points for the "Core Method / Technical Approach" section of an academic poster. | ## Paper Content
# Paper Title
Enhancing 3D Gaze Estimation in the Wild using Weak Supervision with Gaze Following Labels
## Abstract
Accurate 3D gaze estimation in unconstrained real-world environments remains a significant challenge due to variations in appearance, head pose, occlusion, and the limited availabili... | • ST-WSGE: a Self-Training Weakly Supervised Gaze Estimation framework using both model prediction and weak label
• 1. Supervised Training: GaT is composed of Encoder (image/video processing) and Decoder (shared MLP on each frame); GaT is pretrained on available 3D Gaze data
• 2. Pseudo labels: Use pretrained GaT on ga... |
Generate 3 bullet points for the "Background / Related Work" section of an academic poster. | ## Paper Content
# Paper Title
Dr.Hair: Reconstructing Scalp-Connected Hair Strands without Pre-Training via Differentiable Rendering of Line Segments
## Abstract
In the film and gaming industries, achieving a realistic hair appearance typically involves the use of strands originating from the scalp. However, recon... | • Problem: Hair strand reconstruction from multi-view images
• Learning-based methods require pre-training with CG data, slow runtime
• Optimization-based methods have no internal hair, 180° ambiguity |
Generate 3 bullet points for the "Experimental Results / Performance Analysis" section of an academic poster. | ## Paper Content
# Paper Title
CoMapGS: Covisibility Map-based Gaussian Splating for Sparse Novel View Synthesis
## Abstract
We propose Covisibility Map-based Gaussian Splatting (CoMapGS), designed to recover underrepresented sparse regions in sparse novel view synthesis. CoMapGS addresses both high- and low-uncert... | • Quantitative results on LLFF and Mip-NeRF360 datasets with 3, 6, 9, 12, and 24 training views.
• CoMapGS outperforms baselines (FSGS, CoR-GS) in PSNR, SSIM, and LPIPS metrics.
• Visual comparisons show improved rendering quality, especially in sparse-view settings. |
Generate 2 bullet points for the "Other Content" section of an academic poster. | ## Paper Content
# Paper Title
XCube: Large-Scale 3D Generative Modeling using Sparse Voxel Hierarchies
## Abstract
We present XCube (abbreviated as $\mathcal{X}^3$ ), a novel generative model for high-resolution sparse 3D voxel grids with arbitrary attributes. Our model can generate millions of voxels with a fine... | • For more information, visit the project website and check out the paper.
• Code is available online. |
Generate 3 bullet points for the "Ablation Study" section of an academic poster. | ## Paper Content
# Paper Title
SegEarth-OV: Towards Training-Free Open-Vocabulary Segmentation for Remote Sensing Images
## Abstract
Current remote sensing semantic segmentation methods are mostly built on the close-set assumption, meaning that the model can only recognize pre-defined categories that exist in the t... | • Detailed ablation results for each component: Evaluates contributions of baseline, FeatUp, +ours, +RS Data, +Rec. Image, +Alleviate Global Bias, and +Large Kernel on OpenEarthMap and WBS-SI.
• The proposed method results as a plug-and-play module: Shows performance when integrated with MaskCLIP, SCLIP, ClearCLIP, and... |
Generate 2 bullet points for the "Experimental Results / Performance Analysis" section of an academic poster. | ## Paper Content
# Paper Title
Uncertainty-guided Perturbation for Image Super-Resolution Diffusion Model
## Abstract
Diffusion-based image super-resolution methods have demonstrated significant advantages over GAN-based approaches, particularly in terms of perceptual quality. Building upon a lengthy Markov chain, ... | • Quantitative results on ImageNet-Test and RealSR datasets comparing GAN-based and diffusion-based methods across PSNR, SSIM, LPIPS, CLIP, MUSIQ, MANIQ, NIQE metrics.
• Our method achieves top scores in multiple metrics, e.g., 65.583 PSNR on RealSR with CLIPQA+. |
Generate 3 bullet points for the "Core Method / Technical Approach" section of an academic poster. | ## Paper Content
# Paper Title
High-Fidelity Lightweight Mesh Reconstruction from Point Clouds
## Abstract
Recently, learning signed distance functions (SDFs) from point clouds has become popular for reconstruction. To ensure accuracy, most methods require using high-resolution Marching Cubes for surface extraction... | • Component 1: SDF Learning uses hybrid 2D/3D features for detailed implicit surface modeling.
• Component 2: Vertex Generator uses point representation for curvature-adaptive vertex distribution.
• Component 3: Delaunay Meshing achieves robust tetrahedral classification for accurate meshes. |
Generate 2 bullet points for the "Core Method / Technical Approach" section of an academic poster. | ## Paper Content
# Paper Title
Insightful Instance Features for 3D Instance Segmentation
## Abstract
Recent 3D Instance Segmentation methods typically encode hundreds of instance-wise candidates with instancespecific information in various ways and refine them into final masks. However, they have yet to fully explo... | • Total loss combines BCE, dice, cls, KA, and SG losses: L_total = λ_BCE L_BCE + λ_dice L_dice + λ_cls L_cls + λ_KA L_KA + λ_SG L_SG.
