instruction stringclasses 95
values | input stringlengths 141 265k | output stringlengths 15 1.61k |
|---|---|---|
Generate 3 bullet points for the "Core Method / Technical Approach" section of an academic poster. | ## Paper Content
# Paper Title
MoST: Motion Style Transformer between Diverse Action Contents
## Abstract
While existing motion style transfer methods are effective between two motions with identical content, their performance significantly diminishes when transferring style between motions with different contents.... | • Uses a Transformer-based encoder-generator with Siamese motion encoder.
• Introduces Part-attentive Style Modulator (PSM) for per-body-part style recombination.
• PSM uses cross-attention to identify relevant body parts and modulate style vectors. |
Generate 2 bullet points for the "Research Motivation / Problem Background" section of an academic poster. | ## Paper Content
# Paper Title
PointOBB: Learning Oriented Object Detection via Single Point Supervision
## Abstract
Single point-supervised object detection is gaining attention due to its cost-effectiveness. However, existing approaches focus on generating horizontal bounding boxes (HBBs) while ignoring oriented ... | • Manually annotating fine-grained oriented bounding boxes (OBB) is time-consuming and expensive.
• To reduce annotation cost, we propose PointOBB to obtain OBB via more cost-effective and efficient single-point annotation. |
Generate 2 bullet points for the "Core Method / Technical Approach" section of an academic poster. | ## Paper Content
## Abstract
In recent years, numerous tasks have been proposed to encourage model to develop specified capability in understanding audio-visual scene, primarily categorized into temporal localization, spatial localization, spatio-temporal reasoning, and pixel-level understanding. Instead, human posse... | • Data: Refine existing datasets to construct an Audio-Visual Unified Instruction-tuning dataset (AV-UIE) that clarifies cooperative relationships.
• Model: Design an interaction-aware LoRA structure with multiple LoRA heads to learn different aspects of audiovisual data interaction. |
Generate 3 bullet points for the "Core Method / Technical Approach" section of an academic poster. | ## Paper Content
# Paper Title
Make It Count: Text-to-Image Generation with an Accurate Number of Objects
## Abstract
Despite the unprecedented success of text-to-image diffusion models, controlling the number of depicted objects using text is surprisingly hard. This is important for various applications from techn... | • Introduces test-time optimization loss to guide layout-to-image transformation.
• Uses self-attention masking and cross-attention to align layout with generated image.
• Ensures object count accuracy during diffusion process. |
Generate 2 bullet points for the "Research Motivation / Problem Background" section of an academic poster. | ## Paper Content
# Paper Title
Mind the Gap: Detecting Black-box Adversarial Attacks in the Making through Query Update Analysis
## Abstract
Adversarial attacks remain a significant threat that can jeopardize the integrity of Machine Learning (ML) models. In particular, query-based black-box attacks can generate ma... | • Query-based black-box adversarial attacks use zeroth-order optimization to estimate gradients.
• Existing countermeasures rely on input space analysis and are evaded by Adaptive Attack (OARS1) that identifies perturbation direction per step. |
Generate 3 bullet points for the "Experimental Results / Performance Analysis" section of an academic poster. | ## Paper Content
# Paper Title
HARNESSING DIVERSITY FOR IMPORTANT DATA SELECTION IN PRETRAINING LARGE LANGUAGE MODELS
## Abstract
Data selection is of great significance in pretraining large language models, given the variation in quality within the large-scale available training corpora. To achieve this, researche... | • Quad (ours) achieves the highest overall score (42.94) and FLOPs efficiency (9.15) across multiple tasks.
• Outperforms baselines like DSIR, PPL, Semdedup, and Quating in General Knowledge, Commonsense Reasoning, and Reading Comprehension.
• Ablation bar charts show our method (Ours) outperforms No-Hess, MLP, and Ind... |
Generate 2 bullet points for the "Experimental Results / Performance Analysis" section of an academic poster. | ## Paper Content
# Paper Title
Efficient Diffusion Transformer with Step-wise Dynamic Attention Mediators
## Experiments
We present the attention mediators to regulate the attention between every two query and key pairs. The high-level idea of attention mediators is to use an additional group of tokens to compress ... | • Higher Quality: Lower FID under same FLOPs.
• Linear Complexity: Much lower FLOPs in higher resolution. |
Generate 3 bullet points for the "Core Method / Technical Approach" section of an academic poster. | ## Paper Content
# Paper Title
Fine-Grained Bipartite Concept Factorization for Clustering
## Abstract
In this paper, we propose a novel concept factorization method that seeks factor matrices using a cross-order positive semi-definite neighbor graph, which provides comprehensive and complementary neighbor informat... | • F is obtained by computing eigenvectors of L associated with top c eigenvalues.
• W and V are updated iteratively using multiplicative rules derived from Lagrangian.
• Constraints ensure rank and connectivity properties for clustering. |
Generate 2 bullet points for the "Other Content" section of an academic poster. | ## Paper Content
# Paper Title
GenNBV: Generalizable Next-Best-View Policy for Active 3D Reconstruction
## Abstract
While recent advances in neural radiance field enable realistic digitization for large-scale scenes, the image-capturing process is still time-consuming and labor-intensive. Previous works attempt to ... | • Project Page: gennbv.github.io
• Contact: cx123@ie.cuhk.edu.hk |
Generate 3 bullet points for the "Core Method / Technical Approach" section of an academic poster. | ## Paper Content
# Paper Title
L-DiffER: Single Image Reflection Removal with Language-based Diffusion Model
## Method
In this paper, we address the problem of language-based reflection removal using diffusion models. We first provide a brief overview of the framework of our language-based reflection removal diffus... | • Describes the diffusion process in both pixel and latent space with iterative condition refinement.
• Includes noise prediction network, compression decoder, and structure encoder components.
• Uses maximum value and saturation-aware structure constraints to guide the denoising process with language prompts. |
Generate 3 bullet points for the "Core Method / Technical Approach" section of an academic poster. | ## Paper Content
# Paper Title
NightAdapter: Learning a Frequency Adapter for Generalizable Night-time Scene Segmentation
## Abstract
Night-time scene segmentation is a critical yet challenging task in the real-world applications, primarily due to the complicated lighting conditions. However, existing methods lack ... | • Apply Discrete Sine Transform (DST) on a frozen VFM to analyze frequency bands.
• Rejection analysis across eight datasets reveals domain gaps concentrate in high-frequency bands.
• Illumination-sensitive bands show large domain gaps; illumination-insensitive bands retain similar patterns in low frequencies. |
Generate 6 bullet points for the "Core Method / Technical Approach" section of an academic poster. | ## Paper Content
# Paper Title
Identifying and Mitigating Position Bias of Multi-image Vision-Language Models
## Abstract
The evolution of Large Vision-Language Models (LVLMs) has progressed from single to multi-image reasoning. Despite this advancement, our findings indicate that LVLMs struggle to robustly utilize... | • Recency bias: for open-source VLMs, they are good at reasoning in the position at the back, while performance is decaying from back to front.
• Lost in the middle: for proprietary VLMs such as GPT-4o, similar to LLMs, where the middle part tends to be ignored by VLMs (U-shape).
• The problem of causal attention for v... |
Generate 2 bullet points for the "Ablation Study" section of an academic poster. | ## Paper Content
# Paper Title
Layout: Flow Matching for Layout Generation
## Experiments
A layout typically consists of various elements, such as text, tables, or images placed on a canvas. Mathematically, we can describe a layout $\mathcal{A}$ as a set of $N$ elements $\mathbf{e}^k$ with $k\in [1,N]$ in t... | • Different conditional vector fields (Linear, Sine/Cosine, Sine) and distributions (Gaussian, Uniform, Mixture) are evaluated.
