instruction stringclasses 95
values | input stringlengths 141 265k | output stringlengths 15 1.61k |
|---|---|---|
Generate 3 bullet points for the "Research Motivation / Problem Background" section of an academic poster. | ## Paper Content
# Paper Title
UltrAvatar: A Realistic Animatable 3D Avatar Diffusion Model with Authenticity Guided Textures
## Abstract
Recent advances in 3D avatar generation have gained significant attention. These breakthroughs aim to produce more realistic animatable avatars, narrowing the gap between virtual... | • Explore the relationship between self-attention features and lighting effects, propose a diffuse-color-extraction model.
• Propose an authenticity guided diffusion model to generate high-quality PBR textures aligned with 3D meshes.
• Introduce UltrAvatar, a novel framework for generating 3D animatable avatars. |
Generate 3 bullet points for the "Core Method / Technical Approach" section of an academic poster. | ## Paper Content
# Paper Title
Dependency-aware Differentiable Neural Architecture Search
## Method
Differential Architecture Search approaches utilize a continuous relaxation technique to effectively explore the discrete search space for architecture. Initially, within the DARTS search space, every edge $e^{(i,j)... | • We regard architecture weights α of each edge as a random variable from a multivariate normal distribution with learnable weight vector μ and dependency matrix Σ.
• During forward phase, α is sampled from this distribution.
• We prove that the distribution of each α value affects others according to μ and Σ, ensuring... |
Generate 2 bullet points for the "Experimental Results / Performance Analysis" section of an academic poster. | ## Paper Content
# Paper Title
Adaptive Multi-task Learning for Few-shot Object Detection
## Experiments
Our evaluation primarily centers on the detection performance of novel classes, utilizing the FSOD evaluation protocol that is commonly employed in the SOTA
![](images/e55792352597855a9f4c23cb0a71148d57c39bf117... | • Ablation studies on COCO-10shot dataset show the contribution of each component.
• Detection results visualization compares proposed method with baseline across diverse scenes. |
Generate 4 bullet points for the "Core Method / Technical Approach" section of an academic poster. | ## Paper Content
# Paper Title
Rethinking Inductive Biases for Surface Normal Estimation
## Abstract
Despite the growing demand for accurate surface normal estimation models, existing methods use general-purpose dense prediction models, adopting the same inductive biases as other tasks. In this paper, we discuss th... | • Our model generates crisp yet piecewise smooth predictions by learning rotational relationships between nearby pixels.
• We use axis-angle representation and learn e and θ separately.
• For most pixel pairs, θ is 0° or ±90° and e is projected on an edge.
• Inter-pixel relationship can be inferred by assuming symmetry... |
Generate 2 bullet points for the "Method Overview / Framework" section of an academic poster. | ## Paper Content
# Paper Title
Diffperformer: Iterative Learning of Consistent Latent Guidance for Diffusion-based Human Video Generation
## Abstract
Existing diffusion models for pose-guided human video generation mostly suffer from temporal inconsistency in the generated appearance and poses due to the inherent r... | • Introduce an implicit video representation as consistent guidance to enforce the temporal consistency of the diffusion model.
• Present an iterative joint optimization algorithm to integrate the pretrained diffusion model and the implicit video representation in a mutually beneficial way. |
Generate 4 bullet points for the "Background / Related Work" section of an academic poster. | ## Paper Content
# Paper Title
MEET: Towards Memory-Efficient Temporal Sparse Deep Neural Networks
## Abstract
Deep Neural Networks (DNNs) are accurate but compute-intensive, leading to substantial energy consumption during inference. Exploiting temporal redundancy through $\Delta-\Sigma$ convolution [26] in vide... | • The state size of TSNNs often dominates memory usage.
• Storing states in external memory consumes significant energy.
• High accuracy, low sparse computations, reduced memory cost.
• Benefiting from: Sparsity-aware computing, Near-memory computing. |
Generate 4 bullet points for the "Core Method / Technical Approach" section of an academic poster. | ## Paper Content
# Paper Title
Scene-Centric Unsupervised Panoptic Segmentation
## Abstract
Unsupervised panoptic segmentation aims to partition an image into semantically meaningful regions and distinct object instances without training on manually annotated data. In contrast to prior work on unsupervised panoptic... | • CUPS has three stages: (1) Generating panoptic pseudo labels; (2) Bootstrapping a panoptic network with pseudo labels; (3) Self-training the network.
• Stage 1: Pseudo-label generation uses motion clustering, depth-guided semantic segmentation, and alignment to create high-resolution pseudo labels.
• Stage 2: Panopti... |
Generate 3 bullet points for the "Experimental Results / Performance Analysis" section of an academic poster. | ## Paper Content
# Paper Title
VoteFlow: Enforcing Local Rigidity in Self-Supervised Scene Flow
## Abstract
Scene flow estimation aims to recover per-point motion from two adjacent LiDAR scans. However, in real-world applications such as autonomous driving, points rarely move independently of others, especially for... | • VoteFlow achieves state-of-the-art performance on Argoverse2 test set with EPE of 0.289 (mean) and 0.249 (W.V.)
• Cross-domain performance on Waymo val: 0.142 (FD), 0.014 (FS), 0.012 (BS)
• Outperforms baselines including Flow4D, NSFP, and SeFlow |
Generate 3 bullet points for the "Qualitative Results / Visualization" section of an academic poster. | ## Paper Content
# Paper Title
Towards Training-free Anomaly Detection with Vision and Language Foundation Models
## Abstract
Anomaly detection is valuable for real-world applications, such as industrial quality inspection. However, most approaches focus on detecting local structural anomalies while neglecting comp... | • Shows anomaly-free and anomalous images with segmentation masks.
• Demonstrates the model's ability to segment objects based on open-vocabulary prompts.
• Highlights successful detection of anomalies like extra pushpins or wrong cable colors. |
Generate 3 bullet points for the "Core Method / Technical Approach" section of an academic poster. | ## Paper Content
# Paper Title
UNLOCKING THE POWER OF FUNCTION VECTORS FOR CHARACTERIZING AND MITIGATING CATASTROPHIC FORGETTING IN CONTINUAL INSTRUCTION TUNING
## Abstract
Catastrophic forgetting (CF) poses a significant challenge in machine learning, where a model forgets previously learned information upon learn... | • Reformulate function vector hypothesis: P_M(y|x,θ_T) = Σ_{(l,k)∈S} h_{lk} → f_T(y|x).
• Reformulate LLM as latent variable model: P_M(y|x) = ∫_θ P_M(y|θ,x) P_M(θ|x) dθ.
• Hypothesize that forgetting stems from activation of biased model functions, not overwriting. |
Generate 3 bullet points for the "Qualitative Results / Visualization" section of an academic poster. | ## Paper Content
# Paper Title
Noisy Label Learning with Instance-Dependent Outliers: Identifiability via Crowd Wisdom
## Abstract
The generation of label noise is often modeled as a process involving a probability transition matrix (also interpreted as the annotator confusion matrix) imposed onto the label distrib... | • Shows performance of the proposal on CIFAR-10 with synthetic labels against different numbers of annotators.
• Displays examples from ImageNet-15N with low and high sn values.
• Compares nominal vs outlier samples detected by the method. |
Generate 3 bullet points for the "Experimental Results / Performance Analysis" section of an academic poster. | ## Paper Content
# Paper Title
CORE-MPI: Consistency Object Removal with Embedding MultiPlane Image
## Abstract
Novel view synthesis is attractive for social media, but it often contains unwanted details such as personal information that needs to be edited out for a better experience. Multiplane image (MPI) is desi... | • CORE-MPI achieves highest PSNR, SSIM, LPIPS, FID, and lowest L1, L2 on RealEstate10K and UCSD datasets.
