instruction stringclasses 95
values | input stringlengths 141 265k | output stringlengths 15 1.61k |
|---|---|---|
Generate 3 bullet points for the "Experimental Results / Performance Analysis" section of an academic poster. | ## Paper Content
# Paper Title
A Unified Framework for Human-centric Point Cloud Video Understanding
## Abstract
Human-centric Point Cloud Video Understanding (PVU) is an emerging field focused on extracting and interpreting human-related features from sequences of human point clouds, further advancing downstream h... | • UniPVU-Human achieves state-of-the-art performance on multiple benchmarks including 3D pose estimation and action recognition.
• Ablation studies validate the contribution of each component in the network design.
• Semi-supervised settings show effectiveness of self-learning with varying proportions of fine-tuning da... |
Generate 3 bullet points for the "Experimental Results / Performance Analysis" section of an academic poster. | ## Paper Content
# Paper Title
Zero-Shot Blind-spot Image Denoising via Implicit Neural Sampling
## Abstract
The blind-spot principle has been a widely used tool in zero-shot image denoising but faces challenges with real-world noise that exhibits strong local correlations. Existing methods focus on reducing noise ... | • Quantitative comparison across multiple datasets and methods including SIDD Validation, SIDD Benchmark, and FMDD.
• Our method achieves top performance: 35.31/0.868 (SIDD Validation), 35.05/0.922 (SIDD Benchmark), 33.95/0.885 (FMDD).
• Outperforms baselines such as CVF-SID, LUD-VAE, R2R, APBSN, BM3D, DIP, Self2Self, ... |
Generate 3 bullet points for the "Core Method / Technical Approach" section of an academic poster. | ## Paper Content
# Paper Title
pFedMxF: Personalized Federated Class-incremental Learning with Mixture of Frequency Aggregation
## Abstract
Federated learning (FL) has emerged as a promising paradigm for privacy-preserving collaborative machine learning. However, extending FL to class incremental learning settings ... | • The pFedMixF framework consists of three main components: local training with 2D-DFT, frequency aggregation, and global model update.
• Each client learns unique frequency components, and the server aggregates them via inverse 2D DFT.
• The architecture includes a transformer-based feature extractor and an AAC classi... |
Generate 3 bullet points for the "Experimental Results / Performance Analysis" section of an academic poster. | ## Paper Content
# Paper Title
GigaPose: Fast and Robust Novel Object Pose Estimation via One Correspondence
## Abstract
We present GigaPose, a fast, robust, and accurate method for CAD-based novel object pose estimation in RGB images. GigaPose first leverages discriminative "templates", rendered images of the CAD ... | • 3D reconstruction by Wonder3D [3] enables pose estimation from single image
• Quantitative results on Linemod-Occlusion dataset show GigaPose outperforms MegaPose
• Visual predictions show accurate pose estimation even with occlusions |
Generate 3 bullet points for the "Research Motivation / Problem Background" section of an academic poster. | ## Paper Content
# Paper Title
Optimus-2 : Multimodal Minecraft Agent with Goal-Observation-Action Conditioned Policy
## Abstract
Building an agent that can mimic human behavior patterns to accomplish various open-world tasks is a long-term goal. To enable agents to effectively learn behavioral patterns across dive... | • Problem: Existing policies neglect modeling relationships between observations and actions, and between sub-goals and observation-action sequences.
• Idea: Propose Goal-Observation-Action Conditioned Policy (GOAP) with an Action-guided Behavior Encoder and an MLLM to model sub-goal and observation-action relationship... |
Generate 2 bullet points for the "Qualitative Results / Visualization" section of an academic poster. | ## Paper Content
# Paper Title
FlashTex: Fast Relightable Mesh Texturing with LightControlNet
## Experiments
LightControlNet adapts the ControlNet architecture to enable control over the lighting in the generated image. Specifically, we create a conditioning image for a 3D mesh by rendering it using three pre-defin... | • Demonstrates consistent material appearance under canonical and random lighting conditions.
• Highlights realistic specular and diffuse responses across lighting environments. |
Generate 3 bullet points for the "Research Motivation / Problem Background" section of an academic poster. | ## Paper Content
# Paper Title
AI-Face: A Million-Scale Demographically Annotated AI-Generated Face Dataset and Fairness Benchmark
## Abstract
AI-generated faces have enriched human life, such as entertainment, education, and art. However, they also pose misuse risks. Therefore, detecting AI-generated faces becomes... | • Deepfakes are highly realistic AI-generated multimedia that pose threats like misinformation and political manipulation.
• AI-Face is the first million-scale demographically annotated AI face dataset, including real faces, deepfake videos, and faces generated by GANs and Diffusion Models.
• Detection models show bias... |
Generate 2 bullet points for the "Experimental Results / Performance Analysis" section of an academic poster. | ## Paper Content
# Paper Title
An Information Theoretical View for Out-Of-Distribution Detection
## Experiments
In this section, we propose OER to regularize representation learning for OOD detection. Specifically, we tune the temperature to alleviate the over-confidence issue and increase the detection-relevant in... | • Table 1 compares OOD detection and ID classification performance across multiple methods on CIFAR-100 (ID) with ResNet-34.
• Bold numbers indicate superior results; arrows indicate better performance (↓ smaller, ↑ larger). |
Generate 4 bullet points for the "Background / Related Work" section of an academic poster. | ## Paper Content
# Paper Title
ALGM: Adaptive Local-then-Global Token Merging for Efficient Semantic Segmentation with Plain Vision Transformers
## Abstract
This work presents Adaptive Local-then-Global Merging (ALGM), a token reduction method for semantic segmentation networks that use plain Vision Transformers. A... | • Challenge: Identify tokens that can be merged without harming segmentation quality
• Lesson: Tokens of the same class can be merged safely
• Objective: Find a measure to identify same-class tokens locally and globally with minimal overhead
• Findings: Cosine similarity is suitable — locally in early layers, globally ... |
Generate 1 bullet points for the "Ablation Study" section of an academic poster. | ## Paper Content
# Paper Title
Efficient Event-Based Object Detection: A Hybrid Neural Network with Spatial and Temporal Attention
## Abstract
Event cameras offer high temporal resolution and dynamic range with minimal motion blur, making them promising for robust object detection. While Spiking Neural Networks (SN... | • An in-depth ablation study was conducted for each component of the proposed ASAB module, along with various configurations of the hybrid architecture. |
Generate 2 bullet points for the "Background / Related Work" section of an academic poster. | ## Paper Content
# Paper Title
AWOL: Analysis WithOut synthesis using Language
## Introduction
We address the problem of generating new, realistic samples from various 3D shape models using language. The key idea is to relate language (e.g. names of dog breeds or types of trees) to the model's parameters and then l... | • SMAL+ is a multi-species 3D shape model trained on toy scans
• Parameters sampled from normal distribution: β ~ N(0,1) |
Generate 3 bullet points for the "Core Method / Technical Approach" section of an academic poster. | ## Paper Content
# Paper Title
Stochastic Human Motion Prediction with Memory of Action Transition and Action Characteristic
## Abstract
Action-driven stochastic human motion prediction aims to generate future motion sequences of a pre-defined target action based on given past observed sequences performing non-targ... | • Presents the full architecture including Encoder (Enc), MPM, STAB, ACB, and ARM modules.