• Performance metrics (FID, Alignment, Overlap, mIoU) vary slightly across configurations, with Sine/Cosine and Mixture showing competitive results. |
Generate 3 bullet points for the "Core Method / Technical Approach" section of an academic poster. | ## Paper Content
# Paper Title
A COMPUTATIONAL FRAMEWORK FOR MODELING EMERGENCE OF COLOR VISION IN THE HUMAN BRAIN
## Abstract
It is a mystery how the brain decodes color vision purely from the optic nerve signals it receives, with a core inferential challenge being how it disentangles internal perception with the ... | • We simulate the eye to generate biologically realistic optic nerve signals from hyperspectral images.
• Uses a simulation engine of biological eyes processing scene stimuli (400-700nm).
• Includes a realistic model of 2-degree human central fovea with cone cells, bipolar signals, and RGC spikes. |
Generate 2 bullet points for the "Qualitative Results / Visualization" section of an academic poster. | ## Paper Content
# Paper Title
Learning Unsigned Distance Functions from Multi-view Images with Volume Rendering Priors
## Experiments
We evaluate our method in surface reconstruction from multi-view RGB images. We report numerical and visual comparisons with the latest methods of learning UDFs under the same exper... | • We highlight multi-view reconstruction results on open surface scenes and indoor scenes from DTU dataset.
• Our method produces more complete and accurate reconstructions compared to NeUDF, Neus, and NeAT. |
Generate 2 bullet points for the "Core Method / Technical Approach" section of an academic poster. | ## Paper Content
# Paper Title
Instruction Tuning Large Language Models to Understand Electronic Health Records
## Abstract
Large language models (LLMs) have shown impressive capabilities in solving a wide range of tasks based on human instructions. However, developing a conversational AI assistant for electronic h... | • (a) Generation Pipeline for Schema Alignment Data: Clinical Event Sequence → Apply Template → Paraphrase With GPT
• (b) Generation Pipeline for Instruction Following Data: Discharge Summary → Generate With GPT |
Generate 5 bullet points for the "Core Method / Technical Approach" section of an academic poster. | ## Paper Content
# Paper Title
EmoDubber: Towards High Quality and Emotion Controllable Movie Dubbing
## Abstract
Given a piece of text, a video clip, and a reference audio, the movie dubbing task aims to generate speech that aligns with the video while cloning the desired voice. The existing methods have two prima... | • Architecture consists of four main components: Lip-related Prosody Aligning (LPA), Pronunciation Enhancing (PE), Speaker Identity Adapting (SIA), and Flow-based User Emotion Controlling (FUEC).
• LPA focuses on learning consistency between lip motion and phoneme prosody via duration level contrastive learning.
• PE f... |
Generate 3 bullet points for the "Core Method / Technical Approach" section of an academic poster. | ## Paper Content
# Paper Title
Diffusion Models for Open-Vocabulary Segmentation
## Method
We present OVDiff, a method for open-vocabulary segmentation, i.e., semantic segmentation of any category described in natural language. We achieve this goal in three steps: (1) we leverage text-to-image generative models to ... | • Sampling: Bridge vision-language gap, explore multiple possible instances, sample relevant contextual backgrounds.
• Representation: Use multiple complementary feature extractors, use many diverse prototypes for foreground and background.
• Matching: Compare image features to prototypes quickly, use category filter t... |
Generate 3 bullet points for the "Research Motivation / Problem Background" section of an academic poster. | ## Paper Content
# Paper Title
Contrasting intra-modal and ranking cross-modal hard negatives to enhance visio-linguistic compositional understanding
## Abstract
Vision-Language Models (VLMs), such as CLIP, exhibit strong image-text comprehension abilities, facilitating advances in several downstream tasks such as ... | • Vision-language compositional understanding requires distinguishing fine-grained differences in image-text pairs.
• Standard contrastive learning fails to differentiate similar captions (e.g., 'dog on left' vs 'dog on right').
• Our method introduces hard negative generation and fine-grained alignment to improve comp... |
Generate 4 bullet points for the "Core Method / Technical Approach" section of an academic poster. | ## Paper Content
# Paper Title
Progress-Aware Video Frame Captioning
## Abstract
While image captioning provides isolated descriptions for individual images, and video captioning offers one single narrative for an entire video clip, our work explores an important middle ground: progress-aware video captioning at th... | • We develop ProgressCaptioner for fine-grained frame-level captioning.
• Key insight: direct learning of correct caption sequencing as actions unfold.
• Two stages: frame pair captioning and frame sequence captioning.
• Uses supervised fine-tuning (SFT) then direct preference optimization (DPO). |
Generate 2 bullet points for the "Experimental Results / Performance Analysis" 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... | • Table 1: Unconditional generation evaluation shows Ours outperforms baselines (DeepCAD, SolidGen, BrepGen) on COV (75.52), MMD (0.98), JSD (1.17), Novel (98.9), Unique (98.6), Valid (67.7).
• Table 2: Ours uses 7.80×10⁷ parameters, 4 days training, RTX4090x1 GPU, and generates 50 max faces — efficient vs. SolidGen an... |
Generate 3 bullet points for the "Ablation Study" section of an academic poster. | ## Paper Content
# Paper Title
OMG: Occlusion-friendly Personalized Multi-concept Generation in Diffusion Models
## Experiments
Datasets. To evaluate OMG method, we collect a dataset that encompasses 15 distinct concepts. This dataset comprises 7 real-world characters, 3 anime characters, and 5 real-world objects, ... | • Ablation of Layout Preservation: Removing it causes identity degradation.
• Ablation of Sample Type: Concept Noise Blending improves identity preservation.
• Ablation of Number of Concepts: Method scales well with increasing concept count. |
Generate 2 bullet points for the "Experimental Results / Performance Analysis" section of an academic poster. | ## Paper Content
# Paper Title
Adaptive Bidirectional Displacement for Semi-Supervised Medical Image Segmentation
## Abstract
Consistency learning is a central strategy to tackle unlabeled data in semi-supervised medical image segmentation (SSMIS), which enforces the model to produce consistent predictions under th... | • Comparisons on ACDC and PROMISE12 datasets show Ours-ABD outperforms baselines in DSC, Jaccard, and ASD metrics.
• Ablation studies confirm the effectiveness of ABD-R and ABD-I modules with consistent improvements across metrics. |
Generate 4 bullet points for the "Research Motivation / Problem Background" section of an academic poster. | ## Paper Content
# Paper Title
Effective Video Mirror Detection with Inconsistent Motion Cues
## Abstract
Image-based mirror detection has recently undergone rapid research due to its significance in applications such as robotic navigation, semantic segmentation and scene reconstruction. Recently, VMD-Net was propo... | • Mirrors and reflective surfaces are abundant and reduce errors in applications like salient object detection and scene reconstruction.
• Mirror detection is a per-pixel binary classification task to determine if pixels are inside or outside the mirror.
• Mirrors lack their own visual appearance, making detection chal... |
Generate 2 bullet points for the "Ablation Study" section of an academic poster. | ## Paper Content
# Paper Title
Guided Slot Attention for Unsupervised Video Object Segmentation
## Abstract
Unsupervised video object segmentation aims to segment the most prominent object in a video sequence. However, the existence of complex backgrounds and multiple foreground objects make this task challenging. ... | • Compares performance of Random Initialized Slots vs Guided Slots.