• L_SG increases correlations for similar instances and decreases for unrelated ones. |
Generate 4 bullet points for the "Method Overview / Framework" section of an academic poster. | ## Paper Content
# Paper Title
GraphI2P: Image-to-Point Cloud Registration with Exploring Pattern of Correspondence via Graph Learning
## Abstract
Although the fusion of images and LiDAR point clouds is crucial to many applications in computer vision, the relative poses of cameras and LiDAR scanners are often unkno... | • Introduces virtual point cloud as a bridge to alleviate cross-modality gap between images and LiDAR point clouds.
• Proposes virtual-spherical representation for orthogonal decoupling between pixel location and predicted depth.
• Uses distribution-based adaptive sample module to generate unified distribution of point... |
Generate 2 bullet points for the "Method Overview / Framework" section of an academic poster. | ## Paper Content
# Paper Title
DetailSemNet: Elevating Signature Verification through Detail-Semantic Integration
## Introduction
Handwritten offline signature verification (OSV) is a pivotal biometric technology, especially in sectors like banking and commerce. The core goal of this technology is to authenticate a... | • Introduces DetailSemNet, a new model for offline signature verification (OSV).
• Aims to distinguish genuine vs. forged signatures by emphasizing fine-grained local differences. |
Generate 2 bullet points for the "Core Method / Technical Approach" section of an academic poster. | ## Paper Content
# Paper Title
CoSeR: Bridging Image and Language for Cognitive Super-Resolution
## Abstract
Existing super-resolution (SR) models primarily focus on restoring local texture details, often neglecting the global semantic information within the scene. This oversight can lead to the omission of crucial... | • Advocates comprehensive integration of all conditional info via All-in-Attention (AiA) module.
• Introduces 'one-hot attention' to enhance LR image with most relevant reference feature. |
Generate 3 bullet points for the "Background / Related Work" section of an academic poster. | ## Paper Content
# Paper Title
Implicit Regularization for Tubal Tensor Factorizations via Gradient Descent
## Abstract
We provide a rigorous analysis of implicit regularization in an overparametrized tensor factorization problem beyond the lazy training regime. For matrix factorization problems, this phenomenon ha... | • Model for tensors of low tubal rank r ≤ n: T := X * X^T for X ∈ R^{n×r×k}.
• Linear data y = A(X * X^T) s.t. y_i = <A_i, X * X^T> for i = 1,…,m and a tubal-symmetric tensor A_i.
• RIP(r,δ): Operator A: R^{n×n×k} → R^m with y = A(X * X^T) has Restricted Isometry Property of rank-r with constant δ > 0, if (1−δ)‖Z‖_F^2 ... |
Generate 3 bullet points for the "Method Overview / Framework" section of an academic poster. | ## Paper Content
# Paper Title
Named Entity Driven Zero-Shot Image Manipulation
## Abstract
We introduced StyleEntity, a zero-shot image manipulation model that utilizes named entities as proxies during its training phase. This strategy enables our model to manipulate images using unseen textual descriptions during... | • Introduces StyleEntity, a novel approach to image manipulation using named entities in CLIP text space.
• Finds that named entities can serve as compositional proxies for descriptive text components.
• Proposes Prompt Ensemble Latent Averaging (PELA) to improve generative quality and stability in zero-shot scenarios. |
Generate 3 bullet points for the "Core Method / Technical Approach" section of an academic poster. | ## Paper Content
# Paper Title
COBRA: COmBinatorial Retrieval Augmentation for Few-Shot Adaptation
## Abstract
Retrieval augmentation, the practice of retrieving additional data from large auxiliary pools, has emerged as an effective technique for enhancing model performance in the low-data regime. Prior approaches... | • Existing retrieval strategies are instances of nearest-neighbor based selection
• Given a small target dataset V^tar and auxiliary pool V^aux, they solve: g(A; V^tar, W) = sum_{j∈A} sum_{i∈V^tar} w_ij, A* = argmax_{A⊆V^aux, |A|≤k} g(A; V^tar, W)
• Nearest neighbor retrieval is an instance of CMI and prone to redundan... |
Generate 4 bullet points for the "Core Method / Technical Approach" section of an academic poster. | ## Paper Content
# Paper Title
From Sparse to Dense: Camera Relocalization with Scene-Specific Detector from Feature Gaussian Splatting
## Abstract
This paper presents a novel camera relocalization method, STDLoc, which leverages Feature Gaussian as scene representation. STDLoc is a full relocalization pipeline tha... | • Embeds off-the-shelf dense local features into 3DGS.
• Supports various dense local features, like SuperPoint, R2D2, etc.
• Compatible with different Gaussians, like 3DGS, 2DGS, etc.