• Shows input sequences (X, Y) and sampling process with GRU cells and concatenation.
• Highlights the integration of action transition memory (STAB) and action characteristic memory (ACB). |
Generate 3 bullet points for the "Experimental Results / Performance Analysis" section of an academic poster. | ## Paper Content
# Paper Title
Generalizing 6-DoF Grasp Detection via Domain Prior Knowledge
## Abstract
We focus on the generalization ability of the 6-DoF grasp detection method in this paper. While learning-based grasp detection methods can predict grasp poses for unseen objects using the grasp distribution lear... | • Baseline performance without augmentation.
• Augmentation improves performance on Seen and Similar but not on Novel.
• Our method (Ours + Augmentation) achieves the best performance on Seen and Novel categories. |
Generate 3 bullet points for the "Background / Related Work" section of an academic poster. | ## Paper Content
# Paper Title
Diffusion-based Adversarial Purification from the Perspective of the Frequency Domain
## Abstract
The diffusion-based adversarial purification methods attempt to drown adversarial perturbations into a part of isotropic noise through the forward process, and then recover the clean imag... | • Introduces the paradox of diffusion-based adversarial purification.
• Explains that large time-step t damages true semantic information, while small t omits adversarial perturbations.
• References prior work on diffusion models for adversarial purification (Nie et al., ICML 2022). |
Generate 2 bullet points for the "Method Overview / Framework" section of an academic poster. | ## Paper Content
# Paper Title
LabelDistill: Label-guided Cross-modal Knowledge Distillation for Camera-based 3D Object Detection
## Introduction
3D object detection is an essential task in various applications, such as autonomous driving and robotics. In recent years, camera-based methods [35,44, 55, 56] have attr... | • Images lack geometric information while LiDAR point clouds have accurate spatial information.
• Goal: Transfer accurate knowledge from LiDAR model to Image model. |
Generate 3 bullet points for the "Core Method / Technical Approach" section of an academic poster. | ## Paper Content
# Paper Title
Backpropagation-free Network for 3D Test-time Adaptation
## Abstract
Real-world systems often encounter new data over time, which leads to experiencing target domain shifts. Existing Test-Time Adaptation (TTA) methods tend to apply computationally heavy and memory-intensive backpropag... | • Backpropagation-Free: BFTT3D adapts networks to new 3D data at test time without relying on backpropagation.
• Dual-Stream Architecture: Maintains knowledge of the source domain while integrating target-domain-specific information from non-parametric network effectively.
• Entropy-Based Adaptive Fusion: Aligns the tw... |
Generate 1 bullet points for the "Qualitative Results / Visualization" section of an academic poster. | ## Paper Content
# Paper Title
Rethinking Transformers Pre-training for Multi-Spectral Satellite Imagery
## Abstract
Recent advances in unsupervised learning have demonstrated the ability of large vision models to achieve promising results on downstream tasks by pre-training on large amount of unlabelled data. Such... | • SatMAE++ provides better reconstruction results at (H,W), (2H,2W), and (4H,4W) resolutions compared to SatMAE baseline. |
Generate 2 bullet points for the "Qualitative Results / Visualization" 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... | • UniPose: Turn slightly right, straighten right arm and leg, move right hand back, shift left arm forward.
• GPT-4o: Posture shifts from relaxed standing to forward-leaning stance with arm movements and torso tilt. |
Generate 2 bullet points for the "Qualitative Results / Visualization" section of an academic poster. | ## Paper Content
# Paper Title
Visual Fact Checker: Enabling High-Fidelity Detailed Caption Generation
## Abstract
Existing automatic captioning methods for visual content face challenges such as lack of detail, content hallucination, and poor instruction following. In this work, we propose VisualFactChecker (VFC),... | • Qualitative examples show VFC generating detailed captions for 2D images (e.g., a woman on a scooter) and 3D objects (e.g., a stuffed teddy bear).
• VFC mitigates hallucination by fact-checking captions using object detection and VQA tools. |
Generate 4 bullet points for the "Core Method / Technical Approach" section of an academic poster. | ## Paper Content
# Paper Title
AnySat: One Earth Observation Model for Many Resolutions, Scales, and Modalities
## Abstract
Geospatial models must adapt to the diversity of Earth observation data in terms of resolutions, scales, and modalities. However, existing approaches expect fixed input configurations, which l... | • Patch encoder φ^patch works with any patch size.
• Modality combiner φ^comb merges modality tokens.
• JEPA-like scheme reconstructs in latent space.
• 75% weights shared across modalities & scales. |
Generate 3 bullet points for the "Experimental Results / Performance Analysis" section of an academic poster. | ## Paper Content
# Paper Title
Trusted Multi-View Classification with Expert Knowledge Constraints
## Abstract
Trusted multi-view classification (TMVC) based on the Dempster-Shafer theory has gained significant recognition for its reliability in safety-critical applications. However, existing methods predominantly ... | • Sleep Stage Classification table shows TMCEK (Ours) achieving top accuracy and kappa scores across multiple datasets.
• Hypnogram Visualization displays learned Gabor kernels and their importance ranking for different sleep stages.
• Uncertainty Estimation plots compare clean and noisy conditions, showing TMCEK’s rob... |
Generate 2 bullet points for the "Core Method / Technical Approach" section of an academic poster. | ## Paper Content
# Paper Title
MagicQuill: An Intelligent Interactive Image Editing System
## Abstract
As a highly practical application, image editing encounters a variety of user demands and thus prioritizes excellent ease of use. In this paper, we unveil MagicQuill, an integrated image editing system designed to... | • User-friendly interface supports consecutive editing.