• Shows that Guided Slots improve segmentation accuracy in complex scenes with similar objects. |
Generate 2 bullet points for the "Experimental Results / Performance Analysis" section of an academic poster. | ## Paper Content
# Paper Title
SUGAR: Pre-training 3D Visual Representations for Robotics
## Abstract
Learning generalizable visual representations from Internet data has yielded promising results for robotics. Yet, prevailing approaches focus on pre-training 2D representations, being sub-optimal to deal with occlu... | • Bar chart comparing performance of different pre-training methods (from scratch, SN_s, SN_m, Ens_m) with and without grasp modeling
• Average success rates shown for 10 demonstrations vs 100 demonstrations |
Generate 3 bullet points for the "Experimental Results / Performance Analysis" section of an academic poster. | ## Paper Content
# Paper Title
Training-Free Open-Vocabulary Segmentation with Offline Diffusion-Augmented Prototype Generation
## Abstract
Open-vocabulary semantic segmentation aims at segmenting arbitrary categories expressed in textual form. Previous works have trained over large amounts of image-caption pairs t... | • FreeDA (ViT-B) achieves 85.6 J(%) and 43.1 F on COCO Captions, outperforming baselines like ReCo and MaskCLIP.
• FreeDA (ViT-L) achieves 87.9 J(%) and 43.5 F, showing consistent gains across metrics including mIoU and ADE.
• FreeDA uses no training and leverages text-aligned global features for robust segmentation. |
Generate 4 bullet points for the "Experimental Results / Performance Analysis" section of an academic poster. | ## Paper Content
# Paper Title
DGR-MIL: Exploring Diverse Global Representation in Multiple Instance Learning for Whole Slide Image Classification
## Experiments
To validate the effectiveness of the proposed DGR-MIL, we conduct experiments on the CAMELYON16 [5] and TCGA-lung cancer dataset (TCGA-NSCLC).
Dataset an... | • Ablation studies show global representation improves AUC to 0.922 (CAMELYON16) and 0.928 (TCGA-NSCLC).
• Tokenized global vector improves accuracy, F1, and AUC.
• Optimal global vector count is 5 for CAMELYON16 and 3 for TCGA-NSCLC.
• Best performance achieved with λ_tri = 0.1 and λ_div = 0.1. |
Generate 3 bullet points for the "Method Overview / Framework" section of an academic poster. | ## Paper Content
# Paper Title
Self-Supervised Video Desmoking for Laparoscopic Surgery
## Introduction
Laparoscopy is employed to capture videos of the surgical sites to aid surgeons' decision-making, and it has found extensive application in the medical field [37]. However, during the surgery, the activation of h... | • We suggest utilizing the internal characteristics of real-world surgery videos for effective self-supervised video desmoking, and propose a SelfSVD solution, in which the pre-smoke (PS) frame serves as an unaligned supervision.
• We propose to take PS frame as an additional input to further improve desmoking performa... |
Generate 3 bullet points for the "Experimental Results / Performance Analysis" section of an academic poster. | ## Paper Content
# Paper Title
Enhancing Intrinsic Features for Debiasing via Investigating Class-Discerning Common Attributes in Bias-Contrastive Pair
## Abstract
In the image classification task, deep neural networks frequently rely on bias attributes that are spuriously correlated with a target class in the pres... | • Shows performance metrics (BFFHQ, Waterbirds, BAR) across multiple methods including Vanilla, HINT, LfF, and proposed TSAM.
• TSAM achieves state-of-the-art performance: 83.64 on Waterbirds, 85.22 on BFFHQ, 86.33 on BAR.
• Ablation study demonstrates effectiveness of each component. |
Generate 3 bullet points for the "Background / Related Work" section of an academic poster. | ## Paper Content
# Paper Title
Simulator HC: Regression-based Online Simulation of Starting Problem-Solution Pairs for Homotopy Continuation in Geometric Vision
## Abstract
While automatically generated polynomial elimination templates have sparked great progress in the field of 3D computer vision, there remain man... | • Camera calibration is essential for applications like SfM and SLAM.
• These problems are often formulated as polynomials and solved via RANSAC.
• Example: multi-camera system with absolute/relative pose estimation. |
Generate 3 bullet points for the "Qualitative Results / Visualization" section of an academic poster. | ## Paper Content
# Paper Title
BOTH2Hands: Inferring 3D Hands from Both Text Prompts and Body Dynamics
## Abstract
The recently emerging text-to-motion advances have inspired numerous attempts for convenient and interactive human motion generation. Yet, existing methods are largely limited to generating body motion... | • Shows inference results with BOTH57M prompts, including hand motions like finger extension and grasping.
• Demonstrates results with HumanML3D prompts, such as dancing, stepping, and striking motions.
• Visualizes diverse hand poses and interactions with body movements. |
Generate 2 bullet points for the "Core Method / Technical Approach" section of an academic poster. | ## Paper Content
# Paper Title
Compositional Substitutivity of Visual Reasoning for Visual Question Answering
## Method
VQA aims to provide an answer $A$ for a natural language question $Q$ about an image $V$ . For a given training sample $(Q, V)$ with the ground-truth answer $A$ , the question can be denot... | • Method-Specific Loss: L_m = loss(P(Q,V),A) + λ_q loss(P(Q_r,V),A) + λ_v loss(P(Q,V_r),A).
• Contrastive Learning Loss: L_c = -log( e^g(h_q,h_qr) + e^g(h_v,h_vr) ) / ( e^g(h_q,h_qr) + e^g(h_v,h_vr) + e^g(h_q,h_q-) + e^g(h_v,h_v-) ). |
Generate 3 bullet points for the "Core Method / Technical Approach" section of an academic poster. | ## Paper Content
# Paper Title
Difference Inversion: Interpolate and Isolate the Difference with Token Consistency for Image Analogy Generation
## Abstract
How can we generate an image $B'$ that satisfies $A: A' :: B: B'$ , given the input images $A, A'$ and $B$ ? Recent works have tackled this challenge thro... | • Delta Interpolation: Computes an interpolated delta from image delta and text delta to perform prompt optimization.
• Token Consistency Loss: Introduces a Cycle Consistency Loss to ensure A' is generated when the DIFF tokens are absent.
• Zero-Initialization of Token Embeddings: Initializes token embeddings to zero t... |
Generate 3 bullet points for the "Ablation Study" section of an academic poster. | ## Paper Content
# Paper Title
OCAI: Improving Optical Flow Estimation by Occlusion and Consistency Aware Interpolation
## Abstract
The scarcity of ground-truth labels poses one major challenge in developing optical flow estimation models that are both generalizable and robust. While current methods rely on data au... | • Ablates components of semi-supervised flow training: EMA, Confidence-based loss masking, and image distortion.
• Compares variants using different hole filling methods: BHF and CBF.
• Shows OCAI with all components achieves best performance on KITTI. |
Generate 2 bullet points for the "Background / Related Work" section of an academic poster. | ## Paper Content
# Paper Title
Automatically Identify and Rectify: Robust Deep Contrastive Multi-view Clustering in Noisy Scenarios
## Abstract
Leveraging the powerful representation learning capabilities, deep multi-view clustering methods have demonstrated reliable performance by effectively integrating multi-sou... | • Deep multi-view clustering methods leverage powerful representation learning to integrate multi-source information from diverse views.