• Outperforms Embedding-Base, Embedding-Disparity, and Embedding-Guide across all metrics.
• Quantitative table includes metrics for Pre-inpaint, Post-inpaint, Layer-inpaint, and CORE-MPI. |
Generate 2 bullet points for the "Core Method / Technical Approach" section of an academic poster. | ## Paper Content
# Paper Title
Part-aware Unified Representation of Language and Skeleton for Zero-shot Action Recognition
## Abstract
While remarkable progress has been made on supervised skeleton-based action recognition, the challenge of zero-shot recognition remains relatively unexplored. In this paper, we argu... | • Describes how to generate short textual descriptions for body parts (Head, Hands, Torso, Legs) and motion phases (Start, Middle, End) for given actions.
• Examples include 'Hit another person with something' and 'Shoot at the basket' with corresponding part-based and phase-based descriptions. |
Generate 3 bullet points for the "Experimental Results / Performance Analysis" section of an academic poster. | ## Paper Content
# Paper Title
An Adaptive Screen-Space Meshing Approach for Normal Integration
## Experiments
Edges in screen-space are subject to foreshortening, i.e. appear shorter on screen than in 3D, see Fig. 4a. This foreshortening is exactly described by the first fundamental form $\pmb{I}$ , Eq. (5). We h... | • At higher resolutions, we achieve even more compression.
• Number of vertices grows significantly slower than number of pixels.
• Fewer variables lead to faster linear solver convergence. |
Generate 1 bullet points for the "Background / Related Work" section of an academic poster. | ## Paper Content
# Paper Title
Towards Efficient Replay in Federated Incremental Learning
## Abstract
In Federated Learning (FL), the data in each client is typically assumed fixed or static. However, data often comes in an incremental manner in real-world applications, where the data domain may increase dynamicall... | • Federated Incremental Learning (FIL) enables distributed devices to collaboratively learn novel concepts from shifting data while avoiding knowledge degradation on previously seen classes. |
Generate 4 bullet points for the "Background / Related Work" section of an academic poster. | ## Paper Content
# Paper Title
RNb-NeuS: Reflectance and Normal-based Multi-View 3D Reconstruction
## Abstract
This paper introduces a versatile paradigm for integrating multi-view reflectance (optional) and normal maps acquired through photometric stereo. Our approach employs a pixel-wise joint re-parameterization... | • Photometric Stereo (PS) relies on images under varying lighting to recover high-frequency details via normal maps.
• PS is the only photographic technique that can estimate reflectance.
• PS handles non-Lambertian surfaces and complex illumination.
• State-of-the-art Multi-View Photometric Stereo (MVPS) methods still... |
Generate 2 bullet points for the "Method Overview / Framework" section of an academic poster. | ## Paper Content
# Paper Title
Gaze-LLE: Gaze Target Estimation via Large-Scale Learned Encoders
## Abstract
We address the problem of gaze target estimation, which aims to predict where a person is looking in a scene. Predicting a person's gaze target requires reasoning both about the person's appearance and the c... | • Leverages the power of visual foundation models
• Gaze-LLE consists of a frozen scene encoder and a lightweight gaze decoder with a positional prompting mechanism to predict a specific person's gaze. |
Generate 4 bullet points for the "Core Method / Technical Approach" section of an academic poster. | ## Paper Content
# Paper Title
Link to the Past: Temporal Propagation for Fast 3D Human Reconstruction from Monocular Video
## Abstract
Fast 3D clothed human reconstruction from monocular video remains a significant challenge in computer vision, particularly in balancing computational efficiency with reconstruction... | • Plug-and-play: any pixel-aligned reconstruction that uses the SMPL parametric model as shape prior.
• Volumetric boundary filter: designed to prevent false-positives but also reduces computation by removing improbable coordinates.
• SMPL-based coordinate mapping: compute skinning by propagating SMPL vertices transfor... |
Generate 3 bullet points for the "Background / Related Work" section of an academic poster. | ## Paper Content
# Paper Title
Thompson Sampling For Combinatorial Bandits: Polynomial Regret and Mismatched Sampling Paradox
## Abstract
We consider Thompson Sampling (TS) for linear combinatorial semi-bandits and subgaussian rewards. We propose the first known TS whose finite-time regret does not scale exponentia... | • The function g(t) is chosen to ensure that P(sup_{s≤t} |A^T (μ̂(s) - μ*)| / √(A^T V(s) A) ≥ σ √(2 ln(t) g(t))) ≤ 1 / (t ln(t)^2).
• Where V(t) = diag((N_i(t)^-1)_{i∈[d]}).
• This boosts the proof relies on showing that with high probability: ∀t ∈ [T], ∑_s 1{A^*(s) > A^*} θ(s) > c t^β with constant β > 0. |
Generate 3 bullet points for the "Implementation Details" section of an academic poster. | ## Paper Content
# Paper Title
EmoVIT: Revolutionizing Emotion Insights with Visual Instruction Tuning
## Abstract
Visual Instruction Tuning represents a novel learning paradigm involving the fine-tuning of pre-trained language models using task-specific instructions. This paradigm shows promising zero-shot results... | • Shows sample context types: Caption, Attribute, Categorical, Conversation, Advanced Interaction, Reasoning.
• Example: For a woman singing, model selects emotion 'excitement' and explains reasoning based on attire, action, and context.
• Demonstrates multi-level instruction alignment for emotion understanding. |
Generate 2 bullet points for the "Experimental Results / Performance Analysis" section of an academic poster. | ## Paper Content
# Paper Title
Decoupling Fine Detail and Global Geometry for Compressed Depth Map Super-Resolution
## Abstract
Recovering high-quality depth maps from compressed sources has gained significant attention due to the limitations of consumer-grade depth cameras and the bandwidth restrictions during dat... | • Our method achieves MAE of 0.0322 and RMSE of 0.0739, significantly outperforming all baselines including UniDepth V2 and SGNet.
• Baselines like Depth Anything V2-Large and iDisc show higher errors, confirming the effectiveness of our decoupling strategy. |
Generate 8 bullet points for the "Core Method / Technical Approach" 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. ... | • Tasks have different levels of difficulty and redundancy. Learning some tasks might help in learning others.
• A reference model is finetuned on 5% of randomly selected images-instruction pairs (D_ref).
• On D_ref, each instruction (Y) is split into questions (Q) and responses (R) to compute the Instruction Relevance... |
Generate 3 bullet points for the "Experimental Results / Performance Analysis" section of an academic poster. | ## Paper Content
# Paper Title
LangSplat: 3D Language Gaussian Splatting
## Abstract
Humans live in a 3D world and commonly use natural language to interact with a 3D scene. Modeling a 3D language field to support open-ended language queries in 3D has gained increasing attention recently. This paper introduces Lang... | • Main results show that LangSplat beats SOTA methods by a large margin.
• Ablations show that LangSplat is 199X faster than LERF.
• Visualization results show the effectiveness of LangSplat. |
Generate 3 bullet points for the "Experimental Results / Performance Analysis" section of an academic poster. | ## Paper Content
# Paper Title
Bayesian Evidential Deep Learning for Online Action Detection
## Experiments
THUMOS'14 [31] is a dataset for video-based temporal action localization. We use the validation set with 200 videos for training and test set with 213 videos for evaluation. There are 20 action classes and a ... | • BEDL maintains performance with limited training data.
• Outperforms baselines across THUMOS14-A, THUMOS14-K400, TVSeries-A, and TVSeries-K400 datasets.