• Demonstrates editing workflow from raw image to edited output with parameter controls. |
Generate 3 bullet points for the "Core Method / Technical Approach" section of an academic poster. | ## Paper Content
# Paper Title
ANALOGGENIE: A GENERATIVE ENGINE FOR AUTOMATIC DISCOVERY OF ANALOG CIRCUIT TOPOLOGIES
## Abstract
The massive and large-scale design of foundational semiconductor integrated circuits (ICs) is crucial to sustaining the advancement of many emerging and future technologies, such as gener... | • Represents circuit as graph (Eulerian circuit)
• Uses pretrained decoder-only transformer for graph generation
• Advantages: versatile, scalable, and accurate |
Generate 4 bullet points for the "Conclusion / Future Work" section of an academic poster. | ## Paper Content
# Paper Title
Sim-to-Real Causal Transfer: A Metric Learning Approach to Causally-Aware Interaction Representations
## Abstract
Modeling spatial-temporal interactions among neighboring agents is at the heart of multi-agent problems such as motion forecasting and crowd navigation. Despite notable pr... | • Analyzed and offered an effective approach for causally-aware multi-agent representations.
• Revealed underestimation of indirect causal effects.
• Introduced regularization (contrastive and ranking) using varied causal annotations.
• Approach enables faster learning, better generalization, and stronger transfer, eve... |
Generate 3 bullet points for the "Experimental Results / Performance Analysis" section of an academic poster. | ## Paper Content
# Paper Title
HIRA: PARAMETER-EFFICIENT HADAMARD HIGH-RANK ADAPTATION FOR LARGE LANGUAGE MODELS
## Abstract
We propose Hadamard High-Rank Adaptation (HiRA), a parameter-efficient finetuning (PEFT) method that enhances the adaptability of Large Language Models (LLMs). While Low-rank Adaptation (LoRA... | • Experiments validate HiRA’s effectiveness across multiple tasks.
• Achieves competitive performance with significantly reduced parameter count.
• Maintains efficiency comparable to standard low-rank adaptation methods. |
Generate 4 bullet points for the "Core Method / Technical Approach" section of an academic poster. | ## Paper Content
# Paper Title
Mitigating Background Shift in Class-Incremental Semantic Segmentation
## Method
Class-Incremental Semantic Segmentation (CISS) aims for the model to learn new classes while retaining previously learned knowledge, using supervision only for the novel classes. Basically, the model for ... | • Introduce our method combining SPL, AFD, and Separation loss.
• Selective Pseudo-Labeling (SPL): Selectively assigns ambiguous pixels (lower than threshold) the ignore label, excluding them from learning to mitigate background shift.
• Adaptive Feature Distillation (AFD): Distills reliable features using a reliabilit... |
Generate 7 bullet points for the "Core Method / Technical Approach" section of an academic poster. | ## Paper Content
# Paper Title
Towards Practical Real-Time Neural Video Compression
## Abstract
We introduce a practical real-time neural video codec (NVC) designed to deliver high compression ratio, low latency and broad versatility. In practice, the coding speed of NVCs depends on 1) computational costs, and 2) n... | • We introduce a lean and streamlined pipeline for substantial speedup, resulting in DCVC-RT.
• Accelerating NVC by reducing operational complexity:
• MEMC-free implicit temporal modeling
• Learn latent representation at a single low 1/8 resolution
• Integrating practical techniques into NVC:
• Model Integerization → c... |
Generate 3 bullet points for the "Qualitative Results / Visualization" section of an academic poster. | ## Paper Content
# Paper Title
SG-NeRF: Neural Surface Reconstruction with Scene Graph Optimization
## Experiments
We evaluate the effectiveness of our method through extensive experiments on various datasets, which includes a new inward-facing dataset containing significant outlier camera poses produced by the SfM... | • Shows 3D reconstructions of 7 objects: Baby Bear, Bell, Clock, Deaf Farmer, Pavilion, Sculpture, and others.
• Compares input images with reconstructions from NeuS, Neuralangelo, BARF*, SCNeRF*, L2G-NeRF*, Joint-TensoRF*, PoRF, and SG-NeRF (Ours).
• SG-NeRF reconstructions show sharper details and fewer artifacts. |
Generate 3 bullet points for the "Qualitative Results / Visualization" section of an academic poster. | ## Paper Content
# Paper Title
$\mathbf{M}^{3}$ -VOS: Multi-Phase, Multi-Transition, and Multi-Scenery Video Object Segmentation
## Abstract
Intelligent robots need to interact with diverse objects across various environments. The appearance and state of objects frequently undergo complex transformations depending ... | • Shows qualitative segmentation results for GT, Cutie, SAM2, and Ours on two scenarios: object on table and hand interacting with object.
• Ours produces more accurate and complete masks compared to baselines.
• Highlights superior boundary adherence and object coverage in complex transitions. |
Generate 3 bullet points for the "Core Method / Technical Approach" section of an academic poster. | ## Paper Content
# Paper Title
Mitigating Motion Blur in Neural Radiance Fields with Events and Frames
## Abstract
Neural Radiance Fields (NeRFs) have shown great potential in novel view synthesis. However, they struggle to render sharp images when the data used for training is affected by motion blur. On the other... | • RGB supervision via blur rendering: recover rigid motion of rays and use tensorial fields for faster training.
• Event-based supervision via eCRF: render ΔL via NeRF and use a learnable CRF to model event pixel and non-idealities.
• Model-based prior via EDI: exploit relation between events triggered during exposure ... |
Generate 4 bullet points for the "Core Method / Technical Approach" section of an academic poster. | ## Paper Content
# Paper Title
Semantic Line Combination Detector
## Abstract
A novel algorithm, called semantic line combination detector (SLCD), to find an optimal combination of semantic lines is proposed in this paper. It processes all lines in each line combination at once to assess the overall harmony of the ... | • We extract a feature map F by using ResNet50.
• Semantic feature grouping module groups pixel features into regions.
• Compositional feature extraction generates three types of feature maps: line, region, and positional.
• Score regression estimates composition score; highest score determines optimal combination. |
Generate 1 bullet points for the "Experimental Results / Performance Analysis" section of an academic poster. | ## Paper Content
# Paper Title
No-Regret Learning in Harmonic Games: Extrapolation in the Face of Conflicting Interests
## Abstract
The long-run behavior of multi-agent learning – and, in particular, no-regret learning – is relatively well-understood in potential games, where players have aligned interests. By cont... | • In any harmonic game, for (Lebesgue) almost every initial condition x(0) = Q(y(0)) ∈ X, there exists an increasing sequence of times tn ↑ ∞ such that the corresponding solution orbit x(t) of FTRL fulfills x(tn) → x(0). |
Generate 3 bullet points for the "Core Method / Technical Approach" section of an academic poster. | ## Paper Content
# Paper Title
Conditional Balance: Improving Multi-Conditioning Trade-Offs in Image Generation
## Abstract
Balancing content fidelity and artistic style is a pivotal challenge in image generation. While traditional style transfer methods and modern Denoising Diffusion Probabilistic Models (DDPMs) s... | • Our core idea: different conditionals impact generation parts with varying degrees of importance.