• These methods have shown reliable performance in recent years by effectively combining information across views. |
Generate 3 bullet points for the "Core Method / Technical Approach" section of an academic poster. | ## Paper Content
# Paper Title
Video Prediction by Modeling Videos as Continuous Multi-Dimensional Processes
## Abstract
Diffusion models have made significant strides in image generation, mastering tasks such as unconditional image synthesis, text-image translation, and image-to-image conversions. However, their c... | • Derives loss function using forward (FWD) and reverse (RWS) processes
• Uses KL divergence between two Gaussians for simplification
• Final loss expression involves expected value of squared error |
Generate 3 bullet points for the "Core Method / Technical Approach" section of an academic poster. | ## Paper Content
# Paper Title
MICDrop: Masking Image and Depth Features via Complementary Dropout for Domain-Adaptive Semantic Segmentation
## Method
Overview. In Fig.2, we present our method, featuring two novel modules that can be plugged into various UDA methods to leverage geometric cues. Our feature fusion mo... | • Easy Integration: Strategy can be easily added to existing networks.
• Low computational complexity: Does not require additional forward passes over backbone.
• Diagram shows student and teacher encoders/decoders with complementary masking and pseudo-labeling. |
Generate 2 bullet points for the "Experimental Results / Performance Analysis" section of an academic poster. | ## Paper Content
# Paper Title
3DToonify: Creating Your High-Fidelity 3D Stylized Avatar Easily from 2D Portrait Images
## Abstract
Visual content creation has aroused a surge of interest given its applications in mobile photography and AR/VR. Portrait style transfer and 3D recovery from monocular images as two rep... | • Artistic portrait generation: diverse stylized 3D avatars from input portraits.
• 3D assets export:展示了 stylized avatar, mesh, and UV texture outputs. |
Generate 4 bullet points for the "Research Motivation / Problem Background" section of an academic poster. | ## Paper Content
# Paper Title
PIDLoc: Cross-View Pose Optimization Network Inspired by PID Controllers
## Abstract
Accurate localization is essential for autonomous driving, but GNSS-based methods struggle in challenging environments such as urban canyons. Cross-view pose optimization offers an effective solution ... | • GNSS can degrade in urban canyons and bad weather.
• 3D maps are expensive to generate and maintain.
• Geo-tagged satellite images can be a lightweight alternative.
• Cross-view pose optimization (CVPO) is proposed as a solution. |
Generate 3 bullet points for the "Research Motivation / Problem Background" section of an academic poster. | ## Paper Content
# Paper Title
Proximal Algorithm Unrolling: Flexible and Efficient Reconstruction Networks for Single-Pixel Imaging
## Abstract
Deep-unrolling and plug-and-play (PnP) approaches have become the de-facto standard solvers for single-pixel imaging (SPI) inverse problem. PnP approaches, a class of iter... | • Task: In single-pixel cameras, an image X is compressed into measurement Y via Y = HXW^T, creating an inverse problem.
• Problem: Most deep reconstruction models are CR-specific, lacking flexibility for varying compression ratios.
• Contribution: Propose ProxUnroll to train unrolled networks with an ideal regularizer... |
Generate 3 bullet points for the "Core Method / Technical Approach" section of an academic poster. | ## Paper Content
# Paper Title
Bridging Gait Recognition and Large Language Models Sequence Modeling
## Abstract
Gait sequences exhibit sequential structures and contextual relationships similar to those in natural language, where each element—whether a word or a gait step—is connected to its predecessors and succe... | • Proposes an add-on G2L-L2G framework to convert gait features into language tokens and vice versa.
• G2L maps gait sequences to linguistic representations; L2G reconstructs gait from language tokens.
• Enables gait recognition via LLM-based sequence modeling. |
Generate 3 bullet points for the "Core Method / Technical Approach" section of an academic poster. | ## Paper Content
# Paper Title
Unveiling Differences in Generative Models: A Scalable Differential Clustering Approach
## Abstract
A fine-grained comparison of generative models requires the identification of sample types generated differently by each of the involved models. While quantitative scores have been prop... | • Proposes a spectral method on matrix K_X|ρY for unsupervised cluster identification.
• Matrix is non-symmetric and may not be diagonalizable, making it hard to apply to large datasets like ImageNet or MS-COCO.
• Complexity is O(n³), which is prohibitive for millions of samples. |
Generate 3 bullet points for the "Core Method / Technical Approach" section of an academic poster. | ## Paper Content
# Paper Title
FreeKD: Knowledge Distillation via Semantic Frequency Prompt
## Abstract
Knowledge distillation (KD) has been applied to various tasks successfully, and mainstream methods typically boost the student model via spatial imitation losses. However, the consecutive downsamplings induced in... | • Relational attention from multi-receptive fields represented as A = Softmax(ψ(F)F^T).
• Gating operation generates position-aware loss relation weight ω = G(A) ∈ ℝ^(1×C).
• Final frequency distillation loss: L_FreeKD = Σ ω^(r) ||M⊗a_k - M⊗b_k||_1. |
Generate 3 bullet points for the "Experimental Results / Performance Analysis" section of an academic poster. | ## Paper Content
# Paper Title
Naturally Supervised 3D Visual Grounding with Language-Regularized Concept Learners
## Abstract
3D visual grounding is a challenging task that often requires direct and dense supervision, notably the semantic label for each object in the scene. In this paper, we instead study the natu... | • LARC outperforms prior methods (NS, LAR, MVT, etc.) on test, symmetry, and exclusivity accuracy
• Achieves 38.4% exclusivity accuracy, significantly higher than prior methods
• Shows strong data efficiency: higher accuracy with less training data |
Generate 2 bullet points for the "Experimental Results / Performance Analysis" section of an academic poster. | ## Paper Content
# Paper Title
Bridging The Gap between Low-rank and Orthogonal Adaptation via Householder Reflection Adaptation
## Abstract
While following different technical routes, both low-rank and orthogonal adaptation techniques can efficiently adapt large-scale pre-training models in specific tasks or domai... | • Scatter plot shows HRA achieving high GLUE scores with fewer trainable parameters compared to OFT, BOFT, AdaLoRA, P-Adapter, and H-Adapter.
• Highlights efficiency and effectiveness of HRA in balancing performance and parameter count. |
Generate 5 bullet points for the "Implementation Details" section of an academic poster. | ## Paper Content
# Paper Title
PEA-Diffusion: Parameter-Efficient Adapter with Knowledge Distillation in non-English Text-to-Image Generation
## Method
A naive approach for generating images from non-English text is to substitute the CLIP text encoder in SD with a language-specific CLIP encoder. The text encoder is... | • The adapter is trained with a small parallel data corpus and uses feature-level and logits knowledge distillation.
• Training uses MSE, KL, and MiniLLM losses with lambda weights for balancing objectives.
• The model supports multiple teacher models including SD1.5, SDXL, Stable Cascade, and others.
• Adaptability te... |
Generate 3 bullet points for the "Ablation Study" section of an academic poster. | ## Paper Content
# Paper Title
Boosting Spike Camera Image Reconstruction from a Perspective of Dealing with Spike Fluctuations
## Abstract
As a bio-inspired vision sensor with ultra-high speed, spike cameras exhibit great potential in recording dynamic scenes with high-speed motion or drastic light changes. Differ... | • Ablating MODF components (spike, D(1,1), DCN, CAPA) shows each contributes to performance; removing any reduces PSNR.
• Increasing pyramid levels (Pym) and CAPA layers improves PSNR, with optimal settings at Pym=3, CAPA=3.
• Increasing input frames from 21 to 81 improves PSNR slightly, confirming temporal information... |
Generate 3 bullet points for the "Core Method / Technical Approach" section of an academic poster. | ## Paper Content
# Paper Title
Tetrahedron Splatting for 3D Generation
## Abstract
3D representation is essential to the significant advance of 3D generation with 2D diffusion priors. As a flexible representation, NeRF has been first adopted for 3D representation. With density-based volumetric rendering, it however... | • Overview of TeT-Splattting: pre-filter, project to 2D splats, alpha-blend using SDF-derived opacity.