• Robust to data scarcity due to Bayesian uncertainty modeling. |
Generate 3 bullet points for the "Core Method / Technical Approach" section of an academic poster. | ## Paper Content
# Paper Title
N2F2: Hierarchical Scene Understanding with Nested Neural Feature Fields
## Method
In this section, we describe Nested Neural Feature Fields (N2F2), an approach to learning multi-scale semantic representations for 3D scenes that uses a form of hierarchical supervision. We first descri... | • Instead of searching for a scale at test time, we use canonical phrases (e.g. object, part) to precompute scale-relevancy γ^3D
• γ^3D = Softmax(max(W₁:dΘᵢ,₁:d)ᵀφ^canon)
• We aggregate information across all scales in a single forward pass → Θ̃ᵢ = [Σγ^3D; Σγ^3D; ...] ⊙ Θᵢ |
Generate 3 bullet points for the "Ablation Study" section of an academic poster. | ## Paper Content
# Paper Title
Communication-Efficient Federated Learning with Accelerated Client Gradient
## Abstract
Federated learning often suffers from slow and unstable convergence due to the heterogeneous characteristics of participating client datasets. Such a tendency is aggravated when the client particip... | • Ablation on server update, accelerated gradient, and local regularization components.
• Removing any component degrades performance, confirming each is essential.
• FedACG (ours) achieves best results across CIFAR-10 and CIFAR-100. |
Generate 3 bullet points for the "Experimental Results / Performance Analysis" section of an academic poster. | ## Paper Content
# Paper Title
No Thing, Nothing: Highlighting Safety-Critical Classes for Robust LiDAR Semantic Segmentation in Adverse Weather
## Abstract
Existing domain generalization methods for LiDAR semantic segmentation under adverse weather struggle to accurately predict "things" categories compared to "st... | • Table compares mIoU scores across methods for various classes (car, bicycle, motorcycle, etc.) on SemanticKITTI → SemanticSTF.
• NTN (Ours) achieves highest scores for critical things classes like car (83.3) and motorcycle (17.2).
• Improvements over SJ+LPD shown in parentheses, e.g., +2.5 for car, +4.1 for motorcycl... |
Generate 1 bullet points for the "Experimental Results / Performance Analysis" section of an academic poster. | ## Paper Content
# Paper Title
MemFlow: Optical Flow Estimation and Prediction with Memory
## Abstract
Optical flow is a classical task that is important to the vision community. Classical optical flow estimation uses two frames as input, whilst some recent methods consider multiple frames to explicitly model long-... | (Top) End-point-error of flow prediction. (Bottom) Next frame prediction on KITTI. |
Generate 3 bullet points for the "Core Method / Technical Approach" section of an academic poster. | ## Paper Content
# Paper Title
SAR3D: Autoregressive 3D Object Generation and Understanding via Multi-scale 3D VQVAE
## Abstract
Autoregressive models have demonstrated remarkable success across various fields, from large language models (LLMs) to large multimodal models (LMMs) and 2D content generation, moving clo... | • Training of VQVAE: Render multi-view RGB-D, encode to multi-scale planes, quantize, decode to 3D triplane, render and calculate loss.
• For 3D Generation: Full tokens train an autoregressive transformer.
• For 3D Understanding: Only 3/8 tokens (first 8 scales out of 10) are used for finetuning LLM. |
Generate 3 bullet points for the "Core Method / Technical Approach" section of an academic poster. | ## Paper Content
# Paper Title
LogiCzSL: Exploring Logic-induced Representation for Compositional Zero-shot Learning
## Abstract
Compositional zero-shot learning (CZSL) aims to recognize unseen attribute-object compositions by learning the primitive concepts (i.e., attribute and object) from the training set. While... | • Example: Shows logic rules for attribute-object compositions like 'Var(dog(x) → animal(x))'.
• Induction: Describes induction rules for logic-based learning.
• Loss: Defines loss functions based on logic rules with mathematical formulas. |
Generate 2 bullet points for the "Research Motivation / Problem Background" section of an academic poster. | ## Paper Content
# Paper Title
UniPose: A Unified Multimodal Framework for Human Pose Comprehension, Generation and Editing
## Abstract
Human pose plays a crucial role in the digital age. While recent works have achieved impressive progress in understanding and generating human poses, they often support only a sing... | • Traditional human pose-related works often operate in isolation.
• Existing MLLMs lack a comprehensive analysis of human poses, particularly in fine-grained part semantics and complex relationships between pose pairs. |
Generate 3 bullet points for the "Method Overview / Framework" section of an academic poster. | ## Paper Content
# Paper Title
Visual Prompting for Generalized Few-shot Segmentation: A Multi-scale Approach
## Abstract
The emergence of attention-based transformer models has led to their extensive use in various tasks, due to their superior generalization and transfer properties. Recent research has demonstrate... | • Prompting at multiple scales through learnable visual prompts with few examples.
• A causal novel-to-base prompt attention mechanism that refines the novel prompts.
• Finetuning the prompts using transduction losses utilizing the unlabelled target images. |
Generate 3 bullet points for the "Experimental Results / Performance Analysis" section of an academic poster. | ## Paper Content
# Paper Title
RoGSplat: Learning Robust Generalizable Human Gaussian Splatting from Sparse Multi-View Images
## Abstract
This paper presents RoGSplat, a novel approach for synthesizing high-fidelity novel views of unseen human from sparse multi-view images, while requiring no cumbersome per-subject... | • Compares RoGSplat against SHERF, NHP, GP-NeRF, and TransHuman on THuman2.0, RenderPeople, and ZJU-MoCap datasets.
• Metrics include PSNR, SSIM, LPIPS, and parameter count.
• Our method achieves top scores on most metrics, e.g., 28.94 PSNR on THuman2.0 and 31.89 PSNR on ZJU-MoCap. |
Generate 3 bullet points for the "Background / Related Work" section of an academic poster. | ## Paper Content
# Paper Title
Advancing Spiking Neural Networks for Sequential Modeling with Central Pattern Generators
## Abstract
Spiking neural networks (SNNs) represent a promising approach to developing artificial neural networks that are both energy-efficient and biologically plausible. However, applying SNN... | • Illustrates the basic spiking neuron model with membrane potential U(t), threshold comparison, and spike generation.
• Includes mathematical formulation for U(t), I(t), and S(t) with sigmoid approximation.
• Shows a plot of the sigmoid surrogate function used for gradient computation. |
Generate 4 bullet points for the "Core Method / Technical Approach" section of an academic poster. | ## Paper Content
# Paper Title
Mani-GS: Gaussian Splating Manipulation with Triangular Mesh
## Abstract
Neural 3D representations, such as Neural Radiation Fields (NeRF), excel at producing photorealistic rendering results but lack the flexibility for manipulation and editing which is crucial for content creation. ... | • Step (1): Extract mesh from 3DGS and NeuS representations.
• Step (2): Apply Gaussian Splatting (GS) binding using triangle-aware parameters (μ, R, s, o, c).
• Step (3): Perform GS manipulation with deformation and rendering updates.
• Pipeline enables controllable manipulation while preserving rendering fidelity. |
Generate 3 bullet points for the "Experimental Results / Performance Analysis" section of an academic poster. | ## Paper Content
# Paper Title
BilevelPruning: Unified Dynamic and Static Channel Pruning for Convolutional Neural Networks
## Abstract
Most existing dynamic or runtime channel pruning methods have to store all weights to achieve efficient inference, which brings extra storage costs. Static pruning methods can redu... | • Compares accuracy, FLOPs reduction, and parameter count across multiple pruning methods on ImageNet.
• BilevelPruning achieves competitive or superior performance with significant parameter reduction.