• We analyze the model to identify these sensitivities and inject each conditional only into its most responsive layers.
• This ensures all conditions work in harmony, resulting in a balanced output. |
Generate 3 bullet points for the "Experimental Results / Performance Analysis" section of an academic poster. | ## Paper Content
# Paper Title
NVILA: Efficient Frontier Visual Language Models
## Abstract
Visual language models (VLMs) have made significant advances in accuracy in recent years. However, their efficiency has received much less attention. This paper introduces NVILA, a family of open VLMs designed to optimize bo... | • NVILA outperforms VILA-1.5, LLaVA-OV, InternVL2, and Qwen2-VL across multiple benchmarks including AIZD, ChartQA, and MVBench.
• NVILA achieves 405 GPU days training time for image tasks and 208 for video, significantly lower than LLaVA-OV.
• NVILA reduces pre-filling latency and increases decoding speed compared to ... |
Generate 3 bullet points for the "Experimental Results / Performance Analysis" section of an academic poster. | ## Paper Content
# Paper Title
DI-PCG: Diffusion-based Efficient Inverse Procedural Content Generation for High-quality 3D Asset Creation
## Abstract
Procedural Content Generation (PCG) is powerful in creating high-quality 3D contents, yet controlling it to produce desired shapes is difficult and often requires ext... | • Qualitative Results: Shows diverse 3D assets generated from input images, including chairs, tables, vases.
• Quantitative Results: Compares DI-PCG against baselines on ShapeNet Chairs using CD↓, EMD↓, F-Score↑ metrics.
• Editing Results: Demonstrates editable outputs like changing arm, leg, or height of chairs. |
Generate 2 bullet points for the "Research Motivation / Problem Background" section of an academic poster. | ## Paper Content
# Paper Title
FreeUV: Ground-Truth-Free Realistic Facial UV Texture Recovery via Cross-Assembly Inference Strategy
## Abstract
Recovering high-quality 3D facial textures from single-view 2D images is a challenging task, especially under the constraints of limited data and complex facial details suc... | • Recovering high-quality 3D facial UV textures from single-view 2D images is challenging due to complex details (e.g., wrinkles, makeup, facial hair).
• Current methods rely on costly ground-truth UV data, which is hard to obtain. |
Generate 3 bullet points for the "Ablation Study" section of an academic poster. | ## Paper Content
# Paper Title
Multi-subject Open-set Personalization in Video Generation
## Abstract
Video personalization methods allow us to synthesize videos with specific concepts such as people, pets, and places. However, existing methods often focus on lim
ited domains, require time-consuming optimization p... | • Evaluates impact of CLIP usage, word tokens, and data augmentation.
• Shows that using CLIP improves results compared to no word token or no augmentation.
• Demonstrates Video Alchemist's robustness across ablation conditions. |
Generate 3 bullet points for the "Method Overview / Framework" section of an academic poster. | ## Paper Content
# Paper Title
ADVERSARIAL PERTURBATIONS CANNOT RELIABLY PROTECT ARTISTS FROM GENERATIVE AI
## Abstract
Artists are increasingly concerned about advancements in image generation models that can closely replicate their unique artistic styles. In response, several protection tools against style mimicr... | • Generative models can mimic art styles using public art.
• Protections fail against robust mimicry techniques.
• Adversarial perturbations are ineffective for long-term protection. |
Generate 3 bullet points for the "Research Motivation / Problem Background" section of an academic poster. | ## Paper Content
# Paper Title
Barely Random Algorithms and Collective Metrical Task Systems
## Abstract
We consider metrical task systems on general metric spaces with $n$ points, and show that any fully randomized algorithm can be turned into a randomized algorithm that uses only $2\log n$ random bits, and ac... | • Motivates the problem: when should an agent move between positions to minimize service cost (rent) and movement cost (transport fees)?
• Illustrates with US maps showing rent prices in 1960, 1980, and 2000.
• Defines competitive ratio as cost due to algorithm lacking hindsight, equivalent to multiplicative version of... |
Generate 3 bullet points for the "Core Method / Technical Approach" section of an academic poster. | ## Paper Content
# Paper Title
Customized Condition Controllable Generation for Video Soundtrack
## Abstract
Recent advances in latent diffusion models (LDMs) have enabled data-driven paradigms for video soundtrack generation, improving multimodal alignment capabilities. However, current two-stage frameworks—which ... | • Stage a: Contrastive training is used to independently optimize video generation features for music and sound effects.
• Stage b: Train the denoising process using the Spectrum Divergence Masked Attention module to fuse sound effect and music information.
• Stage c: Customize and refine the musical content of the aud... |
Generate 2 bullet points for the "Core Method / Technical Approach" section of an academic poster. | ## Paper Content
# Paper Title
DiffusionTrack: Point Set Diffusion Model for Visual Object Tracking
## Abstract
Existing Siamese or transformer trackers commonly pose visual object tracking as a one-shot detection problem, i.e., locating the target object in a single forward evaluation scheme. Despite the demonstra... | • We construct a denoising diffusion process, originally proposed for generative image tasks, to infer the target from random hypotheses on the entire frame step by step.
• Algorithm 1 outlines the training algorithm including feature extraction, target initialization, noise construction, and loss computation. |
Generate 4 bullet points for the "Research Motivation / Problem Background" section of an academic poster. | ## Paper Content
# Paper Title
DesignDiffusion: High-Quality Text-to-Design Image Generation with Diffusion Models
## Abstract
In this paper, we present DesignDiffusion, a simple yet effective framework for the novel task of synthesizing design images from textual descriptions. A primary challenge lies in generatin... | • Existing text-to-image models (e.g., SDXL, IF, DALLE-3, FLUX) struggle with layout planning and visual text generation.
• Visual Text Rendering methods separate text and image generation, leading to unnatural or inaccurate results.
• Two-stage process: generate image from text, then insert text — or learn layout in a... |
Generate 1 bullet points for the "Experimental Results / Performance Analysis" section of an academic poster. | ## Paper Content
# Paper Title
PromptKD: Unsupervised Prompt Distillation for Vision-Language Models
## Abstract
Prompt learning has emerged as a valuable technique in enhancing vision-language models (VLMs) such as CLIP for downstream tasks in specific domains. Existing work mainly focuses on designing various lea... | • Comparison of PromptKD with CLIP, PromptSRC, CLIP-PR, UPL, LaFTer, FPL, IFPL, GRIP across Zero-shot, Few-shot, and Unlabeled settings with Base, Novel, HM metrics. |
Generate 5 bullet points for the "Conclusion / Future Work" section of an academic poster. | ## Paper Content
# Paper Title
Aligning with Logic: Measuring, Evaluating and Improving Logical Preference Consistency in Large Language Models
## Abstract
Large Language Models (LLMs) are expected to be predictable and trustworthy to support reliable decision-making systems. Yet current LLMs often show inconsisten... | • We explored the role of logical preference consistency in enhancing reliability and trustworthiness of LLMs.
