instruction stringclasses 95
values | input stringlengths 141 265k | output stringlengths 15 1.61k |
|---|---|---|
Generate 2 bullet points for the "Research Motivation / Problem Background" section of an academic poster. | ## Paper Content
# Paper Title
Causal Composition Diffusion Model for Closed-loop Traffic Generation
## Abstract
Simulation is critical for safety evaluation in autonomous driving, particularly in capturing complex interactive behaviors. However, generating realistic and controllable traffic scenarios in long-tail ... | • Goal: control traffic generation to create challenging, safety-critical scenarios that remain realistic.
• Approach: add spatial reasoning to diffusion-based generation. |
Generate 2 bullet points for the "Ablation Study" section of an academic poster. | ## Paper Content
# Paper Title
Spherical Manifold Guided Diffusion Model for Panoramic Image Generation
## Abstract
Panoramic image essentially acts as a pivotal role in emerging virtual reality and augmented reality scenarios; however, the generation of panoramic images are essentially challenging due to the intri... | • Results demonstrate that SMConv exhibits superior performance compared to directly using existing spherical convolution.
• Performance degradation occurs when not using SMConv, validating its contribution to the framework. |
Generate 3 bullet points for the "Experimental Results / Performance Analysis" section of an academic poster. | ## Paper Content
# Paper Title
KITRO: Refining Human Mesh by 2D Clues and Kinematic-tree Rotation
## Abstract
2D keypoints are commonly used as an additional cue to refine estimated 3D human meshes. Current methods optimize the pose and shape parameters with a reprojection loss on the provided 2D keypoints. Such an... | • Significant improvements over optimization-based methods: 15% improvement in PA-MPJPE, 20% in MPJPE.
• Refinement on Top of Refinement: Can gain extra improvement.
• Factors Impact: Pose > Shape > Camera. |
Generate 4 bullet points for the "Core Method / Technical Approach" section of an academic poster. | ## Paper Content
# Paper Title
$V_{k}D:$ Improving Knowledge Distillation using Orthogonal Projections
## Abstract
Knowledge distillation is an effective method for training small and efficient deep learning models. However, the efficacy of a single method can degenerate when transferring to other tasks, modalitie... | • A learnable projection layer is needed to match feature dimensions.
• Goal is to mitigate possibility of learning new data representation not shared by feature extractor.
• Propose projection preserving pairwise similarity of features using kernel function.
• Constraint defines set of matrices with orthonormal rows; ... |
Generate 1 bullet points for the "Experimental Results / Performance Analysis" section of an academic poster. | ## Paper Content
# Paper Title
Dynamic Content Prediction with Motion-aware Priors for Blind Face Video Restoration
## Abstract
Blind Face Video Restoration (BFVR) focuses on reconstructing high-quality facial image sequences from degraded video inputs. The main challenge is address unknown degradations, while main... | • Visual comparison of feature heatmaps from low-quality/high-quality features and corrected features. |
Generate 2 bullet points for the "Core Method / Technical Approach" section of an academic poster. | ## Paper Content
# Paper Title
Shape and Texture: What Influences Reliable Optical Flow Estimation?
## Abstract
Recent methods have made significant progress in optical flow estimation. However, the evaluation of these methods focuses mainly on improved accuracy in benchmarks and often overlooks the analysis of net... | • Given data from a benchmark, change the texture and shape to generate new data (Flow-R-T, Flow-R-S, Flow-R-O).
• Use object extraction and manipulation to create controlled perturbations. |
Generate 3 bullet points for the "Core Method / Technical Approach" section of an academic poster. | ## Paper Content
# Paper Title
LC-Mamba: Local and Continuous Mamba with Shifted Windows for Frame Interpolation
## Abstract
In this paper, we propose LC-Mamba, a Mamba-based model that captures fine-grained spatiotemporal information in video frames, addressing limitations in current interpolation methods and enha... | • Captures complex spatiotemporal correlations between frames.
• Uses interleaved H-SS2D scanning for relationship modeling.
• Includes rearrange, selective scan, and revert/deinterleaving steps. |
Generate 7 bullet points for the "Core Method / Technical Approach" section of an academic poster. | ## Paper Content
# Paper Title
RICA $^2$ : Rubric-Informed, Calibrated Assessment of Actions
## Method
Steps and rubric as graph. We encode action steps and the corresponding scoring rubric with a directed acyclic graph (DAG). This DAG is denoted as $\mathcal{G} = (\mathbb{V},\mathbb{E})$ with $\mathbb{V}$ as t... | • Inputs: Video, Step Descriptions, Scoring Rubric
• Step 1: Extract embeddings from step descriptions using LLM
• Step 2: Extract I3D features from videos
• Step 3: Encode text/video features into stochastic embedding using x-attn
• Step 4: Decode score distribution from stochastic embedding using a DAG representing s... |
Generate 2 bullet points for the "Qualitative Results / Visualization" section of an academic poster. | ## Paper Content
# Paper Title
Denoising Functional Maps: Diffusion Models for Shape Correspondence
## Abstract
Estimating correspondences between pairs of deformable shapes remains a challenging problem. Despite substantial progress, existing methods lack broad generalization capabilities and require category-spec... | • Training data: 64,000 animal shapes (SMAL).
• Our method is applicable to various shape classes. |
Generate 2 bullet points for the "Conclusion / Future Work" section of an academic poster. | ## Paper Content
# Paper Title
OpenHumanVid: A Large-Scale High-Quality Dataset for Enhancing Human-Centric Video Generation
## Abstract
Recent advancements in visual generation technologies have markedly increased the scale and availability of video datasets, which are crucial for training effective video generati... | • Current limitations: lack of caption diversity and human action representation across cultures.
• Future work will focus on enhancing caption generation models and expanding the dataset. |
Generate 3 bullet points for the "Experimental Results / Performance Analysis" section of an academic poster. | ## Paper Content
# Paper Title
MaxQ: Multi-Axis Query for N:M Sparsity Network
## Abstract
N:M sparsity has received increasing attention due to its remarkable performance and latency trade-off compared with structured and unstructured sparsity. However, existing N:M sparsity methods do not differentiate the relati... | • Shows parameter distribution from SR-STE and MaxQ at inference.
• Highlights 8bit quantization performance: 77.1% to 77.6%.
• Notes that MaxQ is friendly to quantization with consistent kernel value distributions. |
Generate 3 bullet points for the "Core Method / Technical Approach" section of an academic poster. | ## Paper Content
# Paper Title
TFMQ-DM: Temporal Feature Maintenance Quantization for Diffusion Models
## Abstract
The Diffusion model, a prevalent framework for image generation, encounters significant challenges in terms of broad applicability due to its extended inference times and substantial memory requirement... | • Introduces Temporal Information Block to preserve temporal consistency.
• Proposes Temporal Information Aware Reconstruction loss to align features.
• Implements Finite Set Calibration to map quantized values to optimal ranges. |
Generate 3 bullet points for the "Method Overview / Framework" section of an academic poster. | ## Paper Content
# Paper Title
MoFlow: One-Step Flow Matching for Human Trajectory Forecasting via Implicit Maximum Likelihood Estimation based Distillation
## Abstract
In this paper, we address the problem of human trajectory forecasting, which aims to predict the inherently multimodal future movements of humans b... | • Present a novel Motion prediction Flow matching model for trajectory prediction, with a novel flow matching loss encouraging diverse, socially/physically compliant trajectories.
• Propose a one-step distillation method for flow models based on IMLE, accelerating inference 100x without sacrificing quality.