• Includes results for ResNet-18, ResNet-50, MobileNet-V2, and others with various architectures. |
Generate 3 bullet points for the "Research Motivation / Problem Background" section of an academic poster. | ## Paper Content
# Paper Title
Learning to Control Camera Exposure via Reinforcement Learning
## Abstract
Adjusting camera exposure in arbitrary lighting conditions is the first step to ensure the functionality of computer vision applications. Poorly adjusted camera exposure often leads to critical failure and perf... | • Exposure control finds optimal parameters (Exp Time, Gain) given lighting conditions to achieve desired image quality.
• Improper control causes under/over-exposure, image noise, or motion blur.
• Objective function f(x) is maximized to find optimal exposure settings x*. |
Generate 3 bullet points for the "Core Method / Technical Approach" section of an academic poster. | ## Paper Content
# Paper Title
Pixel-level and Semantic-level Adjustable Super-resolution: A Dual-LoRA Approach
## Abstract
Diffusion prior-based methods have shown impressive results in real-world image super-resolution (SR). However, most existing methods entangle pixel-level and semantic-level SR objectives in t... | • PiSA-SR learns two LoRA modules upon a pre-trained stable-diffusion (SD) model for improved and adjustable SR.
• (a) Pixel-level LoRA is trained using ℓ2 loss.
• (b) Semantic-level LoRA is trained using LPIPS and classifier score distillation (CSD) losses to extract semantic info from pre-trained classification and S... |
Generate 4 bullet points for the "Implementation Details" section of an academic poster. | ## Paper Content
# Paper Title
SDDGR: Stable Diffusion-based Deep Generative Replay for Class Incremental Object Detection
## Abstract
In the field of class incremental learning (CIL), generative replay has become increasingly prominent as a method to mitigate the catastrophic forgetting, alongside the continuous i... | • Details 1: Generation process uses grounding inputs like class names and locations with stable diffusion prompts.
• Details 2: Iterative refinement generates a specific number of images per class, iteratively refining by lowering thresholds.
• Details 3: Pseudo-labeling applies during new dataset training to prevent ... |
Generate 3 bullet points for the "Core Method / Technical Approach" section of an academic poster. | ## Paper Content
# Paper Title
Resolving Scale Ambiguity in Multi-view 3D Reconstruction using Dual-Pixel Sensors
## Method
Our method resolves the scale ambiguity in multi-view 3D reconstruction using multi-view DP images. The overview is illustrated in Fig. 3. Our method infers the unknown scene scale just by cap... | • Focus distance ambiguity: Depth cannot be determined from blur size when focus distance g is unknown.
• Scale ambiguity: Scene scale s is unknown in multi-view reconstruction.
• Formula: b = (lf / (1 - f/g)) * (1/g - 1/z), where l is lens aperture, f is focal length. |
Generate 3 bullet points for the "Method Overview / Framework" section of an academic poster. | ## Paper Content
# Paper Title
Model Poisoning Attacks to Federated Learning via Multi-Round Consistency
## Abstract
Model poisoning attacks are critical security threats to Federated Learning (FL). Existing model poisoning attacks suffer from two key limitations: 1) they achieve suboptimal effectiveness when defen... | • State-of-the-art Attack Effectiveness: PoisonedFL achieves high attack success.
• Breaking Most Defenses with 5% Fake Clients: Effective even with minimal malicious participation.
• Breaking Defenses under Different Non-IID degrees / Local Epochs / Participation Rates: Robust across varying federated learning setting... |
Generate 3 bullet points for the "Implementation Details" section of an academic poster. | ## Paper Content
# Paper Title
Low-Rank Knowledge Decomposition for Medical Foundation Models
## Abstract
The popularity of large-scale pre-training has promoted the development of medical foundation models. However, some studies have shown that although foundation models exhibit strong general feature extraction c... | • LoRKD reduces training and deployment costs via parameter sharing and low-rank decomposition.
• Mathematical formulation shows improved operational efficiency with constraints on rank and parameter budget.
• Achieves 2.531x lower FLOPs compared to baseline while maintaining performance. |
Generate 2 bullet points for the "Core Method / Technical Approach" section of an academic poster. | ## Paper Content
# Paper Title
One Model for ALL: Low-Level Task Interaction Is a Key to Task-Agnostic Image Fusion
## Abstract
Advanced image fusion methods mostly prioritise high-level missions, where task interaction struggles with semantic gaps, requiring complex bridging mechanisms. In contrast, we propose to ... | • Shows inference flow: input images → S-Enc → MM/DP branches → CFGM → G-Dec → fusion result.
• Highlights skip connection and branch-specific processing. |
Generate 3 bullet points for the "Core Method / Technical Approach" 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... | • Multiple manipulation scenarios from the collected object repository.
• Complex tasks with task goals and initial conditions.
• Recipe: mix one slice of apple, three cubes of sugar, and half a bowl of herbal tea, then heat it up. |
Generate 3 bullet points for the "Conclusion / Future Work" section of an academic poster. | ## Paper Content
# Paper Title
"Where am I?" Scene Retrieval with Language
## Conclusion
In conclusion, we present Text2SceneGraphMatcher for language-based scene retrieval. We demonstrate that open-set natural language text-queries are able to retrieve the corresponding scene they describe. We do so by training a ... | • We show the potential for using language and scene graphs for performing scene retrieval.
• We show generalizability to open-set human annotated language queries.
• Our joint embedding model also works in a "retrieval-based" fashion, meaning we can embed our scene graphs in a first step and then retrieve the most sim... |
Generate 3 bullet points for the "Core Method / Technical Approach" section of an academic poster. | ## Paper Content
# Paper Title
Anchor-based Robust Finetuning of Vision-Language Models
## Abstract
We aim at finetuning a vision-language model without hurting its out-of-distribution (OOD) generalization. We address two types of OOD generalization, i.e., i) domain shift such as natural to sketch images, and ii) z... | • Propose Anchor-based Robust Finetuning (ARF) using generated texts and retrieved image-text pairs.
• Uses a captioner to generate text descriptions and a retriever to fetch relevant image-text pairs.
• Regularizes the finetuning process with text-compensated anchor generation and image-text anchor retrieval. |
Generate 3 bullet points for the "Method Overview / Framework" section of an academic poster. | ## Paper Content
# Paper Title
BlockGCN: Redefine Topology Awareness for Skeleton-Based Action Recognition
## Abstract
Graph Convolutional Networks (GCNs) have long set the state-of-the-art in skeleton-based action recognition, leveraging their ability to unravel the complex dynamics of human joint topology through... | • Identifying and restoring the overlooked skeletal topology in advanced GCNs via novel topological encoding schemes.
• Devising BlockGC, an efficient and powerful graph convolutional block.
• Establishing new state-of-the-art performance on standard benchmarks. |
Generate 3 bullet points for the "Experimental Results / Performance Analysis" section of an academic poster. | ## Paper Content
# Paper Title
EmoVIT: Revolutionizing Emotion Insights with Visual Instruction Tuning
## Abstract
Visual Instruction Tuning represents a novel learning paradigm involving the fine-tuning of pre-trained language models using task-specific instructions. This paradigm shows promising zero-shot results... | • Shows accuracy improvements of fine-tuned LLaVA and InstructBLIP models over base models.
• Gains range from +11.34% to +30.12% across datasets including EmoSet, FI, and Emotion6.
• Demonstrates generalizability of the instruction tuning approach across different base models. |
Generate 2 bullet points for the "Other Content" section of an academic poster. | ## Paper Content
# Paper Title
Minimalist Vision with Freeform Pixels
## Task
Section Title: Acknowledgements
Generate poster content for this section. | • This work was supported by the Office of Naval Research and the National Defense Science and Engineering Graduate (NDSEG) fellowship.