• Introduced a framework to quantify three key properties: Transitivity, Commutativity, and Negation Invariance.
• Proposed REPAIR, a data refinement and augmentation framework, demonstrating that models train... |
Generate 3 bullet points for the "Ablation Study" section of an academic poster. | ## Paper Content
# Paper Title
Towards Learning a Generalist Model for Embodied Navigation
## Abstract
Building a generalist agent that can interact with the world is the intriguing target of AI systems, thus spurring the research for embodied navigation, where an agent is required to navigate according to instruct... | • Multi-task learning enhances performance across all tasks.
• LLM plays a key role in the method.
• Pre-training on augmented data has limited benefits. |
Generate 2 bullet points for the "Research Motivation / Problem Background" section of an academic poster. | ## Paper Content
# Paper Title
CVE-Bench: A Benchmark for AI Agents' Ability to Exploit Real-World Web Application Vulnerabilities
## Abstract
Large language model (LLM) agents are increasingly capable of autonomously conducting cyberattacks, posing significant threats to existing applications. This growing risk hi... | • CVE-Bench is the first cybersecurity benchmark to:
1. Use real-world vulnerabilities instead of capture-the-flags
2. Focus on critical vulnerabilities
3. Support diverse cyberattacks
• Comparison table with Cybench and Fang et al. showing CVE-Bench supports real-world, critical, and diverse attacks. |
Generate 2 bullet points for the "Background / Related Work" section of an academic poster. | ## Paper Content
# Paper Title
Learn To be Efficient: Build Structured Sparsity in Large Language Models
## Abstract
Large Language Models (LLMs) have achieved remarkable success with their billion-level parameters, yet they incur high inference overheads. The emergence of activation sparsity in LLMs provides a nat... | • Softmax router fails to converge for satisfying performance.
• Demonstrated via accuracy vs. sparsity plots on RoBERTa-base models. |
Generate 4 bullet points for the "Core Method / Technical Approach" section of an academic poster. | ## Paper Content
# Paper Title
MaxFusion: Plug&Play Multi-Modal Generation in Text-to-Image Diffusion Models
## Method
ControlNet [49] performs structural conditioning by using dedicated layers designed to handle different types of inputs, enabling it to process diverse input conditions. Once these layers process t... | • Variance maps across the channels of a diffusion model capture the salient regions within the conditioning inputs.
• This conditioning method is effective for text prompts as well.
• Intermediate features are merged based on their correlation to integrate information from different modalities.
• Variance-based featur... |
Generate 2 bullet points for the "Research Motivation / Problem Background" section of an academic poster. | ## Paper Content
# Paper Title
YOUR MIXTURE-OF-EXPERTS LLM IS SECRETLY AN EMBEDDING MODEL FOR FREE
## Abstract
While large language models (LLMs) excel on generation tasks, their decoder-only architecture often limits their potential as embedding models if no further representation finetuning is applied. Does this ... | • Decoder-only LLMs underperform as embedding models compared to encoders because Hidden State (HS) focuses on next-token prediction, missing input semantics.
• MoE Opportunity: Expert routers naturally capture input semantics through dynamic routing. |
Generate 3 bullet points for the "Qualitative Results / Visualization" section of an academic poster. | ## Paper Content
# Paper Title
Image Restoration by Denoising Diffusion Models with Iteratively Preconditioned Guidance
## Abstract
Training deep neural networks has become a common approach for addressing image restoration problems. An alternative for training a "task-specific" network for each observation model i... | • Visual comparisons on ImageNet, CelebA-HQ, and Sparse-View CT datasets
• Shows Ground Truth, Observed Image, and outputs from DDRM, DPS, DIP, IDPG, and DDPG
• IDPG and DDPG produce sharper, more faithful reconstructions |
Generate 2 bullet points for the "Conclusion / Future Work" section of an academic poster. | ## Paper Content
# Paper Title
Efficient Dataset Distillation via Minimax Diffusion
## Abstract
Dataset distillation reduces the storage and computational consumption of training a network by generating a small surrogate dataset that encapsulates rich information of the original large-scale one. However, previous d... | • We eagerly anticipate exploring further possibilities of using generative models to produce datasets for training.
• Please don’t hesitate to contact us for questions or collaborations! |
Generate 2 bullet points for the "Method Overview / Framework" section of an academic poster. | ## Paper Content
# Paper Title
LINEAR SPHERICAL SLICED OPTIMAL TRANSPORT: A FAST METRIC FOR COMPARING SPHERICAL DATA
## Abstract
Efficient comparison of spherical probability distributions becomes important in fields such as computer vision, geosciences, and medicine. Sliced optimal transport distances, such as sph... | • Proposes embedding spherical distributions into an L² space via Linear Spherical Sliced Optimal Transport (LSSOT).
• LSSOT embedding induces a novel spherical metric and enhances efficiency of group analysis in spherical domains. |
Generate 4 bullet points for the "Conclusion / Future Work" section of an academic poster. | ## Paper Content
# Paper Title
Diff-Palm: Realistic Palmprint Generation with Polynomial Creases and Intra-Class Variation Controllable Diffusion Models
## Abstract
Palmprint recognition is significantly limited by the lack of large-scale publicly available datasets. Previous methods have adopted Bezier curves to s... | • We propose a polynomial-based palm crease representation for generating large-scale datasets resembling real palm creases.
• Our intra-class variation controllable diffusion model enables adjustable intra-class variations.
• This is the first time a palmprint recognition model trained solely on our generated data out... |
Generate 3 bullet points for the "Core Method / Technical Approach" section of an academic poster. | ## Paper Content
# Paper Title
Interpreting Object-level Foundation Models via Visual Precision Search
## Abstract
Advances in multimodal pre-training have propelled object-level foundation models, such as Grounding DINO and Florence-2, in tasks like visual grounding and object detection. However, interpreting thes... | • Proposes Visual Precision Search to interpret object-level foundation models by generating attribution maps via submodular functions.
• Uses Clue Score to locate and identify objects, and Collaboration Score to assess sub-regions with high sensitivity.