• Two-stage 3D generation: Stage1 uses TeT-Splattting for geometry; Stage2 uses Marching Tetrahedra for texture optimization.
• Loss functions include SDS, Eikonal, and normal consistency terms. |
Generate 2 bullet points for the "Core Method / Technical Approach" section of an academic poster. | ## Paper Content
# Paper Title
SignGraph: A Sign Sequence is Worth Graphs of Nodes
## Abstract
Despite the recent success of sign language research, the widely adopted CNN-based backbones are mainly migrated from other computer vision tasks, in which the contours and texture of objects are crucial for identifying o... | • Local sign graph module (LSG) builds graphs by connecting K nearest neighbors for each node to learn the correlation of cross-region features in one frame.
• Temporal sign graph module (TSG) builds graphs by choosing top K node pairs between two adjacent frames to learn the interaction of cross-region features among ... |
Generate 4 bullet points for the "Research Motivation / Problem Background" section of an academic poster. | ## Paper Content
# Paper Title
DA-BEV: Unsupervised Domain Adaptation for Bird's Eye View Perception
## Introduction
3D visual perception [32, 45] aims at sensing and understanding surrounding environments in 3D space, which plays an important role in various applications such as mobile robotics [3, 20], autonomous... | • Camera-only BEV models trained on source domain suffer performance degradation on target domain due to cross-domain discrepancy.
• BEV features show severe domain discrepancy from explicit camera configuration changes.
• Image-view features have less domain discrepancy but lack global 3D information.
• Goal: bridge i... |
Generate 3 bullet points for the "Experimental Results / Performance Analysis" section of an academic poster. | ## Paper Content
# Paper Title
WhodunitBench: Evaluating Large Multimodal Agents via Murder Mystery Games
## Abstract
Recently, large language models (LLMs) have achieved superior performance, empowering the development of large multimodal agents (LMAs). An LMA expected to perform practical tasks must possess a ran... | • Agent vs. Agent table shows performance across perception, role-play, decision-making, and cognition.
• GPT-4V achieves 24.2% Avg. Win Ratio, outperforming others.
• Higher CoE scores correlate with higher win rates in the arena. |
Generate 3 bullet points for the "Qualitative Results / Visualization" section of an academic poster. | ## Paper Content
# Paper Title
Discover and Mitigate Multiple Biased Subgroups in Image Classifiers
## Abstract
Machine learning models can perform well on in-distribution data but often fail on biased subgroups that are underrepresented in the training data, hindering the robustness of models for reliable applicat... | • Compares subgroup visualizations from Jain et al., Domino, and DIM (Ours).
• DIM shows clearer subgroup separation (e.g., Forest/Mountain, Plain/Cloud).
• Demonstrates improved subgroup discovery and bias mitigation. |
Generate 2 bullet points for the "Qualitative Results / Visualization" section of an academic poster. | ## Paper Content
# Paper Title
A Simple Baseline for Spoken Language to Sign Language Translation with 3D Avatars
## Experiments
Datasets and Evaluation Metrics. Due to the absence of SMPL-X annotations in existing sign language datasets, back-translation is adopted as the primary metric [56-58] to evaluate the Spo... | • Qualitative results on Phoenix-2014T and CSL-Daily show natural 3D avatar motions.
• Ablation studies show improvements with L_unseen and L_upright constraints. |
Generate 3 bullet points for the "Method Overview / Framework" section of an academic poster. | ## Paper Content
# Paper Title
Improving Adversarial Transferability via Model Alignment
## Introduction
Adversarial examples with small perturbations can mislead deep neural networks to make wrong predictions [50]. It has been observed that perturbations created to attack one model can also fool other models, whic... | • Illustrates how adversarial attacks transfer from source to target models.
• Shows perturbed input generation via attack algorithms and model augmentation.
• Highlights the goal: transform source models to generate more transferable perturbations. |
Generate 4 bullet points for the "Method Overview / Framework" section of an academic poster. | ## Paper Content
# Paper Title
SinGS: Animatable Single-Image Human Gaussian Splats with Kinematic Priors
## Abstract
Despite significant advances in accurately estimating geometry in contemporary single-image 3D human reconstruction, creating a high-quality, efficient, and animatable 3D avatar remains an open chal... | • Created Rotation-Human Datasets with 10k+ mono-videos for turning motions.
• KHD infers unseen views and motion from video using Kinematic Sampling.
• GPGS builds compact, accurate avatars under imperfect priors.
• SinGS enables high-quality, efficient, animatable 3D avatar reconstruction from a single image. |
Generate 3 bullet points for the "Ablation Study" 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... | • Evaluates impact of adding data on Top-1 accuracy.
• Compares settings with and without shuffling, adding labels, and adding both.
• Shows consistent improvement with additional data and labels. |
Generate 3 bullet points for the "Core Method / Technical Approach" section of an academic poster. | ## Paper Content
# Paper Title
CAPE: CAM as a Probabilistic Ensemble for Enhanced DNN Interpretation
## Abstract
Deep Neural Networks (DNNs) are widely used for visual classification tasks, but their complex computation process and black-box nature hinder decision transparency and interpretability. Class activation... | • Introduces CAPE (CAM as a Probabilistic Ensemble) to measure probabilistic contribution of image regions (pi) toward model prediction (pc = Σpi).
• Proposes μ-CAPE to restore class mutual regions by including class mutual region inclusive explanation.
• Two training setups: TS (training both vanilla classifier and CA... |
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... | • Evaluates performance with 1, 2, or 3 adaptive layers.
• 3 layers achieve 34.82 PSNR at r=1/4 with 33.01% parameters.
• 1 layer achieves 34.63 PSNR at r=1/4 with 32.45% parameters. |
Generate 1 bullet points for the "Other Content" 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... | This work was supported in part by NSF 1910284, 2142725, and 2103829. |
Generate 3 bullet points for the "Research Motivation / Problem Background" section of an academic poster. | ## Paper Content
# Paper Title
Frugal 3D Point Cloud Model Training via Progressive Near Point Filtering and Fused Aggregation
## Introduction
3D point cloud is a set of points representing scenes or objects with 3D geometry information from sensors (e.g., LiDAR and RGB-D cameras). Deep neural networks (DNNs) [3,7,... | • FPS takes up 44.69% of training time due to quadratic complexity.
• Aggregation takes up 22.84% due to redundant GPU memory accesses.
• These bottlenecks motivate the need for lightweight sampling and fused aggregation. |
Generate 3 bullet points for the "Method Overview / Framework" section of an academic poster. | ## Paper Content
# Paper Title
Noisy-Correspondence Learning for Text-to-Image Person Re-identification
## Abstract
Text-to-image person re-identification (TIREID) is a compelling topic in the cross-modal community, which aims to retrieve the target person based on a textual query. Although numerous TIREID methods ... | • Introduces RDE: cross-modal embedding model with BGE and TSE modules.
• Core components: Confident Consensus Division (CCD) and Triplet Alignment Loss (TAL).
• CCD filters noisy pairs; TAL uses an upper bound to consider all negative pairs for stable learning. |
Generate 5 bullet points for the "Core Method / Technical Approach" section of an academic poster. | ## Paper Content
# Paper Title
Gradient Inversion Attacks on Parameter-Efficient Fine-Tuning
## Abstract
Federated learning (FL) allows multiple data-owners to collaboratively train machine learning models by exchanging local gradients, while keeping their private data on device. To simultaneously enhance privacy a... | • Malicious design of parameters: E = 0.5 I_{D×D}, E^{(n)}_{pos} ~ N(0, σ) with σ = 10.