• MoFlow an... |
Generate 2 bullet points for the "Experimental Results / Performance Analysis" section of an academic poster. | ## Paper Content
# Paper Title
RCBEVDet: Radar-camera Fusion in Bird's Eye View for 3D Object Detection
## Abstract
Three-dimensional object detection is one of the key tasks in autonomous driving. To reduce costs in practice, low-cost multi-view cameras for 3D object detection are proposed to replace the expansive... | • On nuScenes test set, RCBEVDet (Ours) achieves 63.9 NDS and 55.0 mAP, outperforming all listed radar-camera fusion methods.
• Demonstrates superior accuracy while maintaining real-time inference speed. |
Generate 2 bullet points for the "Conclusion / Future Work" section of an academic poster. | ## Paper Content
# Paper Title
DragAnything: Motion Control for Anything using Entity Representation
## Conclusion
In this paper, we reevaluate the current trajectory-based video generation task and introduce two new insights: 1) Trajectory points on objects cannot adequately represent the entity. 2) For the trajec... | • DragAnything achieved superior performance in terms of motion control and video quality.
• Demonstrates effectiveness of entity representation for precise motion control in video generation. |
Generate 2 bullet points for the "Research Motivation / Problem Background" section of an academic poster. | ## Paper Content
# Paper Title
Forecasting Future Videos from Novel Views via Disentangled 3D Scene Representation
## Introduction
The Video Extrapolation in Space and Time (VEST) task, as introduced in [55], combines the tasks of novel view synthesis and video prediction. VEST involves extrapolating scenes to nove... | • Video extrapolation in space and time (VEST) enables viewers to forecast a 3D scene into the future and view it from novel viewpoints.
• Our approach disentangles scene geometry from motion by lifting 2D scenes to 3D point clouds, enabling high-quality future video rendering from novel views. |
Generate 3 bullet points for the "Method Overview / Framework" section of an academic poster. | ## Paper Content
# Paper Title
Continual Learning for Remote Physiological Measurement: Minimize Forgetting and Simplify Inference
## Introduction
The measurement of physiological signals, heart rate (HR) and heart rate variability (HRV) for instance, holds significant importance across various fields, such as medi... | • Adapter Tuning: Adapter are leveraged to efficiently adapt the backbone to new tasks and enhance its stability.
• Prototype-based Augmentation: Style and noise prototypes are designed to capture factors related to domain shift and then employed to augment training samples to consolidate previous knowledge.
• Prototyp... |
Generate 3 bullet points for the "Qualitative Results / Visualization" section of an academic poster. | ## Paper Content
# Paper Title
Mitigating Hallucinations in Large Vision-Language Models via DPO: On-Policy Data Hold the Key
## Abstract
Hallucination remains a major challenge for Large Vision-Language Models (LVLMs). Direct Preference Optimization (DPO) has gained increasing attention as a simple solution to hal... | • Qualitative comparison of LLaVA-Instruct-13B outputs with and without OPA and +OPA-DPO.
• Shows corrected responses for image description tasks, highlighting how OPA-DPO reduces hallucinations.
• Includes visual examples of generated text responses for the same image prompt under different training conditions. |
Generate 4 bullet points for the "Core Method / Technical Approach" section of an academic poster. | ## Paper Content
# Paper Title
VidLA: Video-Language Alignment at Scale
## Abstract
In this paper, we propose VidLA, an approach for video-language alignment at scale. There are two major limitations of previous video-language alignment approaches. First, they do not capture both short-range and long-range temporal... | • Uses MLLM for caption generation and LLM for text summarization
• Combines text encoder and video encoder with hierarchical temporal attention
• Employs spatially-local and global spatio-temporal attention mechanisms
• Integrates [CLS] tokens, [MST] tokens, and patch tokens from different frames |
Generate 3 bullet points for the "Research Motivation / Problem Background" section of an academic poster. | ## Paper Content
# Paper Title
MITIGATING MEMORIZATION IN LANGUAGE MODELS
## Abstract
Language models (LMs) can "memorize" information, i.e., encode training data in their weights in such a way that inference-time queries can lead to verbatim regurgitation of that data. This ability to extract training data can be ... | • Language models exhibit memorization when they reproduce exact training data during inference.
• Example: if prompt[50:100] == LM(prompt[0:50])[50:100], it indicates memorization.
• Otherwise, no memorization occurs. |
Generate 3 bullet points for the "Ablation Study" section of an academic poster. | ## Paper Content
# Paper Title
GET: Unlocking the Multi-modal Potential of CLIP for Generalized Category Discovery
## Abstract
Given unlabelled datasets containing both old and new categories, generalized category discovery (GCD) aims to accurately discover new classes while correctly classifying old classes. Curre... | • Ablation study shows the contribution of TES and CICO components to accuracy.
• Two line charts show accuracy improvement with TES and CICO on Stanford Cars dataset.
• Table compares different pseudo text embeddings (Baseline, VQA, Captioning, Feature-Generation) with GET (ours) achieving highest performance. |
Generate 3 bullet points for the "Qualitative Results / Visualization" section of an academic poster. | ## Paper Content
# Paper Title
SCPNet: Unsupervised Cross-modal Homography Estimation via Intra-modal Self-supervised Learning
## Experiments
We first denote the image pair from modality A and B as $\mathbf{I}_{\mathrm{A}}$ and $\mathbf{I}_{\mathrm{B}}$ , with homography deformation between them. To train a homo... | • Visualization shows how correlation facilitates consistent feature map generation.
• Input images are shown alongside concatenated and correlated feature maps.
• Correlation enforces inner product constraint, improving feature alignment across modalities. |
Generate 4 bullet points for the "Experimental Results / Performance Analysis" section of an academic poster. | ## Paper Content
# Paper Title
SPARSE COMPONENTS DistinguISH VISUAL PATHWAYS & THEIR ALIGNMENT TO NEURAL NETWORKS
## Abstract
The ventral, dorsal, and lateral streams in high-level human visual cortex are implicated in distinct functional processes. Yet, deep neural networks (DNNs) trained on a single task model th... | • Sparse component alignment respects a system’s native axes of neural tuning.
• The result is increased sensitivity to rotations in the neural space.
• Standard methods like RSA are less sensitive to rotations, resulting in similar alignment across the three visual pathways.
• In contrast, SCA reveals markedly higher ... |
Generate 3 bullet points for the "Experimental Results / Performance Analysis" section of an academic poster. | ## Paper Content
# Paper Title
AccDiffusion: An Accurate Method for Higher-Resolution Image Generation
## Experiments
AccDiffusion is a plug-and-play extension to stable diffusion without additional training costs. We mainly validate the feasibility of AccDiffusion using the pretrained SDXL [17]. More results for o... | • AccDiffusion achieves state-of-the-art performance across multiple resolutions and metrics.
• Outperforms baselines including SDXL-DI, Attn-SF, MultiDiffusion, and ScaleCrafter.
• Metrics include FID, IS, FID↑, IS↑, CLIP↑, and Time (minutes). |
Generate 4 bullet points for the "Method Overview / Framework" section of an academic poster. | ## Paper Content
# Paper Title
Towards Transformer-Based Aligned Generation with Self-Coherence Guidance
## Abstract
We introduce a novel, training-free approach for enhancing alignment in Transformer-based Text-Guided Diffusion Models (TGDMs). Existing TGDMs often struggle to generate semantically aligned images, ... | • We introduce a novel, training-free approach for enhancing alignment in Transformer-based Text-Guided Diffusion Models (TGDMs).
• Existing TGDMs struggle with semantically aligned images, especially for complex prompts or multi-concept attribute binding.