• Submodular function combines Clue and Collaboration Scores for ... |
Generate 5 bullet points for the "Other Content" section of an academic poster. | ## Paper Content
# Paper Title
BRIGHT: A REALISTIC AND CHALLENGING BENCHMARK FOR REASONING-INTENSIVE RETRIEVAL
## Abstract
Existing retrieval benchmarks primarily consist of information-seeking queries (e.g., aggregated questions from search engines) where keyword or semantic-based retrieval is usually sufficient. ... | • Provides links to project, paper, code, and data repositories.
• Project: https://brightbenchmark.github.io/
• Paper: https://arxiv.org/pdf/2407.12883
• Code: https://github.com/xlang-ai/BRIGHT
• Data: https://huggingface.co/datasets/xlangai/BRIGHT |
Generate 1 bullet points for the "Core Method / Technical Approach" section of an academic poster. | ## Paper Content
# Paper Title
BiTT: Bi-directional Texture Reconstruction of Interacting Two Hands from a Single Image
## Abstract
Creating personalized hand avatars is important to offer a realistic experience to users on AR / VR platforms. While most prior studies focused on reconstructing 3D hand shapes, some r... | • Given coarse estimated both hands textures, the bi-directional decoding block optimizes the entire hand texture feature using symmetric information. |
Generate 4 bullet points for the "Core Method / Technical Approach" section of an academic poster. | ## Paper Content
# Paper Title
4K4D: Real-Time 4D View Synthesis at 4K Resolution
## Abstract
This paper targets high-fidelity and real-time view synthesis of dynamic 3D scenes at $4K$ resolution. Recent methods on dynamic view synthesis have shown impressive rendering quality. However, their speed is still limit... | • (a) Point Cloud Sequence: 4D feature grid over time
• (b) Geometry: Small network regressing point radius and density using MLPs
• (c) Appearance: Combines image-based rendering with Spherical Harmonics (IBR + SH)
• Intuition: semi-transparent point clouds with density decreasing from center |
Generate 3 bullet points for the "Experimental Results / Performance Analysis" section of an academic poster. | ## Paper Content
# Paper Title
MCCD: Multi-Agent Collaboration-based Compositional Diffusion for Complex Text-to-Image Generation
## Abstract
Diffusion models have shown excellent performance in text-to-image generation. Nevertheless, existing methods often suffer from performance bottlenecks when handling complex ... | • Compares MCCD against baselines (Composable Diffusion, Structured Diffusion, Attn-Exct v2, GORS, DALL-E 2, PixArt-α, SD1.5, SD2.0, SDXL-base) across metrics: Color↑, Shape↑, Texture↑, Spatial↑, Non-Spatial↑, Complex↑.
• MCCD variants (e.g., SD1.5 + MCCD, SD2.0 + MCCD, SDXL-base + MCCD) consistently outperform their b... |
Generate 3 bullet points for the "Ablation Study" section of an academic poster. | ## Paper Content
# Paper Title
AdaDistill: Adaptive Knowledge Distillation for Deep Face Recognition
## Experiments
Training: We use MS1MV2 [14, 19] to train our proposed models for fair comparisons with SOTA KD approaches [24, 27]. MS1MV2 is a refined version of
MS-Celeb-1M [19] by [14], containing 5.8M images of... | • Evaluates impact of KD method, margin penalty type/value, and teacher architecture.
• Shows performance gains with AdaArcDistill variants across multiple datasets.
• Demonstrates robustness to different teacher models (ResNet50, ResNet18, ResNet100, TransFace-B). |
Generate 3 bullet points for the "Core Method / Technical Approach" section of an academic poster. | ## Paper Content
# Paper Title
Channel-wise Noise Scheduled Diffusion for Inverse Rendering in Indoor Scenes
## Abstract
We propose a diffusion-based inverse rendering framework that decomposes a single RGB image into geometry, material, and lighting. Inverse rendering is inherently ill-posed, making it difficult t... | • Our conditional diffusion model generates intrinsic modalities by concatenating them and conditioning on the RGB image.
• We adopt a two-stage approach: first generate low-res outputs, then upsample them.
• We modify the lighting representation from MAIR++ by modeling it with an autoencoder for compact and learnable ... |
Generate 2 bullet points for the "Experimental Results / Performance Analysis" section of an academic poster. | ## Paper Content
# Paper Title
Learn to Memorize and to Forget: A Continual Learning Perspective of Dynamic SLAM
## Experiments
The objective is to guarantee a photorealistic and multi-view consistent map representation by minimizing the photometric and geometric errors in static areas between the rendered images a... | • Quantitative results compare ReFusion, StaticSLAM, DynaSLAM, Co-SLAM, and the proposed method on multiple metrics.
• The proposed method achieves state-of-the-art performance on dynamic SLAM benchmarks. |
Generate 3 bullet points for the "Method Overview / Framework" section of an academic poster. | ## Paper Content
# Paper Title
Integrating Efficient Optimal Transport and Functional Maps For Unsupervised Shape Correspondence Learning
## Abstract
In the realm of computer vision and graphics, accurately establishing correspondences between geometric 3D shapes is pivotal for applications like object tracking, re... | • Presents a novel unsupervised shape correspondence learning framework combining spectral methods (deep functional maps) with efficient optimal transport.
• Proposes adaptive refinement using entropic optimal transport to enhance correspondence performance.
• Excels in non-rigid shape matching (near-isometric and non-... |
Generate 3 bullet points for the "Research Motivation / Problem Background" section of an academic poster. | ## Paper Content
# Paper Title
Learn To be Efficient: Build Structured Sparsity in Large Language Models
## Abstract
Large Language Models (LLMs) have achieved remarkable success with their billion-level parameters, yet they incur high inference overheads. The emergence of activation sparsity in LLMs provides a nat... | • Not all parameters are necessary for model inference.
• Goal: Reduce model size and inference FLOPs while preserving performance.
• Propose efficiency-aware training to make sparsity an optimization goal. |
Generate 4 bullet points for the "Research Motivation / Problem Background" section of an academic poster. | ## Paper Content
# Paper Title
Event-based Head Pose Estimation: Benchmark and Method
## Introduction
Head pose estimation (HPE) has established itself as a crucial task in computer vision, tasked with determining the three-dimensional orientation of a person's head relative to a reference point, typically the came... | • Head pose estimation (HPE) is crucial for applications like human-computer interaction and driver monitoring.
• Traditional RGB-based methods struggle under sudden movement and extreme lighting.
• Event cameras offer high temporal resolution and dynamic range, but lack paired event and head pose data.