• The authors thank Michael Fridberg, Carl Vondrick, and Behzad Kamgar-Parsi for their feedback and support. |
Generate 2 bullet points for the "Qualitative Results / Visualization" section of an academic poster. | ## Paper Content
# Paper Title
BSNet: Box-Supervised Simulation-assisted Mean Teacher for 3D Instance Segmentation
## Abstract
3D instance segmentation (3DIS) is a crucial task, but point-level annotations are tedious in fully supervised settings. Thus, using bounding boxes (bboxes) as annotations has shown great p... | • Shows step-by-step simulation: from no-overlapping objects, to simulate distribution, to apply gravity/collision constraint, to add background noise points.
• Visualizes how simulated samples evolve to resemble real overlapping scenes. |
Generate 2 bullet points for the "Experimental Results / Performance Analysis" section of an academic poster. | ## Paper Content
# Paper Title
An Upload-Efficient Scheme for Transferring Knowledge From a Server-Side Pre-trained Generator to Clients in Heterogeneous Federated Learning
## Abstract
Heterogeneous Federated Learning (HtFL) enables collaborative learning on multiple clients with different model architectures while... | • Upload and download overhead per iteration on Cifar100 with 20 clients.
• FedKTL achieves 0.09M upload, 7.17M download, and 46.94±0.23 accuracy. |
Generate 3 bullet points for the "Experimental Results / Performance Analysis" section of an academic poster. | ## Paper Content
# Paper Title
Prompting Hard or Hardly Prompting: Prompt Inversion for Text-to-Image Diffusion Models
## Abstract
The quality of the prompts provided to text-to-image diffusion models determines how faithful the generated content is to the user's intent, often requiring 'prompt engineering'. To har... | • Qualitative comparison with inverted prompts and corresponding generated images
• Evaluation of the quality of the images generated with the prompts from inversion using CLIP, LPIPS metrics
• Evaluation of prompt quality with BertScore using Precision, Recall, F1 |
Generate 3 bullet points for the "Core Method / Technical Approach" section of an academic poster. | ## Paper Content
# Paper Title
Sat2Scene: 3D Urban Scene Generation from Satellite Images with Diffusion
## Abstract
Directly generating scenes from satellite imagery offers exciting possibilities for integration into applications like games and map services. However, challenges arise from significant view changes ... | • Pipeline includes 2D background generation and 3D foreground generation
• Uses 3D sparse diffusion model with geometry and point cloud inputs
• Feature extractor and renderer produce final rendered images |
Generate 3 bullet points for the "Core Method / Technical Approach" section of an academic poster. | ## Paper Content
# Paper Title
MAPSeg: Unified Unsupervised Domain Adaptation for Heterogeneous Medical Image Segmentation Based on 3D Masked Autoencoding and Pseudo-Labeling
## Abstract
Robust segmentation is critical for deriving quantitative measures from large-scale, multi-center, and longitudinal medical scans... | • MPL is a standalone self-training component for UDA.
• Uses EMA (Exponential Moving Average) for stable pseudo-label generation.
• Detaches pseudo-labels to avoid error propagation. |
Generate 2 bullet points for the "Background / Related Work" section of an academic poster. | ## Paper Content
# Paper Title
Zero-Shot Image Restoration Using Few-Step Guidance of Consistency Models (and Beyond)
## Abstract
In recent years, it has become popular to tackle image restoration tasks with a single pretrained diffusion model (DM) and data-fidelity guidance, instead of training a dedicated deep ne... | • Train DNN to minimize reconstruction error on training set {x_i*, y_i = Ax_i* + e_i}
• Performance drops when test-time observations mismatch training assumptions |
Generate 3 bullet points for the "Experimental Results / Performance Analysis" section of an academic poster. | ## Paper Content
# Paper Title
Semi-Supervised State-Space Model with Dynamic Stacking Filter for Real-World Video Deraining
## Abstract
Significant progress has been made in video restoration under rainy conditions over the past decade, largely propelled by advancements in deep learning. Nevertheless, existing met... | • Proposes a real-world Rainy Video object Detection and Tracking benchmark (RVDT) with 57 long videos.
• Achieves significant improvements in downstream detection and tracking metrics under rainy conditions.
• VDMamba improves mAP and mAP2 scores across multiple classes compared to baselines. |
Generate 3 bullet points for the "Experimental Results / Performance Analysis" section of an academic poster. | ## Paper Content
# Paper Title
Generative Unlearning for Any Identity
## Abstract
Recent advances in generative models trained on large-scale datasets have made it possible to synthesize high-quality samples across various domains. Moreover, the emergence of strong inversion networks enables not only a reconstructi... | • Qualitative Results: Show unlearned faces across Random, In-Domain (FFHQ), and Out-of-Domain (CelebAHO) datasets.
• Quantitative Results: Table comparing ID, FID, and ΔFID scores for Baseline, + extrapolated w_t, + L_adj, and + L_global (GUIDE).
• Erasing Entire Identity: Visual comparison of source vs. unlearned ide... |
Generate 3 bullet points for the "Experimental Results / Performance Analysis" section of an academic poster. | ## Paper Content
# Paper Title
MAtCha Gaussians: Atlas of Charts for High-Quality Geometry and Photorealism From Sparse Views
## Abstract
We present a novel appearance model that simultaneously realizes explicit high-quality 3D surface mesh recovery and photorealistic novel view synthesis from sparse view samples. ... | • MAtCha renders photorealistic images and recovers sharp meshes from just a few unposed images.
• Qualitative comparisons show MAtCha outperforming baselines like 2DGS+MASI3R-SfM and GOF+MASI3R-SfM.
• Quantitative results on DTU dataset show MAtCha achieves the best CD↓ score (1.04). |
Generate 3 bullet points for the "Method Overview / Framework" section of an academic poster. | ## Paper Content
# Paper Title
Compositional Substitutivity of Visual Reasoning for Visual Question Answering
## Introduction
Compositionality is one of the fundamental properties of human cognition argued by Fodor and Pylyshyn [20]. Compositional generalization, the ability of models to generalize to novel composi... | • First to explore compositional substitutivity under multiple SPS types in VQA, critical for evaluating compositional generalization.
• Model-agnostic training framework that improves substitutivity by encouraging the model to identify synonymous primitives.
• GQA-SPS dataset to evaluate VQA models with different type... |
Generate 2 bullet points for the "Method Overview / Framework" section of an academic poster. | ## Paper Content
# Paper Title
Logits DeConfusion with CLIP for Few-Shot Learning
## Abstract
With its powerful visual-language alignment capability, CLIP performs well in zero-shot and few-shot learning tasks. However, we found in experiments that CLIP's logits suffer from serious inter-class confusion problems in... | • Proposes a Logits DeConfusion method that learns and eliminates inter-class confusion in logits by combining Multi-level Adapter Fusion with Inter-Class Deconfusion module.
• Code is available at https://github.com/LiShuo1001/LDC. |
Generate 2 bullet points for the "Experimental Results / Performance Analysis" section of an academic poster. | ## Paper Content
# Paper Title
Physically Compatible 3D Object Modeling from a Single Image
## Abstract
We present a computational framework that transforms single images into 3D physical objects. The visual geometry of a physical object in an image is determined by three orthogonal attributes: mechanical propertie... | • Compared five methods (Wonder3D, LGM, MeshLRM, TripoSR, TetSphere) on metrics: #CC ↓, Mean Stress ↓ (kPa), Standable ↑ (%), Img. Loss ↓.
• Our method consistently improves standability and reduces stress and image loss across all methods. |
Generate 3 bullet points for the "Core Method / Technical Approach" section of an academic poster. | ## Paper Content
# Paper Title
Learning Audio-guided Video Representation with Gated Attention for Video-Text Retrieval
## Abstract
Video-text retrieval, the task of retrieving videos based on a textual query or vice versa, is of paramount importance for video understanding and multimodal information retrieval. Rec... | • Encoders process audio, video frames, and text independently.