• Ensures: (E^{(t)}_{pos})^T E^{(n)}_{pos} >> (E^{(t)}_{pos})^T E^{(m)}_{pos} for n ≠ t.
• Mean(y^{(n,m)}) ≈ 0.
• Adapter1 targets intervals a and b; Adapter2 targets intervals c and d.
• Recovery from pair of neuron gradients and d... |
Generate 4 bullet points for the "Core Method / Technical Approach" section of an academic poster. | ## Paper Content
# Paper Title
PH-Net: Semi-Supervised Breast Lesion Segmentation via Patch-wise Hardness
## Abstract
We present a novel semi-supervised framework for breast ultrasound (BUS) image segmentation, which is a very challenging task owing to (1) large scale and shape variations of breast lesions and (2) ... | • In the teacher network, split the image into multiple patches and calculate the average entropy as the hardness score.
• Shield hard patches to prevent them from being cropped out while performing random patch cutmix on the remaining patches.
• Incorporate a projector into the student branch to further enhance model ... |
Generate 3 bullet points for the "Ablation Study" section of an academic poster. | ## Paper Content
# Paper Title
Semi-Supervised Video Desnowing Network via Temporal Decoupling Experts and Distribution-Driven Contrastive Regularization
## Experiments
In order to tackle the aforementioned issue of lacking suitable video datasets to generalize the performance of desnowing algorithms in real-world ... | • Ablates components: temporal proposal module, contrastive loss, and unlabeled data usage.
• Removing contrastive regularization drops J(%) by 3.2%.
• Unlabeled data contributes +1.8% J(%) improvement. |
Generate 4 bullet points for the "Core Method / Technical Approach" section of an academic poster. | ## Paper Content
# Paper Title
SimLTD: Simple Supervised and Semi-Supervised Long-Tailed Object Detection
## Abstract
While modern visual recognition systems have made significant advancements, many continue to struggle with the open problem of learning from few exemplars. This paper focuses on the task of object d... | • Overview of SimLTD: partitions long-tailed distribution into frequent head and rare tail classes.
• Step 1: Pre-train on head classes with optional unlabeled images.
• Step 2: Transfer learn on tail classes with optional unlabeled images.
• Step 3: Fine-tune on balanced sample of head and tail classes for exemplar re... |
Generate 3 bullet points for the "Research Motivation / Problem Background" section of an academic poster. | ## Paper Content
# Paper Title
Human-in-the-Loop Visual Re-ID for Population Size Estimation
## Introduction
Computer vision-based re-identification (Re-ID) is increasingly being deployed to analyze large image collections for species-level population monitoring [47,56]. The association of individuals across camera... | • Goal: estimate population size in an image collection.
• Assume an embedding or approximate pairwise similarity ŝ(x,z).
• Reduces to clustering, but only the number of clusters matters, not clustering accuracy. |
Generate 2 bullet points for the "Experimental Results / Performance Analysis" section of an academic poster. | ## Paper Content
# Paper Title
DreamControl: Control-Based Text-to-3D Generation with 3D Self-Prior
## Abstract
3D generation has raised great attention in recent years. With the success of text-to-image diffusion models, the 2D-lifting technique becomes a promising route to controllable 3D generation. However, the... | • Qualitative Results: Visual comparison showing improved consistency and detail over baselines
• Quantitative Results: Metrics including J(%) and F-score demonstrating superior performance |
Generate 3 bullet points for the "Experimental Results / Performance Analysis" section of an academic poster. | ## Paper Content
# Paper Title
Robust Distillation via Untargeted and Targeted Intermediate Adversarial Samples
## Abstract
Adversarially robust knowledge distillation aims to compress large-scale models into lightweight models while preserving adversarial robustness and natural performance on a given dataset. Exis... | • Main results table compares DARWIN with baselines on CIFAR-10 and CIFAR-100 under natural and adversarial settings.
• Shows DARWIN-LF achieves top performance across multiple metrics (e.g., 52.33 F on CIFAR-10 CW).
• Includes ablation study on CIFAR-10/100 showing contribution of each component. |
Generate 4 bullet points for the "Conclusion / Future Work" section of an academic poster. | ## Paper Content
# Paper Title
Saliency-driven Experience Replay for Continual Learning
## Abstract
We present Saliency-driven Experience Replay - SER - a biologically-plausible approach based on replicating human visual saliency to enhance classification models in continual learning settings. Inspired by neurophys... | • SER is a neuro-inspired approach based on replicating human visual saliency to enhance classification models in CL.
• By incorporating saliency-driven modulation, SER improves state-of-the-art CL methods, reducing forgetting.
• The saliency-based modulation significantly enhances robustness to adversarial attacks.
• ... |
Generate 2 bullet points for the "Core Method / Technical Approach" section of an academic poster. | ## Paper Content
# Paper Title
Heavy-Tailed Class Imbalance and Why Adam Outperforms Gradient Descent on Language Models
## Abstract
Adam has been shown to outperform gradient descent on large language models by a larger margin than on other tasks, but it is unclear why. We show that a key factor in this performanc... | • Correlation emerges between gradient and Hessian in last layer W.
• At initialization, gradient scales with πk and Hessian is uniform. |
Generate 2 bullet points for the "Research Motivation / Problem Background" section of an academic poster. | ## Paper Content
# Paper Title
Relational Matching for Weakly Semi-Supervised Oriented Object Detection
## Abstract
Oriented object detection has witnessed significant progress in recent years. However, the impressive performance of oriented object detectors is at the huge cost of labor-intensive annotations, and d... | • Introduce point annotations to balance detection performance and annotation cost for oriented object detection.
• The ambiguity of point annotations includes orientation-agnostic, location-ambiguous, and scale-unconstrained cases. |
Generate 3 bullet points for the "Qualitative Results / Visualization" section of an academic poster. | ## Paper Content
# Paper Title
PURA: Parameter Update-Recovery Test-Time Adaption for RGB-T Tracking
## Abstract
Maintaining stable tracking of objects in domain shift scenarios is crucial for RGB-T tracking, prompting us to explore the use of unlabeled test sample information for effective online model adaptation.... | • Qualitative comparison of our method with other RGB-T trackers across different attributes.
• The figure depicts the tracking results of the object in low-light conditions.
• PURA shows more accurate and stable bounding boxes compared to baselines. |
Generate 3 bullet points for the "Core Method / Technical Approach" section of an academic poster. | ## Paper Content
# Paper Title
A Compact Dynamic 3D Gaussian Representation for Real-Time Dynamic View Synthesis
## Method
Given images with timesteps and camera parameters obtained from videos, our task is to learn a 4D spatial-temporal representation of a dynamic scene that enables fast and high-quality view rend... | • Pipeline: Image sequence → 3D Gaussian splatting → rendered frame
• Objective: Optimize training frame to match rendered output
• Uses canonical coordinates and time-varying parameters for dynamic modeling |
Generate 3 bullet points for the "Experimental Results / Performance Analysis" section of an academic poster. | ## Paper Content
# Paper Title
Convex Combination Star Shape Prior for Data-driven Image Semantic Segmentation
## Abstract
Multi-center star shape is a prevalent object shape feature, which has proven effective in model-based image segmentation methods. However, the shape field function induced by the multi-center ... | • Reports performance on medical image datasets (ISIC, Refuge, Kvasir) using DICE, IOU, and ACC metrics.
• Compares SAM and SAM2 backbones with Finetuned, CCS loss, and CCS module variants.