• Our method introduces Self-Coherence Guidance, which dynamica... |
Generate 3 bullet points for the "Ablation Study" section of an academic poster. | ## Paper Content
# Paper Title
Test-Time Model Adaptation for Image Reconstruction Using Self-supervised Adaptive Layers
## Experiments
We evaluate the proposed framework on two medical image reconstruction tasks: CS-MRI and SV-CT IR. The AdaptNet is tested on the PGDA-based unrolling NN model, same as [29]. See th... | • Compares AdaptNet with Full-tuning, Bias-tuning, Gain-tuning.
• AdaptNet uses 0.24% of parameters and achieves 33.01 PSNR at r=1/4.
• Full-tuning uses 100% parameters but only achieves 33.15 PSNR at r=1/4. |
Generate 3 bullet points for the "Experimental Results / Performance Analysis" section of an academic poster. | ## Paper Content
# Paper Title
Learning Intra-view and Cross-view Geometric Knowledge for Stereo Matching
## Abstract
Geometric knowledge has been shown to be beneficial for the stereo matching task. However, prior attempts to integrate geometric insights into stereo matching algorithms have largely focused on geom... | • Presents in-domain results on KITTI 2012 and KITTI 2015 benchmarks, showing performance across multiple metrics.
• Includes cross-domain results on Middlebury and KITTI datasets, comparing visual disparity outputs of IGEV-Stereo and ICGNet (ours).
• Our method achieves top performance on several metrics, e.g., 2.55 D... |
Generate 3 bullet points for the "Core Method / Technical Approach" section of an academic poster. | ## Paper Content
# Paper Title
Apply Hierarchical-Chain-of-Generation to Complex Attributes Text-to-3D Generation
## Abstract
Recent text-to-3D generation models have demonstrated remarkable abilities in producing high-quality 3D assets. Despite their great advancements, current models struggle to generate satisfyi... | • Hierarchical Blocks (In-to-Out Order): Block 1: Yellow shirt, blue shoes, pink trousers; Block 2: Black coat, green hat
• Initial Input Texts (Coarse): 1. A man in a shirt, shoes and trousers is waving. 2. A man wearing a coat and a hat is waving.
• Input Texts with Attributes (Precise): black coat, yellow shirt, pin... |
Generate 2 bullet points for the "Research Motivation / Problem Background" section of an academic poster. | ## Paper Content
# Paper Title
ADU: Adaptive Detection of Unknown Categories in Black-Box Domain Adaptation
## Abstract
Black-box Domain Adaptation (BDA) utilizes a black-box predictor of the source domain to label target domain data, addressing privacy concerns in Unsupervised Domain Adaptation (UDA). However, BDA... | • Black-box Domain Adaptation (BDA) enables knowledge transfer from a source model to an unlabeled target domain without accessing source data or model parameters, addressing privacy and security.
• Existing methods assume identical label sets and rely on noisy pseudo-labels and fixed thresholds, limiting adaptability ... |
Generate 4 bullet points for the "Core Method / Technical Approach" section of an academic poster. | ## Paper Content
# Paper Title
MemoNav: Working Memory Model for Visual Navigation
## Abstract
Image-goal navigation is a challenging task that requires an agent to navigate to a goal indicated by an image in unfamiliar environments. Existing methods utilizing diverse scene memories suffer from inefficient explorat... | • Agent processes current observation and goal image through encoders.
• Updates map and selectively forgets nodes in short-term memory (STM).
• Generates working memory from retained STM and long-term memory (LTM).
• Policy network outputs actions based on encoded memory states. |
Generate 4 bullet points for the "Method Overview / Framework" section of an academic poster. | ## Paper Content
# Paper Title
The All-Seeing Project V2: Towards General Relation Comprehension of the Open World
## Introduction
The study of artificial general intelligence (AGI) systems that can match human intelligence and excel in any task across domains represents the ultimate goal in the field of artificial... | • We propose a novel task, termed Relation Conversation, and the corresponding formulation method, unifying the modeling of captioning, grounding, and relation tasks flexibly. Based on the task and formulation, we constructed the AS-V2 dataset.
• Combining AS-V2 with other general multimodal corpora for training, we pr... |
Generate 4 bullet points for the "Core Method / Technical Approach" section of an academic poster. | ## Paper Content
# Paper Title
Extend Your Own Correspondences: Unsupervised Distant Point Cloud Registration by Progressive Distance Extension
## Abstract
Registration of point clouds collected from a pair of distant vehicles provides a comprehensive and accurate 3D view of the driving scenario, which is vital for... | • Overview: EYOC is a fully unsupervised outdoor distant PCR method using EMA-synced student-labeler networks.
• Correspondence Filtering: Cuts off near correspondences ≤20m away according to near-far disparity.
• Speculative Registration: Uses SC2-PCR on the fly to assess best correspondence set and output high-fideli... |
Generate 2 bullet points for the "Qualitative Results / Visualization" section of an academic poster. | ## Paper Content
# Paper Title
MatSynth: A Modern PBR Materials Dataset
## Abstract
We introduce MatSynth, a dataset of 4,000+ CCO ultra-high resolution PBR materials. Materials are crucial components of virtual relightable assets, defining the interaction of light at the surface of geometries. Given their importan... | • Comparison on unconditional materials generation for MatFuse trained on Deschaintre et al. 2018 and on MatSynth.
• Shows improved texture diversity and realism with MatSynth training. |
Generate 3 bullet points for the "Experimental Results / Performance Analysis" section of an academic poster. | ## Paper Content
# Paper Title
Feature Learning beyond the Lazy-Rich Dichotomy: Insights from Representational Geometry
## Abstract
Integrating task-relevant information into neural representations is a fundamental ability of both biological and artificial intelligence systems. Recent theories have categorized lear... | • RNNs trained on cognitive tasks with structural constraints show geometric inductive biases.
• Low-rank initializations start with lower capacity but evolve to richer manifolds.
• Manifold geometry reveals biases induced by circuit structures. |
Generate 4 bullet points for the "Ablation Study" section of an academic poster. | ## Paper Content
# Paper Title
Zero-shot RGB-D Point Cloud Registration with Pre-trained Large Vision Model
## Abstract
This paper introduces ZeroMatch, a novel zero-shot RGB-D point cloud registration framework, aimed at achieving robust 3D matching on unseen data without any task-specific training. Our core idea ... | • Table 4 shows ablation studies on coupled-image input mode and coupled consistency attention on ScanNet dataset.
• ZeroMatch with Single Mode achieves 66.1% Accuracy↑, while with Coupled+ achieves 81.6% Accuracy↑, showing the benefit of coupled input.
• ZeroMatch w/o P-CA achieves 66.1% Accuracy↑, while with P-CA ach... |
Generate 3 bullet points for the "Core Method / Technical Approach" section of an academic poster. | ## Paper Content
# Paper Title
Accelerating LLM Inference with Lossless Speculative Decoding Algorithms for Heterogeneous Vocabularies
## Abstract
Accelerating the inference of large language models (LLMs) is a critical challenge in generative AI. Speculative decoding (SD) methods offer substantial efficiency gains... | • SLEM (String-Level Exact Matching) uses strings as common representation and employs exact string matching.
• Steps: k drafter forwards + untokenize to string S, tokenize S w.r.t. target, accept longest matched prefix + first unmatched token.
• Includes a heuristic against normalization rules: s ≠ untokenize(tokenize... |
Generate 3 bullet points for the "Conclusion / Future Work" section of an academic poster. | ## Paper Content
# Paper Title
Prompting Vision Foundation Models for Pathology Image Analysis
## Abstract
The rapid increase in cases of non-alcoholic fatty liver disease (NAFLD) in recent years has raised significant public concern. Accurately identifying tissue alteration regions is crucial for the diagnosis of ... | • A novel visual prompting method, focused on pathology-related features, is proposed.
• Two kinds of quantitative attributes (spatial and morphology attributes) are studied for generating prompts.