• Gated Fusion Transformer fuses video and audio representations.
• Final stage performs video-text alignment with adaptive margin contrastive learning. |
Generate 1 bullet points for the "Core Method / Technical Approach" section of an academic poster. | ## Paper Content
# Paper Title
Ref-GS: Directional Factorization for 2D Gaussian Splatting
## Abstract
In this paper, we introduce Ref-GS, a novel approach for directional light factorization in 2D Gaussian splatting [8], which enables photorealistic view-dependent appearance rendering and precise geometry recovery... | • Proposes a low-rank Tensor Factorization to represent spatio-angular view-dependent effects. |
Generate 2 bullet points for the "Ablation Study" section of an academic poster. | ## Paper Content
# Paper Title
Explain via Any Concept: Concept Bottleneck Model with Open Vocabulary Concepts
## Experiments
In this section, we first demonstrate the performance advantage of our method compared to prior works. Then we show a case study of flexible concept editing, which is a unique capability of ... | • OpenCBM has more compact feature distribution compared to CLIP.
• Precalculated prototype offers meaningful guidance in training. |
Generate 4 bullet points for the "Research Motivation / Problem Background" section of an academic poster. | ## Paper Content
# Paper Title
SkySense-O: Towards Open-World Remote Sensing Interpretation with Vision-Centric Visual-Language Modeling
## Abstract
Open-world interpretation aims to accurately localize and recognize all objects within images by vision-language models (VLMs). While substantial progress has been mad... | • More Accurate Image-Text Alignment: SkySense-O improves alignment over CLIP and existing datasets.
• More Fine-Grained Dataset: Our Sky-SA dataset provides finer-grained annotations compared to existing datasets like Potdam and ISPRS.
• More Open Vocabulary Support: Enables segmentation of novel categories not seen d... |
Generate 5 bullet points for the "Method Overview / Framework" section of an academic poster. | ## Paper Content
# Paper Title
Boosting Vision-Language Models with Transduction
## Abstract
Transduction is a powerful paradigm that leverages the structure of unlabeled data to boost predictive accuracy. We present TransCLIP, a novel and computationally efficient transductive approach designed for Vision-Language... | • We bring the transduction paradigm to vision-language models to improve their performance with TransCLIP.
• We introduce a novel text-based regularization in the form of a Kullback-Leibler divergence term.
• Our method is in line with the blackbox setting, and is applicable on top of popular zero-shot and few-shot mo... |
Generate 2 bullet points for the "Conclusion / Future Work" section of an academic poster. | ## Paper Content
# Paper Title
Gain from Neighbors: Boosting Model Robustness in the Wild via Adversarial Perturbations Toward Neighboring Classes
## Abstract
Recent approaches, such as data augmentation, adversarial training, and transfer learning, have shown potential in addressing the issue of performance degrad... | • Proposed a novel method using gradient-based perturbations targeting neighboring classes and an inter-class distance weighted loss to enhance model robustness in the wild.
• Code is available at: https://github.com/yangzhou321/Gain_from_Neighbors |
Generate 4 bullet points for the "Method Overview / Framework" section of an academic poster. | ## Paper Content
# Paper Title
In2SET: Intra-Inter Similarity Exploiting Transformer for Dual-Camera Compressive Hyperspectral Imaging
## Abstract
Dual-camera compressive hyperspectral imaging (DC-CHI) offers the capability to reconstruct 3D hyperspectral image (HSI) by fusing compressive and panchromatic (PAN) ima... | • Goal: Reconstruct high-quality 3D hyperspectral image (HSI) from CASSI image and panchromatic (PAN) image.
• Motivation: Eliminate hand-crafted priors; exploit PAN guidance for DCCHI reconstruction.
• Key Contributions: Framework using PAN feature pyramid, In2SET module for intra/inter-similarity, improved spatial-sp... |
Generate 3 bullet points for the "Experimental Results / Performance Analysis" section of an academic poster. | ## Paper Content
# Paper Title
Language-Driven Anchors for Zero-Shot Adversarial Robustness
## Abstract
Deep Neural Networks (DNNs) are known to be susceptible to adversarial attacks. Previous researches mainly focus on improving adversarial robustness in the fully supervised setting, leaving the challenging domain... | • Surpassing the previous adversarially robust few-shot methods.
• Establishing the new state-of-the-art of zero-shot adversarial robustness against strong adversarial noises.
• Suggesting that AT in the zero-shot setting could be a promising way to improve practical usefulness. |
Generate 2 bullet points for the "Research Motivation / Problem Background" section of an academic poster. | ## Paper Content
# Paper Title
Generative Multiview Relighting for 3D Reconstruction under Extreme Illumination Variation
## Abstract
Reconstructing the geometry and appearance of objects from photographs taken in different environments is difficult as the illumination and, therefore, the object appearance vary acr... | • Input: collection of internet photos taken under extreme illumination variation
• Output: novel views under consistent illumination |
Generate 2 bullet points for the "Core Method / Technical Approach" section of an academic poster. | ## Paper Content
# Paper Title
Temporal Residual Jacobians for Rig-free Motion Transfer
## Method
Given an unrigged, triangulated mesh of a 3D character, we aim to animate it by motion transfer from an available motion described by relative joint angles (stick figure motion) at each time step. The relative joint an... | • Training: Given motion signatures {M1...Mt} and shape X0, jointly train NJF for per-frame posing and our Temporal Residual Jacobians module to stitch the per-frame mesh predictions across time, supervised with Ground Truth mesh sequences.
• At inference, given the mesh of a target character and motion signatures, we ... |
Generate 3 bullet points for the "Core Method / Technical Approach" section of an academic poster. | ## Paper Content
# Paper Title
MAP: MULTI-HUMAN-VALUE ALIGNMENT PALETTE
## Abstract
Ensuring that generative AI systems align with human values is essential but challenging, especially when considering multiple human values and their potential trade-offs. Since human values can be personalized and dynamically chang... | • Align AI outputs with user-defined 'value palettes' (e.g., ensure average harmlessness exceeds 80% quantile).
• Map palettes to reward functions via constrained optimization.
• Three-step process: Specify palette, check feasibility, align model. |
Generate 5 bullet points for the "Background / Related Work" section of an academic poster. | ## Paper Content
# Paper Title
Learning Non-Linear Invariants for Unsupervised Out-of-Distribution Detection
## Introduction
Deep learning (DL) models can perform remarkably in controlled settings, where samples evaluated come from the same distribution as those seen during training. Unsurprisingly, real-world scen... | • In the supervised setting, a sample is considered OOD if it cannot be assigned to one of the training set classes.
• In the unsupervised setting, we do not know a-priori what classes are present.
• Main idea: Use the invariants in the training set to characterize what should be considered as OOD [1].
• Previous work ... |
Generate 1 bullet points for the "Research Motivation / Problem Background" section of an academic poster. | ## Paper Content
# Paper Title
VAREN: Very Accurate and Realistic Equine Network
## Abstract
Data-driven three-dimensional parametric shape models of the human body have gained enormous popularity both for the analysis of visual data and for the generation of synthetic humans. Following a similar approach for anima... | • Noisy scans with person and holes |
Generate 1 bullet points for the "Qualitative Results / Visualization" section of an academic poster. | ## Paper Content
# Paper Title
Dissecting and Mitigating Diffusion Bias via Mechanistic Interpretability
## Abstract
Diffusion models have demonstrated impressive capabilities in synthesizing diverse content. However, despite their high-quality outputs, these models often perpetuate social biases, including those r... | • Comparison of individual image transformations along the gender axis, with columns displaying images sampled from progressively shifting ratio. (Finding 3) |
Generate 3 bullet points for the "Conclusion / Future Work" section of an academic poster. | ## Paper Content
# Paper Title
APSeg: Auto-Prompt Network for Cross-Domain Few-Shot Semantic Segmentation
## Abstract
Few-shot semantic segmentation (FSS) endeavors to segment unseen classes with only a few labeled samples. Current FSS methods are commonly built on the assumption that their training and application... | • APSeg is an auto-prompt method for guiding SAM to complete CD-FSS tasks.