• Shows CCS module consistently improves performance across datasets. |
Generate 3 bullet points for the "Core Method / Technical Approach" section of an academic poster. | ## Paper Content
# Paper Title
SceneTeller: Language-to-3D Scene Generation
## Method
Fig. 2: Overview of our method. Given a textual prompt in natural language delineating the desired spatial positions and orientations of objects wi... | • Pipeline includes Layout Generation Module, 3D Scene Assembling, and 3DGS initialization.
• Uses InstructPix2Pix for stylization with prompts like 'Make it Pirates of the Caribbean style'.
• Object masks guide placement and orientation of furniture in 3D space. |
Generate 3 bullet points for the "Qualitative Results / Visualization" section of an academic poster. | ## Paper Content
# Paper Title
PoseBH: Prototypical Multi-Dataset Training Beyond Human Pose Estimation
## Abstract
We study multi-dataset training (MDT) for pose estimation, where skeletal heterogeneity presents a unique challenge that existing methods have yet to address. In traditional domains, e.g. regression a... | • Results on COCO (left) and AP-10K (right): Our method predicts more accurate keypoints on the right-hand of the standing person (left) and the right front toe of the dog (right).
• Results on InterHand2.6M (a-c) and 3DPW (e-f): Our method is more robust to the occlusion.
• SMPL fitting results on 3DPW using KITRO: Ou... |
Generate 4 bullet points for the "Core Method / Technical Approach" section of an academic poster. | ## Paper Content
# Paper Title
General Object Foundation Model for Images and Videos at Scale
## Abstract
We present GLEE in this work, an object-level foundation model for locating and identifying objects in images and videos. Through a unified framework, GLEE accomplishes detection, segmentation, tracking, ground... | • The framework includes a Text Encoder, Image Backbone, Visual Prompter, and Object Decoder.
• Applied to image tasks: accepts textual descriptions including object categories, names, captions, and referring expressions.
• Applied to video tasks: supports VIS, MOT, VOS, RVOS, and interactive tracking with open-world/v... |
Generate 3 bullet points for the "Implementation Details" section of an academic poster. | ## Paper Content
# Paper Title
L-DiffER: Single Image Reflection Removal with Language-based Diffusion Model
## Method
In this paper, we address the problem of language-based reflection removal using diffusion models. We first provide a brief overview of the framework of our language-based reflection removal diffus... | • Requires noise prediction network, compression decoder, color encoder, and structure encoder.
• Takes mixture image and language condition as input.
• Outputs recovered latent representation through iterative sampling with constraints and updates. |
Generate 3 bullet points for the "Experimental Results / Performance Analysis" section of an academic poster. | ## Paper Content
# Paper Title
Intriguing Properties of Diffusion Models: An Empirical Study of the Natural Attack Capability in Text-to-Image Generative Models
## Abstract
Denoising probabilistic diffusion models have shown breakthrough performance to generate more photo-realistic images or human-level illustratio... | • Classification accuracy shows no major differences between shared and non-shared training datasets.
• Robustified model against non-robust features also shows high robustness against NDD attacks.
• Attack on robustified model maintains high accuracy even under attack conditions. |
Generate 3 bullet points for the "Research Motivation / Problem Background" section of an academic poster. | ## Paper Content
# Paper Title
Continual Segmentation with Disentangled Objectness Learning and Class Recognition
## Abstract
Most continual segmentation methods tackle the problem as a per-pixel classification task. However, such a paradigm is very challenging, and we find query-based segmenters with built-in obje... | • Unlike conventional per-pixel-based segmenters, mask classification unleashes a hidden property, objectness, facilitating continual learning.
• Objectness generalizes to new classes similar to previous classes.
• After finetuning, objectness is resistant to catastrophic forgetting. |
Generate 5 bullet points for the "Method Overview / Framework" section of an academic poster. | ## Paper Content
# Paper Title
RoHM: Robust Human Motion Reconstruction via Diffusion
## Abstract
We propose RoHM, an approach for robust 3D human motion reconstruction from monocular RGB(-D) videos in the presence of noise and occlusions. Most previous approaches either train neural networks to directly regress mo... | • Reconstruct realistic 3D motions in global space from monocular videos
• Robust to noise and occlusions
• RoHM: a diffusion-based approach
• TrajControl: to model trajectory-pose correlations
• Applications: motion reconstruction, denoising, infilling |
Generate 3 bullet points for the "Research Motivation / Problem Background" section of an academic poster. | ## Paper Content
# Paper Title
FSD-BEV: Foreground Self-Distillation for Multi-view 3D Object Detection
## Introduction
3D object detection is a fundamental task in autonomous driving. Although LiDAR-based methods [31,34,35,38] have demonstrated excellent performance,
![](images/9a5f7d7a7e19c2ff3cbd9c56e09c6f0e187... | • Cross-modal distillation requires a separate teacher model and pre-training.
• Distribution discrepancies exist between teacher and student BEV.
• Our self-distillation approach uses a shared backbone and avoids teacher-student mismatch. |
Generate 3 bullet points for the "Research Motivation / Problem Background" section of an academic poster. | ## Paper Content
# Paper Title
AnyDressing: Customizable Multi-Garment Virtual Dressing via Latent Diffusion Models
## Abstract
Recent advances in garment-centric image generation from text and image prompts based on diffusion models are impressive. However, existing methods lack support for various combinations of... | • Garment Fidelity: Existing methods encounter confusion while preserving the intricate textures of each garment.
• Text-Image Consistency: Ensure faithfulness to the prompts.
• Plugin Compatibility: Support integration with control plugins like ControlNet and LoRA. |
Generate 2 bullet points for the "Qualitative Results / Visualization" section of an academic poster. | ## Paper Content
# Paper Title
Chain of Attack: On the Robustness of Vision-Language Models Against Transfer-Based Adversarial Attacks
## Abstract
Pre-trained vision-language models (VLMs) have showcased remarkable performance in image and natural language understanding, such as image captioning and response genera... | • Demonstrates targeted response generation for ViECap, SmallCap, and UniDiffuser.
• Shows how adversarial examples lead to specific target texts, with CLIP scores indicating success. |
Generate 3 bullet points for the "Core Method / Technical Approach" section of an academic poster. | ## Paper Content
# Paper Title
Consistent Prompting for Rehearsal-Free Continual Learning
## Abstract
Continual learning empowers models to adapt autonomously to the ever-changing environment or data streams without forgetting old knowledge. Prompt-based approaches are built on frozen pre-trained models to learn th... | • Proposes Smooth Regularization to encourage maximum logit values of old task classifiers to be lower than current class by a predefined margin.
• Uses a soft margin function to block gradients when necessary.
• Defines the loss function ℒ_e(i) using softmax and inner product operators. |
Generate 3 bullet points for the "Research Motivation / Problem Background" section of an academic poster. | ## Paper Content
# Paper Title
When Is Inductive Inference Possible?
## Abstract
Can a physicist make only a finite number of errors in the eternal quest to uncover the law of nature? This millennium-old philosophical problem, known as inductive inference, lies at the heart of epistemology. Despite its significance... | • Two ways of human reasoning: deductive (rigorous by logic) and inductive (based on limited observations).
• Induction cannot reliably predict the unexperienced (e.g., black swans).
• Core question: How to model induction and what guarantees can it offer? |
Generate 4 bullet points for the "Core Method / Technical Approach" section of an academic poster. | ## Paper Content
# Paper Title
Higher-order Relational Reasoning for Pedestrian Trajectory Prediction
## Abstract
Social relations have substantial impacts on the potential trajectories of each individual. Modeling these dynamics has been a central solution for more precise and accurate trajectory forecasting. Howe... | • Proposes HighGraph: a graph-based higher-order relational reasoning module.