• Improved diagnosis accuracy and enhanced interpretability. |
Generate 4 bullet points for the "Conclusion / Future Work" section of an academic poster. | ## Paper Content
# Paper Title
MapEval: A Map-Based Evaluation of Geo-Spatial Reasoning in Foundation Models
## Abstract
Recent advancements in foundation models have improved autonomous tool usage and reasoning, but their capabilities in map-based reasoning remain underexplored. To address this, we introduce MapEv... | • No model exceeds 67% accuracy.
• API and Visual tasks show major performance gaps.
• Open-source models significantly underperform.
• Human-level reasoning remains a challenge. |
Generate 2 bullet points for the "Research Motivation / Problem Background" section of an academic poster. | ## Paper Content
# Paper Title
Learning Social Welfare Functions
## Abstract
Is it possible to understand or imitate a policy maker's rationale by looking at past decisions they made? We formalize this question as the problem of learning social welfare functions belonging to the well-studied family of power mean fu... | • COVID-19 pandemic: Decisions on opening/closing public places and masking guidelines.
• Food rescue: Allocation of donated food to recipient organizations by dispatchers. |
Generate 4 bullet points for the "Method Overview / Framework" section of an academic poster. | ## Paper Content
# Paper Title
MORPHODIFF: CELLULAR MORPHOLOGY PAINTING WITH DIFFUSION MODELS
## Abstract
Understanding cellular responses to external stimuli is critical for parsing biological mechanisms and advancing therapeutic development. High-content image-based assays provide a cost-effective approach to exa... | • Motivation: High-content image-based assays provide a cost-effective approach to examine cellular phenotypes induced by diverse interventions, offering valuable insights into cellular states.
• Method: We introduce MorphoDiff, a diffusion based generative pipeline to predict high-resolution cell morphological respons... |
Generate 4 bullet points for the "Core Method / Technical Approach" section of an academic poster. | ## Paper Content
# Paper Title
KeyFace: Expressive Audio-Driven Facial Animation for Long Sequences via KeyFrame Interpolation
## Abstract
Current audio-driven facial animation methods achieve impressive results for short videos but suffer from error accumulation and identity drift when extended to longer durations... | • Emotions: Valence and arousal go through resblocks with timestep.
• Audio: The audio input goes through resnet and attention layers.
• BEATs: Trained with multiple audio types including NSV.
• WavLM: Trained to perform very well on speech. |
Generate 3 bullet points for the "Experimental Results / Performance Analysis" section of an academic poster. | ## Paper Content
# Paper Title
DifFlow3D: Toward Robust Uncertainty-Aware Scene Flow Estimation with Iterative Diffusion-Based Refinement
## Abstract
Scene flow estimation, which aims to predict per-point 3D displacements of dynamic scenes, is a fundamental task in the computer vision field. However, previous works... | • First millimeter-level accuracy directly generalized on KITTI.
• SOTA results on both with and without occlusion.
• Improve estimation accuracy for several SOTA methods in a plug-and-play manner. |
Generate 4 bullet points for the "Core Method / Technical Approach" section of an academic poster. | ## Paper Content
# Paper Title
Your ViT is Secretly an Image Segmentation Model
## Abstract
Vision Transformers (ViTs) have shown remarkable performance and scalability across various computer vision tasks. To apply single-scale ViTs to image segmentation, existing methods adopt a convolutional adapter to generate ... | • Proposes EoMT: keeps only the ViT encoder and adds a small mask module, eliminating the decoder.
• Uses pre-trained Vision Transformer with linear patch embedding and encoder blocks.
• Mask module processes queries via Linear, MLP, and Upscale layers to produce class logits and mask logits.
• Training uses shared wei... |
Generate 1 bullet points for the "Core Method / Technical Approach" section of an academic poster. | ## Paper Content
# Paper Title
Beyond Prompt Learning: Continual Adapter for Efficient Rehearsal-Free Continual Learning
## Method
The continual learning aims to enable the model to learn non-stationary data from sequential tasks while preserving the knowledge obtained from previous tasks. We define a set of tasks ... | • L_or = ||W_dp^t^T * {W_dp^1, ..., W_dp^{t-1}}||_2 + ||W_up^t^T * {W_up^1, ..., W_up^{t-1}}||_2 |
Generate 5 bullet points for the "Research Motivation / Problem Background" section of an academic poster. | ## Paper Content
# Paper Title
Exploring Temporally-Aware Features for Point Tracking
## Abstract
Point tracking in videos is a fundamental task with applications in robotics, video editing, and more. While many vision tasks benefit from pre-trained feature backbones to improve generalizability, point tracking has ... | • Spatially: Models complex real-world data with robust representation for precise matching.
• Temporally: Captures complex motions and point dynamics over time.
• Both spatial and temporal features are crucial but under-explored.
• Previous works use simple backbones without temporal awareness, requiring heavy post-ho... |
Generate 4 bullet points for the "Core Method / Technical Approach" section of an academic poster. | ## Paper Content
# Paper Title
Physical 3D Adversarial Attacks against Monocular Depth Estimation in Autonomous Driving
## Abstract
Deep learning-based monocular depth estimation (MDE), extensively applied in autonomous driving, is known to be vulnerable to adversarial attacks. Previous physical attacks against MDE... | • Problem definition: Find texture seed t_s to minimize distance between estimated and target depth.
• Texture conversion: Convert seed to 3D camouflaged textures using RC and M_robust.
• Adversarial image rendering: Project 3D object with adversarial texture to 2D image via differentiable renderer R.
• Physical augmen... |
Generate 3 bullet points for the "Core Method / Technical Approach" section of an academic poster. | ## Paper Content
# Paper Title
DiET-GS: Diffusion Prior and Event Stream-Assisted Motion Deblurring 3D Gaussian Splatting
## Abstract
Reconstructing sharp 3D representations from blurry multi-view images is a long-standing problem in computer vision. Recent works attempt to enhance high-quality novel view synthesis... | • Leverages diffusion prior to enhance edge details of deblurred images, making them more natural.
• Key idea: Modify RSD Loss from DISR-NeRF to use deblurred image from Stage 1 as conditional input.
• Declares learnable feature to each 3D Gaussian in pretrained DIET-GS, encoding diffusion prior into 3DGS. |
Generate 4 bullet points for the "Method Overview / Framework" section of an academic poster. | ## Paper Content
# Paper Title
SCE-MAE: Selective Correspondence Enhancement with Masked Autoencoder for Self-Supervised Landmark Estimation
## Abstract
Self-supervised landmark estimation is a challenging task that demands the formation of locally distinct feature representations to identify sparse facial landmark... | • First to leverage Masked Autoencoder (MAE) for region-level SSL, better suited for landmark prediction.
• Use vanilla final feature map to avoid expensive hypercolumns.