• Theoretically adapt to emerging technologies in the field of 3DGS. |
Generate 4 bullet points for the "Core Method / Technical Approach" section of an academic poster. | ## Paper Content
# Paper Title
Adaptive Compressed Sensing with Diffusion-Based Posterior Sampling
## Method
We are interested in reconstructing a signal $\mathbf{x} \in \mathbb{R}^D$ from $d < D$ linear measurements of the form
$$
\mathbf {y} = \boldsymbol {H} \mathbf {x}, \tag {1}
$$
where $\pmb{H} \in \ma... | • Goal: progressively construct the most informative matrix H s.t. y = Hx.
• Based on partial existing measurements y_i, the best next measurement y_{i+1} corresponds to the eigenvectors of the posterior covariance.
• Equation (1) defines the selection of H_{i+1} via argmin over posterior covariance.
• Uses posterior s... |
Generate 3 bullet points for the "Method Overview / Framework" section of an academic poster. | ## Paper Content
# Paper Title
Grounded Question-Answering in Long Egocentric Videos
## Abstract
Existing approaches to video understanding, mainly designed for short videos from a third-person perspective, are limited in their applicability in certain fields, such as robotics. In this paper, we delve into open-end... | • Introduces the problem of grounded video question answering in long egocentric videos.
• Proposes GroundVQA: a unified model for temporal grounding and question answering.
• Introduces EgoTimeQA dataset and QAEGO4D-CLOSE benchmark to reduce ambiguity in open-ended answers. |
Generate 3 bullet points for the "Method Overview / Framework" section of an academic poster. | ## Paper Content
# Paper Title
BF-STVSR: B-Splines and Fourier—Best Friends for High Fidelity Spatial-Temporal Video Super-Resolution
## Abstract
While prior methods in Continuous Spatial-Temporal Video Super-Resolution (C-STVSR) employ Implicit Neural Representation (INR) for continuous encoding, they often strugg... | • Axis-Wise Signal Modeling: Fourier for space, B-Spline for time.
• Minimalist Design: No optical flow, fully learnable, robust to unseen scales.
• Best Quality & Speed: SOTA performance, faster inference, lower FLOPs. |
Generate 3 bullet points for the "Core Method / Technical Approach" section of an academic poster. | ## Paper Content
# Paper Title
Catch Your Emotion: Sharpening Emotion Perception in Multimodal Large Language Models
## Abstract
Multimodal large language models (MLLMs) have achieved impressive progress in tasks such as visual question answering and visual understanding, but they still face significant challenges ... | • Two-stage inference: Coarse-grained then Fine-grained Query.
• Confidence Assessment computes confidence score to mitigate errors.
• Focus-on-Emotion Visual Augmentation estimates token significance and augments visual tokens. |
Generate 3 bullet points for the "Research Motivation / Problem Background" section of an academic poster. | ## Paper Content
# Paper Title
PosterLlama: Bridging Design Ability of Language Model to Content-Aware Layout Generation
## Introduction
Layout is a fundamental element in graphic design, harmoniously arranging design elements such as logos and text to capture the reader's attention and effectively convey essential... | • Current content-aware layout generation models struggle to produce well-aligned, high-quality layouts by treating layout generation as numerical optimization.
• We leverage visual-textual content-aware layout generation via two-stage training and HTML sequences.
• Address poster-layout data scarcity by introducing ne... |
Generate 3 bullet points for the "Implementation Details" section of an academic poster. | ## Paper Content
# Paper Title
One-shot 3D Object Canonicalization based on Geometric and Semantic Consistency
## Abstract
3D object canonicalization is a fundamental task, essential for various downstream tasks. Existing methods rely on either cumbersome manual processes or priors learned from extensive, per-categ... | • Based on the proposed approach we contribute a new 3D canonical object dataset, namely COD dataset.
• Data: Our dataset is derived from cleaning and aligning the largest existing category dataset, Objaverse-Ivis.
• We are preparing more data for release. |
Generate 2 bullet points for the "Core Method / Technical Approach" section of an academic poster. | ## Paper Content
# Paper Title
Data Augmentation via Latent Diffusion for Saliency Prediction
## Method
Our approach is depicted in Figure 2. We incorporate the encoder $(\mathcal{E})$ , the decoder $(\mathcal{D})$ , and the denoising U-Net from Stable Diffusion v1.5 [50], all of which are frozen in our framework... | • Our data augmentation method is suitable for saliency prediction, consistently improving SoTA models.
• Leveraging augmentation features for saliency prediction yields a better understanding of visual attention patterns. |
Generate 3 bullet points for the "Experimental Results / Performance Analysis" section of an academic poster. | ## Paper Content
# Paper Title
On the Out-Of-Distribution Generalization of Large Multimodal Models
## Abstract
We investigate the generalization boundaries of current Large Multimodal Models (LMMs) under out-of-distribution scenarios and domain-specific tasks. We evaluate their zero-shot generalization across synt... | • Evaluates ICE performance under domain shifts using GPT-4 and Gemini.
• Bar charts show accuracy improvements across datasets (iWildCam, CT-Xcov, HAM10000) with different ICE configurations.
• Demonstrates consistent gains with ICE, especially under oracle and biased settings. |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.