• Our work intr... |
Generate 4 bullet points for the "Method Overview / Framework" section of an academic poster. | ## Paper Content
# Paper Title
CMD: A Cross Mechanism Domain Adaptation Dataset for 3D Object Detection
## Introduction
As a crucial component in robotics and autonomous driving systems, 3D object detection has garnered increasing attention from researchers. Due to the inherent
depth of information, point cloud da... | • Cross Mechanism Dataset (CMD): first domain adaptation dataset with iDAR/4D Radar sensors.
• 3D cross-mechanism detection baselines.
• Benchmark results of SOTA domain adaptation algorithms on CMD using Density-Intensity-Geometry (DIG).
• Achieved SOTA in cross-mechanism detection. |
Generate 4 bullet points for the "Method Overview / Framework" section of an academic poster. | ## Paper Content
# Paper Title
Show and Tell: Visually Explainable Deep Neural Nets via Spatially-Aware Concept Bottleneck Models
## Abstract
Modern deep neural networks have now reached human-level performance across a variety of tasks. However, unlike humans they lack the ability to explain their decisions by sho... | • Hybrid: provides both semantic (concept-based) and visual (heatmap-based) explanations
• Online: integrates explainability into the model, no post-hoc computations required
• High performance: improves classification results and provides high-quality heatmaps
• Interactive: enables intuitive exploration of how models... |
Generate 4 bullet points for the "Research Motivation / Problem Background" section of an academic poster. | ## Paper Content
# Paper Title
Caltech Aerial RGB-Thermal Dataset in the Wild
## Introduction
Field robots rely predominantly on visual cameras, lidar, and radar to perceive their surroundings [16, 18]. These sensors provide robust perception capabilities,
but falter in low-light and adverse weather conditions [70... | • Field robots can use thermal cameras to see in low-light conditions, but creating vision-based algorithms remains challenging due to dataset gaps and barriers to data collection.
• Dataset Gaps: Existing thermal datasets focus on urban areas and lack natural environments like forests and rivers, which are crucial for... |
Generate 4 bullet points for the "Background / Related Work" section of an academic poster. | ## Paper Content
# Paper Title
Improving Consistency Models with Generator-Augmented Flows
## Abstract
Consistency models imitate the multi-step sampling of score-based diffusion in a single forward pass of a neural network. They can be learned in two ways: consistency distillation and consistency training. The for... | • Goal: learn fθ: (xt, t) ↦ x.
• CD distills the diffusion ODE in fθ: xt = x + tz, xΦt′ = Φ(xt, t, t′) = xt − ∫t′t vs(xs) ds, LCD(θ) = Lconsistency(θ, xΦt′, xt).
• CT removes the distillation with independent coupling (IC): xt = x + tz, xt̂ = z, xt′ = x + t′z, LCT(θ) = Lconsistency(θ, xt′, xt).
• CT and CD have been as... |
Generate 4 bullet points for the "Experimental Results / Performance Analysis" section of an academic poster. | ## Paper Content
# Paper Title
INTERPRETING EMERGENT PLANNING IN MODEL-FREE REINFORCEMENT LEARNING
## Abstract
We present the first mechanistic evidence that model-free reinforcement learning agents can learn to plan. This is achieved by applying a methodology based on concept-based interpretability to a model-free... | • Study agent with 8x8 cell state matching Sokoban board dimensions.
• Conjecture: agent learns a spatial bijection between cell state and grid.
• Train spatially-local linear probes to predict concept labels from cell state activations.
• 1x1 probes correctly decode future agent and box movements. |
Generate 3 bullet points for the "Core Method / Technical Approach" section of an academic poster. | ## Paper Content
# Paper Title
Efficient Test-Time Adaptation of Vision-Language Models
## Abstract
Test-time adaptation with pre-trained vision-language models has attracted increasing attention for tackling distribution shifts during the test time. Though prior studies have achieved very promising performance, th... | • Positive Cache: holds up to k high-confidence samples per class; pseudo label is one-hot of predicted class.
• Negative Cache: holds up to k samples per class focused on uncertain predictions; pseudo label is negative indicator.
• Impact: Positive cache improves accuracy by integrating reliable samples; negative cach... |
Generate 2 bullet points for the "Experimental Results / Performance Analysis" section of an academic poster. | ## Paper Content
# Paper Title
UniRepLKNet: A Universal Perception Large-Kernel ConvNet for Audio, Video, Point Cloud, Time-Series and Image Recognition
## Abstract
Large-kernel convolutional neural networks (ConvNets) have recently received extensive research attention, but two unresolved and critical issues deman... | • UniRepLKNet-S delivers state-of-the-art performance on Global Temperature and Wind Speed Forecasting.
• Achieves 7.602 MSE and 1.832 MAE for temperature, 3.865 MSE and 1.301 MAE for wind speed. |
Generate 3 bullet points for the "Core Method / Technical Approach" section of an academic poster. | ## Paper Content
# Paper Title
Rethinking Spiking Self-Attention Mechanism: Implementing $\alpha$ -XNOR Similarity Calculation in Spiking Transformers
## Abstract
Transformers significantly raise the performance limits across various tasks, spurring research into integrating them into spiking neural networks. Howe... | • Introduces α-XNOR similarity, adapting XNOR to distinguish spikes/non-spikes by assigning lower weight α (0<α<1) to non-spikes, prioritizing spikes.
• Theoretical analysis shows α-XNOR enhances data representation of spiking similarity scores.
• α-SSA employs α-XNOR similarity for more accurate spiking similarity and... |
Generate 2 bullet points for the "Experimental Results / Performance Analysis" section of an academic poster. | ## Paper Content
# Paper Title
Radio Frequency Ray Tracing with Neural Object Representation for Enhanced RF Modeling
## Abstract
Radio frequency (RF) propagation modeling poses unique electromagnetic simulation challenges. While recent neural representations have shown success in visible spectrum rendering, the fu... | • Object-Level Benchmark: Visual meshes scanned with Polycam are inaccurate for RF. Traditional ray tracing yields 15.7 dB error. RFScape reduces error to 2.9 dB using just ~1 sample/sq.ft.
• Room-Level Benchmark: Outperforms NeRF2 by 5.3 dB on WiFi, 7 dB+ on mmWave. Trains with <1% of NeRF2’s data (1.25 vs. 200 sample... |
Generate 4 bullet points for the "Research Motivation / Problem Background" section of an academic poster. | ## Paper Content
# Paper Title
DemoFusion: Democratising High-Resolution Image Generation With No $$
## Abstract
High-resolution image generation with Generative Artificial Intelligence (GenAI) has immense potential but, due to the enormous capital investment required for training, it is increasingly centralised to... | • Directly prompting SDXL to generate a 2K image.
• SDXL inferences on non-overlapping patches at the original resolution.