• Introduce Correspondence Approximation and Refinement Block (CARB) to identify and approximate failures of unimportant regions, and operate a novel... |
Generate 3 bullet points for the "Experimental Results / Performance Analysis" section of an academic poster. | ## Paper Content
# Paper Title
CAT-SAM: Conditional Tuning for Few-Shot Adaptation of Segment Anything Model
## Experiments
We evaluate CAT-SAM's efficacy through comprehensive experiments involving eight segmentation tasks across 11 datasets, all hailing from challenging downstream fields that SAM struggles to add... | • Summarization of evaluation benchmark across multiple segmentation tasks
• Tasks include building, road, polyp, chest X-ray, marine debris, ship instance, shadow, and intricate segmentation
• Datasets span aerial, medical, X-ray, sonar, SAR, and natural images |
Generate 3 bullet points for the "Research Motivation / Problem Background" section of an academic poster. | ## Paper Content
# Paper Title
SCENETAP: Scene-Coherent Typographic Adversarial Planner against Vision-Language Models in Real-World Environments
## Abstract
Large vision-language models (LVLMs) have shown remarkable capabilities in interpreting visual content. While existing works demonstrate these models' vulnera... | • Rise of Large Vision-Language Models (LVLMs)
• Typographic attacks: Inserting adversarial text into images to mislead models
• Key limitations of existing typographic attacks: Predefined adversarial text, fixed placement strategies, not deployable in physical environments |
Generate 3 bullet points for the "Background / Related Work" section of an academic poster. | ## Paper Content
# Paper Title
DNGaussian: Optimizing Sparse-View 3D Gaussian Radiance Fields with Global-Local Depth Normalization
## Abstract
Radiance fields have demonstrated impressive performance in synthesizing novel views from sparse input views, yet prevailing methods suffer from high training costs and slo... | • Depth should target partial parameters, not the entire model, to resolve conflicts between coarse depth and fine color supervision.
• Fixed-scale depth loss is insufficient for geometry learning.
• DNGaussian solves these to achieve outstanding performance. |
Generate 3 bullet points for the "Research Motivation / Problem Background" section of an academic poster. | ## Paper Content
# Paper Title
DreamLIP: Language-Image Pre-training with Long Captions
## Introduction
Language-image pre-training largely relies on how precisely and thoroughly a text describes its paired image. Existing image text pairing datasets typically
Long Caption:
, which are usually missing in existing datasets.
• Each sentence within a long caption is very likely to describe the image partially (e.g., an object).
• We study the usage of the resulting captions ... |
Generate 2 bullet points for the "Qualitative Results / Visualization" section of an academic poster. | ## Paper Content
# Paper Title
Using Diffusion Priors for Video Amodal Segmentation
## Abstract
Object permanence in humans is a fundamental cue that helps in understanding persistence of objects, even when they are fully occluded in the scene. Present day methods in object segmentation do not account for this amod... | • Applications include 4D reconstruction and scene manipulation.
• Pseudo-ground truth generation enables training data augmentation for occluded object tasks. |
Generate 3 bullet points for the "Research Motivation / Problem Background" section of an academic poster. | ## Paper Content
# Paper Title
Uncertainty Visualization via Low-Dimensional Posterior Projections
## Abstract
In ill-posed inverse problems, it is commonly desirable to obtain insight into the full spectrum of plausible solutions, rather than extracting only a single reconstruction. Information about the plausible... | • Inverse problems are often ill-posed, requiring probabilistic posterior modeling.
• Forward process maps x to y; backward process estimates x from y.
• High-dimensional posterior distributions are complex and hard to visualize. |
Generate 3 bullet points for the "Core Method / Technical Approach" section of an academic poster. | ## Paper Content
# Paper Title
Probabilistic Image-Driven Traffic Modeling via Remote Sensing
## Method
Our segmentation architecture has two primary components: 1) a multi-stage visual encoder for extracting features from an input overhead image, and 2) task-specific decoders which use these features to generate a... | • Integrate geo-temporal context via a novel GTPE module operating on location, time, and space-time features.
• Use SIREN as foundation for encoding network, representing complex natural signals with periodic activation functions.
• Fuse GTPE output with visual features in transformer stages via stage-specific project... |
Generate 3 bullet points for the "Research Motivation / Problem Background" section of an academic poster. | ## Paper Content
# Paper Title
Learning Conditional Space-Time Prompt Distributions for Video Class-Incremental Learning
## Abstract
Recent advancements in prompt-based learning have significantly advanced image and video class-incremental learning. However, the prompts learned by these methods often fail to captur... | • Previous prompt-based Class-Incremental Learning (CIL) methods generate prompts deterministically, struggling with spatio-temporal complexity in videos.
• Challenges include difficulty generating diverse prompts for variable video content, overfitting to current tasks, and missing local patch relationships across fra... |
Generate 3 bullet points for the "Core Method / Technical Approach" section of an academic poster. | ## Paper Content
# Paper Title
CriSp: Leveraging Tread Depth Maps for Enhanced Crime-Scene Shoeprint Matching
## Method
Given an input shoeprint image (Fig. 4), our goal is to retrieve the most relevant shoe tread models from a reference database (Fig. 2). A method should retrieve a ranked list $[r_1, r_2, \dots, ... | • Our encoder (Enc) is a modified ResNet50 that allows access to different cue locations on depth maps and shoeprints.
• Training and validation data are globally aligned to enable matching of shoeprint patterns to corresponding locations on depth maps.
• Feature and query masking: a random mask is applied on simulated... |
Generate 3 bullet points for the "Method Overview / Framework" section of an academic poster. | ## Paper Content
## Abstract
Reliable embodied perception from an egocentric perspective is challenging yet essential for autonomous navigation technology of intelligent mobile agents. With the growing demand of social robotics, near-field scene understanding becomes an important research topic in the areas of egocen... | • Unique Collected Domain: Unstructured navigation routes
• Diverse Annotation: 3D box + Trajectory + Occupancy
• Social Crowded Scene: Closer obstacles, with a mode at 5m |
Generate 2 bullet points for the "Qualitative Results / Visualization" section of an academic poster. | ## Paper Content
# Paper Title
Attention Calibration for Disentangled Text-to-Image Personalization
## Abstract
Recent thrilling progress in large-scale text-to-image (T2I) models has unlocked unprecedented synthesis quality of AI-generated content (AIGC) including image generation, 3D and video composition. Furthe... | • Demonstrates compatibility with LoRA for inpainting tasks involving multiple concepts.
• Shows successful generation of complex scenes with three distinct concepts (e.g., chair, lamp, room). |
Generate 4 bullet points for the "Conclusion / Future Work" section of an academic poster. | ## Paper Content
# Paper Title
Power Variable Projection for Initialization-Free Large-Scale Bundle Adjustment
## Conclusion
We implement $pOSE^5$ [11], $PoVar$ and $RiPoBA$ framework in $C++$ , directly on the publicly available implementation of $\mathrm{PoBA}^6$ [23]. That leads to fair comparisons with... | • Scalable solver for initialization-free BA problem.
• Extension of recent inverse expansion techniques to the VarPro algorithm and to Riemannian manifold optimization.
• Significant improvements in speed and accuracy.
• Initialization-free BA problem remains largely uncharted. |
Generate 3 bullet points for the "Experimental Results / Performance Analysis" section of an academic poster. | ## Paper Content
# Paper Title
DreamLIP: Language-Image Pre-training with Long Captions
## Experiments
Pretraining Datasets. To make a fair comparison with the state-of-the-art contrastive vision-language pretraining approaches, we adopt the CC3M, CC12M and YFCC15M datasets. In addition, we construct a 30M version ... | • Evaluates performance on semantic segmentation using datasets like MSCOCO and Flickr30k.
• DreamLIP-30M improves segmentation accuracy compared to CLIP baselines.
• Visual examples show better object boundary detection and class labeling. |
Generate 4 bullet points for the "Core Method / Technical Approach" section of an academic poster. | ## Paper Content
# Paper Title
PatchRefiner: Leveraging Synthetic Data for Real-Domain High-Resolution Monocular Metric Depth Estimation
## Method
In this section, we first present the overall PatchRefiner framework in Sec. 3.1. Then, we introduce the limitation of adopting real-domain data for high-resolution dept... | • Reformulate depth estimation as a refinement process.
• Compare with standard depth model and PatchFusion.
• Framework includes Base Model, Refiner, and Decoder components.