• Uses collision-aware kernel function to assign interaction weights between agents.
• Employs higher-order graph convolutions to model distant social influences.
• Integrates multi-hop interactions via combination layers and MLPs. |
Generate 3 bullet points for the "Experimental Results / Performance Analysis" section of an academic poster. | ## Paper Content
# Paper Title
Neural Implicit Morphing of Face Images
## Abstract
Face morphing is a problem in computer graphics with numerous artistic and forensic applications. It is challenging due to variations in pose, lighting, gender, and ethnicity. This task consists of a warping for feature alignment and... | • Quantitative results show our method outperforms baselines in DET and MorDeeply metrics.
• Blending impacts morphing detection: our warping+diffAE is comparable to pure diffAE, followed by StyleGAN3 and ours+S.Mix.
• Non-linear warping leads to smoother alignment, especially noticeable in videos. |
Generate 4 bullet points for the "Method Overview / Framework" section of an academic poster. | ## Paper Content
# Paper Title
Recurrent Feature Mining and Keypoint Mixup Padding for Category-Agnostic Pose Estimation
## Abstract
Category-agnostic pose estimation aims to locate keypoints on query images according to a few annotated support images for arbitrary novel classes. Existing methods generally extract ... | • We propose a novel framework to recurrently mine fine-grained and structure-aware features from both support and query images, which could extract better support features and locate more precise keypoints on query images.
• We propose a feature mining module based on deformable attention to integrally mine fine-grain... |
Generate 3 bullet points for the "Experimental Results / Performance Analysis" section of an academic poster. | ## Paper Content
# Paper Title
Filter Images First, Generate Instructions Later: Pre-Instruction Data Selection for Visual Instruction Tuning
## Abstract
Visual instruction tuning (VIT) for large vision-language models (LVLMs) requires training on expansive datasets of image-instruction pairs, which can be costly. ... | • Evaluates performance across different sampling ratios (10% to 50%) for PreSel and baselines.
• PreSel consistently outperforms baselines (Random, COINCIDE, TypiClust, EL2N) across all sampling ratios.
• Ablation shows that task-importance weighting significantly improves performance over size-based sampling without ... |
Generate 5 bullet points for the "Background / Related Work" section of an academic poster. | ## Paper Content
# Paper Title
Towards Generalizable Multi-Object Tracking
## Abstract
Multi-Object Tracking (MOT) encompasses various tracking scenarios, each characterized by unique traits. Effective trackers should demonstrate a high degree of generalizability across diverse scenarios. However, existing trackers... | • Motion Complexity reflects the irregularity and unpredictability of target motion within the scenario.
• Variation Amplitude reflects the target’s variability, encompassing both shape and position.
• Target Density reflects the density of the crowds in the scenario.
• Small Target represents the average amount of sma... |
Generate 2 bullet points for the "Qualitative Results / Visualization" section of an academic poster. | ## Paper Content
# Paper Title
Action-slot: Visual Action-centric Representations for Multi-label Atomic Activity Recognition in Traffic Scenes
## Abstract
In this paper, we study multi-label atomic activity recognition. Despite the notable progress in action recognition, it is still challenging to recognize atomic... | • Shows scenario with traffic cones occluding intersection; atomic activities: Z4-Z3:C+ and Z1-Z3:K.
• Shows construction covering corner C4; atomic activities: Z1-Z3:K. |
Generate 3 bullet points for the "Core Method / Technical Approach" section of an academic poster. | ## Paper Content
# Paper Title
FLAIR: VLM with Fine-grained Language-informed Image Representations
## Abstract
CLIP has shown impressive results in aligning images and texts at scale. However, its ability to capture detailed visual features remains limited because CLIP matches images and texts at a global level. T... | • Illustrates positive and negative sample selection for training.
• Shows how meaningful negatives are chosen based on text-image alignment.
• Visualizes attention maps for different negative types. |
Generate 3 bullet points for the "Core Method / Technical Approach" section of an academic poster. | ## Paper Content
# Paper Title
Goal Reduction with Loop-Removal Accelerates RL and Models Human Brain Activity in Goal-Directed Learning
## Abstract
Goal-directed planning presents a challenge for classical RL algorithms due to the vastness of the combinatorial state and goal spaces, while humans and animals adapt ... | • Three sampling strategies: random, trajectory, and loop-removal sampling.
• Loop-removal detects and eliminates loops in memory by selecting a proper filtration radius.
• This filters out 'bad' subgoals, improving subgoal quality and learning efficiency. |
Generate 3 bullet points for the "Core Method / Technical Approach" 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 enables LLMs to self-improve general reasoning capabilities by self-synthesizing reasoning paths from abstract to concrete.
• Framework includes: Question input, Ground Truth, Candidate Reasoning Solution, Reasoning Structure, Task Specific Reasoning Guideline, and Final Reasoning Solution.
• Uses filter by... |
Generate 3 bullet points for the "Core Method / Technical Approach" section of an academic poster. | ## Paper Content
# Paper Title
Flexible task abstractions emerge in linear networks with fast and bounded units
## Abstract
Kai Sandbrink*
Exp. Psychology, Oxford Brain Mind Institute, EPFL
Jan P. Bauer*
ELSC, HebrewU
Gatsby Unit, UCL
Alexandra M. Proca*
Department of Computing
Imperial College London
... | • Gates are accelerated by aligned weights; weights are protected by fast gates.
• Closed-form solution for specialization: w̄ = √(1 - (1/2)(τc/τw)(1 - c̄²)) and ē = c² - c¹.
• Shows prediction error, gate protection, and error reduction mechanisms. |
Generate 3 bullet points for the "Experimental Results / Performance Analysis" section of an academic poster. | ## Paper Content
# Paper Title
OAKINK2 : A Dataset of Bimanual Hands-Object Manipulation in Complex Task Completion
## Abstract
We present OAKINK2, a dataset of bimanual object manipulation tasks for complex daily activities. In pursuit of constructing the complex tasks into a structured representation, OAKINK2 int... | • Task-aware Motion Fulfillment (TaMF): Generate human motion trajectory conditioned on task description, object geometries and desired object motion.
• Baseline model: MF-MDM
• The generated hand motion: examples include 'scrape content with tool', 'cut object with tool', 'uncap container lid', etc. |
Generate 3 bullet points for the "Experimental Results / Performance Analysis" section of an academic poster. | ## Paper Content
# Paper Title
EXPLORING THE CAMERA BIAS OF PERSON RE-IDENTIFICATION
## Abstract
We empirically investigate the camera bias of person re-identification (ReID) models. Previously, camera-aware methods have been proposed to address this issue, but they are largely confined to training domains of the m... | • Reports performance on Market-1501, MSMT17, and VeRi-776 datasets.
• Shows camera-agnostic unsupervised and camera-aware unsupervised settings.
• Demonstrates consistent improvements with proposed methods over baselines. |
Generate 3 bullet points for the "Ablation Study" section of an academic poster. | ## Paper Content
# Paper Title
Towards Modern Image Manipulation Localization: A Large-Scale Dataset and Novel Methods
## Abstract
In recent years, image manipulation localization has attracted increasing attention due to its pivotal role in guaranteeing social media security. However, how to accurately identify th... | • Ablation study on training settings: using CASIAv1, NIST16, IMD20 datasets with different combinations.
• Setting (3) with all datasets achieves best performance: 810.848 IoU and 525.590 F1 on PSCC-Net.
• QES threshold analysis shows optimal performance at 0.3–0.5, with 0.3 yielding 3.03 number ratio of kept predicti... |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.