• Proposes MPG and DPAT modules to achieve fully automatic segmentation in cross-domain scenarios.
• Extensive experiments show APSeg achieves new state-of-the-art in CD-FSS. |
Generate 1 bullet points for the "Experimental Results / Performance Analysis" section of an academic poster. | ## Paper Content
# Paper Title
Cascade Prompt Learning for Vision-Language Model Adaptation
## Experiments
Datasets. For base-to-novel generalization and few-shot experiments, we use 11 datasets following [64, 65]. Specifically, the datasets include ImageNet [4] and Caltech101 [9] for generic objecting, FGVCAircraf... | • CasPL achieves +7.64 HM improvement over CoOp, +4.95 over CoCoOp, +3.41 over MaPLe, and +2.72 over PromptSRC on average across 11 datasets. |
Generate 3 bullet points for the "Core Method / Technical Approach" section of an academic poster. | ## Paper Content
# Paper Title
Gaussian in the Wild: 3D Gaussian Splatting for Unconstrained Image Collections
## Method
Dongbin Zhang*, Chuming Wang*, Weitao Wang, Peihao Li, Minghan Qin, and Haoqian Wang†
Tsinghua Shenzhen International Graduate School, Tsinghua University
![](images/ac24783f6cbc8d4bdf8f50cf757... | • Previous methods share appearance across the whole scene, failing on high-frequency changes.
• We introduce independent appearance features per point via projection and adaptive sampling.
• Our method better captures details like pillars and window colors. |
Generate 6 bullet points for the "Implementation Details" 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... | • Total unique question instances: 700
• Questions with API or textual-context: 300 (42.86%)
• Questions with visual-context: 400 (57.14%)
• Total unique countries: 54; cities: 180
• Maximum textual-context length: 1500; question length: 107
• Average zoom level for visual-context: 15.26 |
Generate 4 bullet points for the "Conclusion / Future Work" section of an academic poster. | ## Paper Content
# Paper Title
AM-RADIO: Agglomerative Vision Foundation Model Reduce All Domains Into One
## Abstract
A handful of visual foundation models (VFMs) have recently emerged as the backbones for numerous downstream tasks. VFMs like CLIP, DINOv2, SAM are trained with distinct objectives, exhibiting uniqu... | • Existing VFMs have major weaknesses in one or more domains
• Distilling from multiple VFMs in different domains creates a broadly powerful unified model
• No need for GT labels
• Targeted teachers can enable improvements in specific domains |
Generate 2 bullet points for the "Experimental Results / Performance Analysis" section of an academic poster. | ## Paper Content
# Paper Title
PanoRecon: Real-Time Panoptic 3D Reconstruction from Monocular Video
## Abstract
We introduce the Panoptic 3D Reconstruction task, a unified and holistic scene understanding task for a monocular video. And we present PanoRecon - a novel framework to address this new task, which realiz... | • Table compares methods (COLMAP, MVDNet, DPSNet, etc.) on metrics: Online, Comp., Acc., Recall, Prec., F-score.
• Ours achieves 8.9 Online, 6.4 Comp., 0.530 Acc., 0.656 Prec., 0.584 F-score. |
Generate 3 bullet points for the "Research Motivation / Problem Background" section of an academic poster. | ## Paper Content
# Paper Title
Lifting Motion to the 3D World via 2D Diffusion
## Abstract
Estimating 3D motion from 2D observations is a longstanding research challenge. Prior work typically requires training on datasets containing ground truth 3D motions, limiting their applicability to activities well-represente... | • Estimating 3D motion from videos usually relies on paired 2D-3D data or motion capture, limiting generalization to out-of-distribution motions.
• MVLift estimates 3D motion in the world coordinate frame from 2D pose sequences without any 3D training data.
• Applicable to human poses, animal poses, and human-object in... |
Generate 4 bullet points for the "Research Motivation / Problem Background" section of an academic poster. | ## Paper Content
# Paper Title
TASTE-Rob: Advancing Video Generation of Task-Oriented Hand-Object Interaction for Generalizable Robotic Manipulation
## Abstract
We address key limitations in existing datasets and models for task-oriented hand-object interaction video generation, a critical approach of generating vi... | • Imitation learning with demonstrations performs well but lacks generalization.
• Generating demonstrations via video generative models is a direct solution.
• Robot data is hard to collect and scale, limiting generalization.
• Question: Can we generate Hand-Object Interaction (HOI) videos as demonstrations for genera... |
Generate 3 bullet points for the "Core Method / Technical Approach" section of an academic poster. | ## Paper Content
# Paper Title
SAMFusion: Sensor-Adaptive Multimodal Fusion for 3D Object Detection in Adverse Weather
## Method
This section describes the early attention fusion schemes of individual sensor features. An illustration of the methodology is shown in Fig. 2b.
In the SAMFusion encoder, early attention... | • Features are blended in the Multi-Modal encoder in an attentive fashion to produce enriched features φ*.
• These are passed to the proposal module where they are refined in the BEV combining gated camera, LiDAR and radar in an adaptive, distance-weighted fashion for initial object proposals.
• The transformer decoder... |
Generate 3 bullet points for the "Method Overview / Framework" section of an academic poster. | ## Paper Content
# Paper Title
SplaTAM: Splat, Track & Map 3D Gaussians for Dense RGB-D SLAM SPLa-tam.github.io
## Abstract
Dense simultaneous localization and mapping (SLAM) is crucial for robotics and augmented reality applications. However, current methods are often hampered by the nonvolumetric or implicit way ... | • SLAM simultaneously estimates camera pose and builds a 3D map
• Visual examples show robot localization and 3D scene reconstruction
• Emphasizes real-time mapping and pose estimation in indoor environments |
Generate 3 bullet points for the "Conclusion / Future Work" section of an academic poster. | ## Paper Content
# Paper Title
MAGMAX: Leveraging Model Merging for Seamless Continual Learning
## Conclusion
In this paper, we introduced MAGMAX, a novel approach to continual learning that leverages model merging via maximum magnitude selection alongside sequential fine-tuning. Our findings underscore the potenti... | • Model merging is a new paradigm of adapting pre-trained models that can be used in CL for large models.
• MagMax learns continually using sequential fine-tuning and maximum magnitude selection to merge task-specific models.
• MagMax outperforms traditional CL methods without requiring training time modifications focu... |
Generate 3 bullet points for the "Ablation Study" section of an academic poster. | ## Paper Content
# Paper Title
Enhanced Contrastive Learning with Multi-view Longitudinal Data for Chest X-ray Report Generation
## Abstract
Automated radiology report generation offers an effective solution to alleviate radiologists' workload. However, most existing methods focus primarily on single or fixed-view ... | • Ablation study on MIMIC-CXR dataset tests contributions of multi-view images, single images, and inclusion of indicators.
• Settings include w/ Ind, w/o Ind, w/o MV, w/o MVL.
• Results show that including multi-view longitudinal data (w/o MVL) yields the best performance. |
Generate 3 bullet points for the "Core Method / Technical Approach" section of an academic poster. | ## Paper Content
# Paper Title
STING-BEE : Towards Vision-Language Model for Real-World X-ray Baggage Security Inspection
## Abstract
Advancements in Computer-Aided Screening (CAS) systems are essential for improving the detection of security threats in X-ray baggage scans. However, current datasets are limited in ... | • STCray Dataset Generation involves collecting diverse X-ray scans with systematic variations in threat type, location, orientation, and concealment, enriched with detailed captions and bounding box annotations.