• Use synthetic data to train Teacher Model for pseudo-labeling real-domain images. |
Generate 2 bullet points for the "Background / Related Work" section of an academic poster. | ## Paper Content
# Paper Title
The Photographer's Eye: Teaching Multimodal Large Language Models to See and Critique like Photographers
## Abstract
"While editing directly from life, photographers have found it too difficult to see simultaneously both the blue and the sky."
John Szarkowski, William Eggleston's Gui... | • Data Scale: Statistics of samples and comparison of description length distributions across PhotoCritique and Q-Instruct.
• Data Quality: PhotoCritique is sourced from images and discussions from photography enthusiasts and professionals. |
Generate 4 bullet points for the "Background / Related Work" section of an academic poster. | ## Paper Content
# Paper Title
Towards More Practical Group Activity Detection: A New Benchmark and Model
## Introduction
Understanding group activities in videos plays a crucial role in numerous applications such as visual surveillance, social scene understanding, and sports analytics. The generic task of group ac... | • Large-scale & rich annotations: 10K video clips, 3.5M bounding box annotations
• Multi-view: Four different viewpoints
• Densely populated multiple non-singleton groups
• Imbalanced class distribution |
Generate 1 bullet points for the "Experimental Results / Performance Analysis" section of an academic poster. | ## Paper Content
# Paper Title
DiffusionAvatars: Deferred Diffusion for High-fidelity 3D Head Avatars
## Abstract
DiffusionAvatars synthesizes a high-fidelity 3D head avatar of a person, offering intuitive control over both pose and expression. We propose a diffusion-based neural renderer that leverages generic 2D ... | • Self-reenactment results averaged over 8 persons |
Generate 2 bullet points for the "Qualitative Results / Visualization" section of an academic poster. | ## Paper Content
# Paper Title
Towards Multi-modal Transformers in Federated Learning
## Experiments
To address the cross-modality and in-modality gaps, a key goal is to unify the learned representation into a shared feature space [56]. Under centralized training, those gaps can be addressed by joint training in a ... | • Parametric Loss Landscape with Hessian Eigenvectors: Smoother loss plateau indicates better generalizability.
• Extracted Features for different modalities: Gaps between uni-modal and multi-modal datasets are reduced. |
Generate 2 bullet points for the "Qualitative Results / Visualization" section of an academic poster. | ## Paper Content
# Paper Title
Unsupervised Discovery of Facial Landmarks and Head Pose
## Abstract
Unsupervised landmark and head pose estimation is fundamental in fields like biometrics, augmented reality, and emotion recognition, offering accurate spatial data without relying on labeled datasets. It enhances sca... | • Shows qualitative landmark detection results on human faces (Mallis, Hedlin, Tourani, Ours) with color-coded landmarks
• Demonstrates cross-species applicability on animals (Alpaca, Donkey, Goat, Giraffe, Monkey, Leopard, Lemur, Ringtail, Seal, Wallaby) with detected landmarks |
Generate 3 bullet points for the "Core Method / Technical Approach" section of an academic poster. | ## Paper Content
# Paper Title
Mitigating Noisy Correspondence by Geometrical Structure Consistency Learning
## Abstract
Noisy correspondence that refers to mismatches in cross-modal data pairs, is prevalent on human-annotated or web-crawled datasets. Prior approaches to leverage such data mainly consider the appli... | • Framework of GSC includes image and text encoders generating representations.
• Structure consistency is enforced via cross-modal and intra-modal discrimination modules.
• True correspondence indicator refines structure consistency. |
Generate 3 bullet points for the "Ablation Study" section of an academic poster. | ## Paper Content
# Paper Title
Modeling Thousands of Human Annotators for Generalizable Text-to-Image Person Re-identification
## Abstract
Text-to-image person re-identification (ReID) aims to retrieve the images of an interested person based on textual descriptions. One main challenge for this task is the high cos... | • Ablation conducted on LLaVA1.6-7B and Qwen-VL-Chat-7B using CLIP-ViT-B/16 and 12-layer transformer.
• Evaluated on CUHK-PEDES, ICFG-PEDES, and RSTPReid datasets.
• Results show HAM+KMeans+UPS with 1000 clusters achieves best performance across metrics. |
Generate 3 bullet points for the "Qualitative Results / Visualization" section of an academic poster. | ## Paper Content
# Paper Title
VIDEOTREE: Adaptive Tree-based Video Representation for LLM Reasoning on Long Videos
## Abstract
Long-form video understanding is complicated by the high redundancy of video data and the abundance of query-irrelevant information. To tackle these challenges, we propose VIDEOTREE, a tra... | • Shows step-by-step visualization of VideoTree processing a video query.
• Demonstrates how the tree structure selects relevant frames and generates accurate answers.
• Includes example questions and corresponding reasoning paths through the tree. |
Generate 3 bullet points for the "Core Method / Technical Approach" section of an academic poster. | ## Paper Content
# Paper Title
Anomize: Better Open Vocabulary Video Anomaly Detection
## Abstract
Open Vocabulary Video Anomaly Detection (OVVAD) seeks to detect and classify both base and novel anomalies. However, existing methods face two specific challenges related to novel anomalies. The first challenge is det... | • Framework includes two stages: Categorization Branch and Detection Branch.
• Uses lightweight temporal encoder and augmenter-detector modules.
• Integrates label features, alignment, and video label prediction for anomaly scoring. |
Generate 2 bullet points for the "Core Method / Technical Approach" section of an academic poster. | ## Paper Content
# Paper Title
LLMs are Good Action Recognizers
## Abstract
Skeleton-based action recognition has attracted lots of research attention. Recently, to build an accurate skeleton-based action recognizer, a variety of works have been proposed. Among them, some works use large model architectures as back... | • Given an input action signal (skeleton sequence), we first perform linguistic projection to acquire an "action sentence".
• Then perform action recognition via the large language model leveraging its pre-learned rich knowledge. |
Generate 4 bullet points for the "Experimental Results / Performance Analysis" section of an academic poster. | ## Paper Content
# Paper Title
Unifying Automatic and Interactive Matting with Pretrained ViTs
## Abstract
Automatic and interactive matting largely improve image matting by respectively alleviating the need for auxiliary input and enabling object selection. Due to different settings on whether prompts exist, they ... | • Compatibility: Comparison of methods on AIM-500 and RefMatte-RW100 datasets with SAD, MSE, MAD, Grad, Conn metrics
• Efficiency: Parameters, speed, and FPS for different methods
• Effect of designs: Ablation study on binary generation, cross attention, and accurate FD
• Robustness to different classes: Performance ac... |
Generate 2 bullet points for the "Experimental Results / Performance Analysis" section of an academic poster. | ## Paper Content
## Abstract
Task automation has been greatly empowered by the recent advances in Large Language Models (LLMs) via Python code, where the tasks range from software engineering development to general-purpose reasoning. While current benchmarks have shown that LLMs can solve tasks using programs like hu... | • Compares BigCodeBench with existing benchmarks like APPS, DS-1000, ODEX, MBPP, etc.
• Shows BigCodeBench has highest complexity (426.0) and diversity (723) |
Generate 4 bullet points for the "Research Motivation / Problem Background" section of an academic poster. | ## Paper Content
# Paper Title
Image Reconstruction from Readout-Multiplexed Single-Photon Detector Arrays
## Abstract
Readout multiplexing is a promising solution to overcome hardware limitations and data bottlenecks in imaging with single-photon detectors. Conventional multiplexed readout processing creates an up... | • Readout multiplexing helps mitigate high data transfer rates and improve array size scaling.
• Methods like sketching and equidepth histogramming are incompatible with superconducting nanowire single-photon detectors (SNSPDs), which cannot incorporate on-chip digital logic.