• The training pipeline includes Multi-task Threat Instruction Tuning using GPT-based captions and bounding... |
Generate 2 bullet points for the "Core Method / Technical Approach" section of an academic poster. | ## Paper Content
# Paper Title
Scaling Proprioceptive-Visual Learning with Heterogeneous Pre-trained Transformers
## Abstract
One of the roadblocks for training generalist robotic models today is heterogeneity. Previous robot learning methods often collect data to train with one specific embodiment for one task, wh... | • Use cross-attention to map arbitrary number/history of camera views or proprioception to same number of tokens
• Enables flexible input handling across embodiments |
Generate 4 bullet points for the "Core Method / Technical Approach" section of an academic poster. | ## Paper Content
# Paper Title
Text-to-3D using Gaussian Splatting
## Abstract
Automatic text-to-3D generation that combines Score Distillation Sampling (SDS) with the optimization of volume rendering has achieved remarkable progress in synthesizing realistic 3D objects. Yet most existing text-to-3D methods by SDS ... | • Optimization strategy includes Geometry Optimization and Appearance refinement.
• Geometry Optimization uses 2D Image Diffusion and 3D Point Cloud Diffusion with Transformer and U-Net components.
• Appearance refinement applies compactness-based densification to add more Gaussians on the surface.
• Initialization fro... |
Generate 4 bullet points for the "Implementation Details" section of an academic poster. | ## Paper Content
# Paper Title
CAT4D: Create Anything in 4D with Multi-View Video Diffusion Models
## Abstract
We present CAT4D, a method for creating 4D (dynamic 3D) scenes from monocular video. CAT4D leverages a multi-view video diffusion model trained on a diverse combination of datasets to enable novel view syn... | • Input views and target views vary by camera motion and scene motion.
• Real data includes CO3D, MVImgNet, Re10K, MC4K, and static-view videos.
• Synthetic data includes Kubric and Objaverse datasets.
• Augmentations include Lumiere static-view videos and CAT3D single images. |
Generate 3 bullet points for the "Research Motivation / Problem Background" section of an academic poster. | ## Paper Content
# Paper Title
TopNet: Transformer-Efficient Occupancy Prediction Network for Octree-Structured Point Cloud Geometry Compression
## Abstract
Efficient Point Cloud Geometry Compression (PCGC) with a lower bits per point (BPP) and higher peak signal-to-noise ratio (PSNR) is essential for the transport... | • Point cloud geometry compression is critical for efficient storage and transmission.
• Octree structures offer hierarchical representation but require efficient occupancy prediction.
• Existing methods struggle with capturing long-range dependencies and local context simultaneously. |
Generate 4 bullet points for the "Research Motivation / Problem Background" section of an academic poster. | ## Paper Content
# Paper Title
nnWNet: Rethinking the Use of Transformers in Biomedical Image Segmentation and Calling for a Unified Evaluation Benchmark
## Abstract
Semantic segmentation is a crucial prerequisite in clinical applications and computer-aided diagnosis. With the development of deep neural networks, b... | • Current biomedical image segmentation models lack a unified evaluation benchmark.
• Significant discrepancies exist in experimental setup, preprocessing, training/validation splits, and evaluation metrics.
• Some models may excel on specific datasets but fail to generalize.
• nnUNet serves as an out-of-the-box soluti... |
Generate 3 bullet points for the "Method Overview / Framework" section of an academic poster. | ## Paper Content
# Paper Title
High-fidelity 3D Object Generation from Single Image with RGBN-Volume Gaussian Reconstruction Model
## Abstract
Recently single-view 3D generation via Gaussian splatt ing has emerged and developed quickly. They learn 3D Gaussians from 2D RGB images generated from pre-trained multi-vie... | • Introduces a novel RGBN-volume Gaussian reconstruction model, called GS-RGBN, to generate high-quality 3D assets from single-view images.
• Proposes a hybrid Voxel-Gaussian model providing a well-structured 3D grid representation for generalizable 3D learning of unstructured Gaussians.
• Designs a cross-volume fusion... |
Generate 4 bullet points for the "Implementation Details" section of an academic poster. | ## Paper Content
# Paper Title
OpticalNet: An Optical Imaging Dataset and Benchmark Beyond the Diffraction Limit
## Abstract
Optical imaging capable of resolving nanoscale features would revolutionize scientific research and engineering applications across biomedicine, smart manufacturing, and semiconductor quality... | • Training samples include building blocks and optical images from scanning electron microscopy (λ/3.6 = 180nm).
• Testing samples include SS (spiral structure) and Light patterns.
• Groundtruth images are provided for supervised training.
• Theoretical simulation data and code are also provided. |
Generate 3 bullet points for the "Experimental Results / Performance Analysis" section of an academic poster. | ## Paper Content
# Paper Title
GOLD: GRAPH OUT-OF-DISTRIBUTION DETECTION VIA IMPLICIT ADVERSARIAL LATENT GENERATION
## Abstract
Despite graph neural networks' (GNNs) great success in modelling graph-structured data, out-of-distribution (OOD) test instances still pose a great challenge for current GNNs. One of the m... | • Shows how energy distributions evolve during adversarial training.
• Illustrates that real OOD energy cannot be effectively separated from ID without adversarial training.
• Demonstrates that GOLD aligns real OOD and generated OOD distributions well. |
Generate 2 bullet points for the "Qualitative Results / Visualization" section of an academic poster. | ## Paper Content
# Paper Title
Compact 3D Scene Representation via Self-Organizing Gaussian Grids
## Experiments
Traditional grid-sorting algorithms are suitable for thousands of items, yet fall short for our application, where even small scenes contain millions of Gaussians across multiple dimensions. To address t... | • Sorted attribute grids of the LEGO scene over the training progress.
• The grid grows as more Gaussians are added. |
Generate 3 bullet points for the "Research Motivation / Problem Background" section of an academic poster. | ## Paper Content
# Paper Title
Self-supervised ControlNet with Spatio-Temporal Mamba for Real-world Video Super-resolution
## Abstract
Existing diffusion-based video super-resolution (VSR) methods are susceptible to introducing complex degradations and noticeable artifacts into high-resolution videos due to their i... | • Unknown complex degradations (e.g., noise, compression) challenge real-world video super-resolution.
• Temporal consistency between adjacent frames is essential for smooth video output.
• SCST addresses both by extracting clean features and modeling spatio-temporal coherence. |
Generate 1 bullet points for the "Qualitative Results / Visualization" section of an academic poster. | ## Paper Content
# Paper Title
One-Shot Open Affordance Learning with Foundation Models
## Abstract
We introduce One-shot Open Affordance Learning (OOAL), where a model is trained with just one example per base object category, but is expected to identify novel objects and affordances. While vision-language models ... | • Demonstrates the model's ability to generalize to novel affordances like 'Sit on', 'Rest on', 'Take a seat', 'Cut', 'Slice', 'Chop' on unseen objects. |
Generate 3 bullet points for the "Core Method / Technical Approach" section of an academic poster. | ## Paper Content
# Paper Title
Gen3DEval: Using vLLMs for Automatic Evaluation of Generated 3D Objects
## Abstract
Rapid advancements in text-to-3D generation require robust and scalable evaluation metrics that align closely with human judgment, a need unmet by current metrics such as PSNR and CLIP, which require g... | • Two-stage training: pre-train (align text and image token space) and finetune (instruction following for pairwise evaluation).
• Uses LiVa architecture trained on curated dataset for pairwise evaluation of generated 3D assets.
• Generates ranking metric output via pairwise comparison of object pairs. |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.