• Row-column multiplexing has enabled kilop... |
Generate 4 bullet points for the "Core Method / Technical Approach" section of an academic poster. | ## Paper Content
# Paper Title
Flexible Biometrics Recognition: Bridging the Multimodality Gap through Attention, Alignment and Prompt Tuning
## Abstract
Periocular and face are complementary biometrics for identity management, albeit with inherent limitations, notably in scenarios involving occlusion due to sungla... | • MFA-ViT consists of multimodal embeddings, multimodal fusion attention (MFA) layer, and multimodal prompt tuning (MPT)
• Multimodal embeddings employ a feature tokenizer to integrate categorical soft-biometric attributes with image-based face and periocular modalities
• MFA aligns facial, periocular, and soft-biometr... |
Generate 3 bullet points for the "Background / Related Work" section of an academic poster. | ## Paper Content
# Paper Title
Contrastive ground-level image and remote sensing pre-training improves representation learning for natural world imagery
## Introduction
In the natural world, there is a vast amount of unlabeled data. Users upload millions of geo-tagged images of plants and animals to citizen science... | • To train CRISP, we introduce NMV, a novel dataset of >3 million ground-level image and aerial image pairs for >6,000 plant taxa across the state of California
• Dataset Strengths: Ecologically diverse: NMV covers 5/10 of the world’s major biome types; Taxonomically diverse: NMV species span over 500 million years of ... |
Generate 3 bullet points for the "Qualitative Results / Visualization" section of an academic poster. | ## Paper Content
# Paper Title
Customized Condition Controllable Generation for Video Soundtrack
## Abstract
Recent advances in latent diffusion models (LDMs) have enabled data-driven paradigms for video soundtrack generation, improving multimodal alignment capabilities. However, current two-stage frameworks—which ... | • Demonstrates generated soundtrack samples under different conditions: style, emotion, and rhythm.
• Shows visual comparison of generated audio waveforms and spectrograms.
• Includes side-by-side comparisons with target samples for evaluation. |
Generate 3 bullet points for the "Experimental Results / Performance Analysis" section of an academic poster. | ## Paper Content
# Paper Title
Mitigating Noisy Correspondence by Geometrical Structure Consistency Learning
## Abstract
Noisy correspondence that refers to mismatches in cross-modal data pairs, is prevalent on human-annotated or web-crawled datasets. Prior approaches to leverage such data mainly consider the appli... | • Comparison to SOTAs on simulated noisy correspondence datasets.
• GSC consistently outperforms SOTA across noise levels on Flickr30K, MS-COCO, and NUS-WIDE.
• On real-world CC152K and CLIP finetuned on MS-COCO, GSC achieves superior performance. |
Generate 2 bullet points for the "Background / Related Work" section of an academic poster. | ## Paper Content
# Paper Title
COSMOS: Cross-Modality Self-Distillation for Vision Language Pre-training
## Abstract
Vision-Language Models (VLMs) trained with contrastive loss have achieved significant advancements in various vision and language tasks. However, the global nature of the contrastive loss makes VLMs ... | • Compares COSMOS with CLIP, SLIP, and SILC using diagrams of their training objectives.
• Shows how COSMOS integrates cross-modality self-distillation with EMA and cross-attention. |
Generate 3 bullet points for the "Method Overview / Framework" section of an academic poster. | ## Paper Content
# Paper Title
STAR-Edge: Structure-aware Local Spherical Curve Representation for Thin-walled Edge Extraction from Unstructured Point Clouds
## Abstract
Extracting geometric edges from unstructured point clouds remains a significant challenge, particularly in thin-walled structures that are commonl... | • Proposed STAR-Edge for robust edge extraction on thin-walled structures with ambiguous local geometry.
• Introduced local spherical curves and SH descriptors to build structure-aware, rotation-invariant representations.
• Designed edge refinement via projection optimization for accurate and stable edge localization. |
Generate 2 bullet points for the "Core Method / Technical Approach" section of an academic poster. | ## Paper Content
# Paper Title
SVDQUANT: ABSORBING OUTLIERS BY LOW-RANK COMPONENTS FOR 4-BIT DIFFUSION MODELS
## Abstract
Diffusion models can effectively generate high-quality images. However, as they scale, rising memory demands and higher latency pose substantial deployment challenges. In this work, we aim to ac... | • Naively running low-rank branch of rank 32 introduces 57% latency overhead due to extra reads/writes in Down/Up Projection.
• Down Projection and Quantize kernels share input; Up Projection and 4-Bit Compute share output. Nunchaku fuses the first two and latter two kernels, achieving 1.43x speedup. |
Generate 3 bullet points for the "Experimental Results / Performance Analysis" section of an academic poster. | ## Paper Content
# Paper Title
Saliuitl: Ensemble Salience Guided Recovery of Adversarial Patches against CNNs
## Abstract
Adversarial patches are capable of misleading computer vision systems based on convolutional neural networks. Existing recovery methods suffer of at least one of three fundamental shortcomings:... | • Computational cost scales linearly with ensemble size.
• Small ensemble = low cost.
• Cost is independent of the number and shape of patches. |
Generate 2 bullet points for the "Experimental Results / Performance Analysis" section of an academic poster. | ## Paper Content
# Paper Title
Scene Map-based Prompt Tuning for Navigation Instruction Generation
## Abstract
Navigation instruction generation (NIG), which provides interactive feedback and guidance to humans along a trajectory, is vital for developing embodied agents capable of human-machine communication and co... | • Table 1 shows quantitative comparison results for NIG on R2R val seen and unseen splits, with MAPINSTRUCTOR (Ours) achieving 0.229 SPICE, 0.746 Bleu-1, 0.301 Bleu-4, 0.576 CIDEr, 0.243 Meteor, 0.497 Rouge.
• Table 2 shows results on REVERIE val seen and unseen, with MAPINSTRUCTOR (Ours) achieving 0.186 SPICE, 0.791 B... |
Generate 3 bullet points for the "Method Overview / Framework" section of an academic poster. | ## Paper Content
# Paper Title
"Where am I?" Scene Retrieval with Language
## Introduction
Localization is central to many important applications, such as: an embodied agent navigating and executing tasks in the real world. Traditional localization methods try to localize within maps such as point clouds or large d... | • We present Text2SceneGraphMatcher, a system that retrieves 3D scenes based on natural language descriptions by matching scene graphs.
• Unlike traditional localization methods that rely on visual or sensor-based data, our approach bridges the gap between language and spatial understanding.
• We match joint embeddings... |
Generate 4 bullet points for the "Conclusion / Future Work" section of an academic poster. | ## Paper Content
# Paper Title
A Versatile Framework for Continual Test-Time Domain Adaptation: Balancing Discriminability and Generalizability
## Abstract
Continual test-time domain adaptation (CTTA) aims to adapt the source pre-trained model to a continually changing target domain without additional data acquisit... | • This paper first proposes a versatile framework that generates high-quality supervision signals from two levels: reliability and diversity
• We calculate an independent threshold for each class through global and local strategies to divide pseudo-labels into reliable and unreliable parts
• We begin by tracking the re... |
Generate 2 bullet points for the "Ablation Study" section of an academic poster. | ## Paper Content
# Paper Title
Collaborative Vision-Text Representation Optimizing for Open-Vocabulary Segmentation
## Experiments
Feature Extraction. We utilize a pre-trained convolutional CLIP-V for extracting features from an input image $I$ . Denoting each stage of CLIP-V's output as $F = \{F^i\}$ , $i \in [... | • Evaluate impact of different components: frozen CLIP, fine-tune CLIP-T, MLP, GPT-Description, Llama-Description, and Content-Dependent Transfer.
• Content-Dependent Transfer consistently improves performance across all benchmarks. |
Generate 3 bullet points for the "Research Motivation / Problem Background" section of an academic poster. | ## Paper Content
# Paper Title
AHIVE: Anatomy-aware Hierarchical Vision Encoding for Interactive Radiology Report Retrieval
## Abstract
Automatic radiology report generation using deep learning models has been recently explored and found promising. Neural decoders are commonly used for the report generation, where ... | • How to improve the clinical accuracy of report retrieval?
• How to incorporate the clinician’s approach of examining X-ray images (anatomical and diagnostic details)?
• How to support user interaction for the retrieval? |
Generate 3 bullet points for the "Core Method / Technical Approach" section of an academic poster. | ## Paper Content
# Paper Title
ATP: Adaptive Threshold Pruning for Efficient Data Encoding in Quantum Neural Networks
## Abstract
Quantum Neural Networks (QNNs) offer promising capabilities for complex data tasks, but are often constrained by limited qubit resources and high entanglement, which can hinder scalabili... | • ATP filters irrelevant features before encoding into the quantum circuit, focusing resources on high-information regions.
• Uses L-BFGS-B algorithm to solve the threshold (T) optimization problem.
• Mathematical formulation defines thresholded pixel values based on a condition involving τ. |
Generate 2 bullet points for the "Core Method / Technical Approach" section of an academic poster. | ## Paper Content
# Paper Title
Ges3ViG: Incorporating Pointing Gestures into Language-Based 3D Visual Grounding for Embodied Reference Understanding
## Abstract
3-Dimensional Embodied Reference Understanding (3D-ERU) combines a language description and an accompanying pointing gesture to identify the most relevant ... | • Humans use less descriptive language when combining language and pointing gestures.
• We use a VLM to augment existing language descriptions (without gestures) to regenerate language that complements the presence of a gesture. |
Generate 3 bullet points for the "Experimental Results / Performance Analysis" section of an academic poster. | ## Paper Content
# Paper Title
SceneDiffuser++: City-Scale Traffic Simulation via a Generative World Model
## Abstract
The goal of traffic simulation is to augment a potentially limited amount of manually-driven miles that is available for testing and validation, with a much larger amount of simulated synthetic mil... | • Compares JS divergence and exiting distance metrics
• Shows SceneDiffuser++ outperforming IDM and SceneDiffusion baselines
• Lower values indicate better performance |
Generate 2 bullet points for the "Core Method / Technical Approach" section of an academic poster. | ## Paper Content
# Paper Title
Balanced Rate-Distortion Optimization in Learned Image Compression
## Abstract
Learned image compression (LIC) using deep learning architectures has seen significant advancements, yet standard rate-distortion (R-D) optimization often encounters imbalanced updates due to diverse gradie... | • Solution 1: Iterative optimization updates weights using gradient descent with step size alpha and Jacobian approximations.
• Solution 2: Quadratic programming (QP) formulates the problem as minimizing a quadratic objective with linear constraints, yielding closed-form solutions for lambda and w_t. |
Generate 3 bullet points for the "Background / Related Work" section of an academic poster. | ## Paper Content
# Paper Title
Perceptual Video Compression with Neural Wrapping
## Abstract
Standard video codecs are rate-distortion optimization machines, where distortion is typically quantified using PSNR versus the source. However, it is now widely accepted that increasing PSNR does not necessarily translate ... | • Conventional codecs are not differentiable.
• Simple motion estimation/prediction blocks are insufficient to capture rate-distortion characteristics.
• Solution: Use a neural codec as a proxy for a conventional codec. |
Generate 2 bullet points for the "Research Motivation / Problem Background" section of an academic poster. | ## Paper Content
# Paper Title
Learning Large-Factor EM Image Super-Resolution with Generative Priors
## Abstract
As the mainstream technique for capturing images of biological specimens at nanometer resolution, electron microscopy (EM) is extremely time-consuming for scanning wide field-of-view (FOV) specimens. In... | • EM image super-resolution (EMSR) can revolutionize EM imaging by enabling faster, less restrictive data acquisition with high-quality wide-field images.
• Existing methods achieve satisfactory results up to 4× magnification but fail for larger factors. |
Generate 2 bullet points for the "Conclusion / Future Work" section of an academic poster. | ## Paper Content
# Paper Title
CheckManual: A New Challenge and Benchmark for Manual-based Appliance Manipulation
## Abstract
Correct use of electrical appliances has significantly improved human life quality. Unlike simple tools that can be manipulated with common sense, different parts of electrical appliances ha... | • Appliance manual understanding presents significant challenges to existing multimodal large models.
• Further exploration is needed for long-horizon appliance manipulation. |
Generate 4 bullet points for the "Core Method / Technical Approach" section of an academic poster. | ## Paper Content
# Paper Title
Revisiting K-mer Profile for Effective and Scalable Genome Representation Learning
## Abstract
Obtaining effective representations of DNA sequences is crucial for genome analysis. Metagenomic binning, for instance, relies on genome representations to cluster complex mixtures of DNA fr... | • Identifiable reads can be uniquely reconstructed from their k-mer profile.
• Theorem provides conditions under which a read has a unique k-mer profile.
• Proposition: k-mer profile mapping is Lipschitz equivalent between metric spaces of reads and k-mer profiles.
• k-mer profile is defined as a weighted sum over cano... |
Generate 3 bullet points for the "Method Overview / Framework" section of an academic poster. | ## Paper Content
# Paper Title
STDD: Spatio-Temporal Dual Diffusion for Video Generation
## Abstract
Diffusion probabilistic model is becoming the cornerstone of data generation, especially generating high-quality images. As an extension, video diffusion generation is in urgent need of a principled temporal-sequenc... | • Proposes an explicit spatio-temporal dual diffusion model integrating temporal and spatial diffusion.
• Theoretically derives a probabilistic reverse process and an accelerated sampling method.
• Achieves more competitive performance on video generation tasks. |
Generate 3 bullet points for the "Background / Related Work" section of an academic poster. | ## Paper Content
# Paper Title
Leak and Learn: An Attacker's Cookbook to Train Using Leaked Data from Federated Learning
## Abstract
Federated learning is a decentralized learning paradigm introduced to preserve privacy of client data. Despite this, prior work has shown that an attacker at the server can still reco... | • Two attack classes: optimization-based (inverting gradients) and linear layer leakage (LOKI).
• Optimization-based attacks require more computation time and larger batch sizes reduce reconstruction quality.
• Linear layer leakage requires model modification but has low computation overhead and reconstructs images nea... |
Generate 5 bullet points for the "Core Method / Technical Approach" section of an academic poster. | ## Paper Content
# Paper Title
Embedding-Free Transformer with Inference Spatial Reduction for Efficient Semantic Segmentation
## Method
This section introduces our Encoder-Decoder Attention Transformer (EDAF-ormer), which is composed of the Embedding-Free Transformer (EFT) encoder and the all-attention decoder. Ad... | • Introduces Embedding-Free Attention (EFA) that removes Q, K, V embedding projections.
• Proposes Inference Spatial Reduction (ISR) to selectively reduce computations at inference without retraining.
• EDAFormer combines EFA with an all-attention decoder for efficient segmentation.
• EFA achieves a better trade-off be... |
Generate 2 bullet points for the "Qualitative Results / Visualization" section of an academic poster. | ## Paper Content
# Paper Title
Projecting Points to Axes: Oriented Object Detection via Point-Axis Representation
## Experiments
DOTA [1] is a large oriented object detection dataset comprising 2,806 images and 188,282 instances distributed across 15 categories, with a substantial variation in orientations, shapes,... | • Qualitative results on text spotting tasks show detection of oriented text in real-world scenes.
• Detected text regions are highlighted with bounding boxes, demonstrating robustness to varying orientations and backgrounds. |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.