• MultiDiffusion fuses multiple overlapping denoising paths to address edge effects.
• DemoFusion achieves global semantic coherence in high-resolution generation. |
Generate 2 bullet points for the "Experimental Results / Performance Analysis" section of an academic poster. | ## Paper Content
# Paper Title
Panda-70M: Captioning 70M Videos with Multiple Cross-Modality Teachers
## Abstract
The quality of the data and annotation upper-bounds the quality of a downstream model. While there exist large text corpora and image-text pairs, high-quality video-text data is much harder to collect. ... | • Panda-70M outperforms Panda-2M and other 2.5M video+image datasets on MSR-VTT and MSRVTT benchmarks.
• Achieves 25.4% B4↑ on MSR-VTT and 32.8% B4↑ on MSRVTT for zero-shot video captioning. |
Generate 3 bullet points for the "Background / Related Work" section of an academic poster. | ## Paper Content
# Paper Title
Position: The Categorization of Race in ML is a Flawed Premise
## Abstract
This position paper critiques the reliance on rigid racial taxonomies in machine learning, exposing their U.S.-centric nature and lack of global applicability—particularly in Europe, where race categories are n... | • Origins of race concept tied to slavery justification and early 20th-century statistics.
• Racial classifications are neither genetically discrete nor scientifically meaningful — they are social constructs.
• In Europe, many countries deny biological races but still prohibit racial discrimination. |
Generate 3 bullet points for the "Background / Related Work" section of an academic poster. | ## Paper Content
# Paper Title
DIFFender: Diffusion-Based Adversarial Defense against Patch Attacks
## Introduction
Deep neural networks are vulnerable to adversarial examples [12, 36], in which imperceptible perturbations are intentionally added to natural examples, leading to incorrect predictions with high confi... | • Uncover the Adversarial Anomaly Perception (AAP) phenomenon in diffusion models.
• AAP leverages distributional discrepancies between adversarial patches and natural images for accurate localization.
• This overcomes the trade-off between patch purification and semantic preservation. |
Generate 1 bullet points for the "Core Method / Technical Approach" section of an academic poster. | ## Paper Content
# Paper Title
MonoPlace3D: Learning 3D-Aware Object Placement for 3D Monocular Detection
## Abstract
Current monocular 3D detectors are held back by the limited diversity and scale of real-world datasets. While data augmentation certainly helps, it's particularly difficult to generate realistic sce... | • Rendering: We render a 3D asset with shadow from a fixed light source, use edge-conditioned ControlNet to generate a realistic car following the same orientation and scale, then combine the shadow, car, and 3D location to create augmented images. |
Generate 3 bullet points for the "Core Method / Technical Approach" section of an academic poster. | ## Paper Content
# Paper Title
Unsupervised Semantic Segmentation Through Depth-Guided Feature Correlation and Sampling
## Abstract
Traditionally, training neural networks to perform semantic segmentation requires expensive human-made annotations. But more recently, advances in the field of unsupervised learning ha... | • Encode images with DINO, then sample features & depth with FPS.
• Feed through S head to create correspondence tensors.
• Calculate loss using L_STEGO+DepthG. |
Generate 4 bullet points for the "Experimental Results / Performance Analysis" section of an academic poster. | ## Paper Content
# Paper Title
Video ReCap: Recursive Captioning of Hour-Long Videos
## Abstract
Most video captioning models are designed to process short video clips of few seconds and output text describing low-level visual concepts (e.g., objects, scenes, atomic actions). However, most real-world videos last fo... | • Evaluate performance using CIDER (C), ROUGE-L (R), and METEOR (M).
• Video ReCap-U achieves 92.67 CIDER, 47.90 ROUGE-L, 28.08 METEOR on clip captioning.
• Outperforms LaViLa on clip captioning by 9.79%.
• Achieves best performance across all three temporal hierarchies with fewer parameters. |
Generate 3 bullet points for the "Research Motivation / Problem Background" section of an academic poster. | ## Paper Content
# Paper Title
BLINK: Multimodal Large Language Models Can See but Not Perceive
## Introduction
Compared to today, computer vision was originally attempting to interpret images as projections of 3D scenes, not just processing 2D arrays of flat "patterns" [25, 59, 62]. In this pursuit, early research... | • Three key features: it includes diverse visual prompts on beyond recognition problems and need “visual” commonsense only.
• Focuses on recognition-based VOA, language-based domain knowledge, and visual commonsense.
• Previous benchmarks focus on text prompts or recognition; BLINK goes beyond with 3D/reflectance estim... |
Generate 3 bullet points for the "Method Overview / Framework" section of an academic poster. | ## Paper Content
# Paper Title
ADU: Adaptive Detection of Unknown Categories in Black-Box Domain Adaptation
## Abstract
Black-box Domain Adaptation (BDA) utilizes a black-box predictor of the source domain to label target domain data, addressing privacy concerns in Unsupervised Domain Adaptation (UDA). However, BDA... | • We propose Selective Amplification Knowledge Distillation (SAKD), a refined knowledge distillation technique for OPBDA and OSBDA scenarios to mitigate noisy pseudo-labels.
• We introduce Entropy-Driven Label Differentiation (EDLD), which categorizes pseudo-labels by quality and applies adaptive entropy-based threshol... |
Generate 3 bullet points for the "Core Method / Technical Approach" section of an academic poster. | ## Paper Content
# Paper Title
Single Mesh Diffusion Models with Field Latents for Texture Generation
## Abstract
We introduce a framework for intrinsic latent diffusion models operating directly on the surfaces of 3D shapes, with the goal of synthesizing high-quality textures. Our approach is underpinned by two co... | • Encodes textured one-ring neighborhoods into vector latent features at each vertex.
• Reconstructs texture values at arbitrary points by decoding coordinate features formed from dot products and determinants of latent vectors with relative position vectors.
• Uses encoder-decoder architecture with geometric operation... |
Generate 3 bullet points for the "Background / Related Work" section of an academic poster. | ## Paper Content
# Paper Title
UNCERTAINTY MODELING IN GRAPH NEURAL NETWORKS VIA STOCHASTIC DIFFERENTIAL EQUATIONS
## Abstract
We propose a novel Stochastic Differential Equation (SDE) framework to address the problem of learning uncertainty-aware representations for graph-structured data. While Graph Neural Ordina... | • A d-dimensional SDE is driven by an m-dimensional Brownian motion.
• Drift and diffusion functions are Lipschitz and satisfy linear growth conditions.
• Ornstein-Uhlenbeck (OU) process is introduced as a specific SDE with mean reversion. |
Generate 1 bullet points for the "Core Method / Technical Approach" section of an academic poster. | ## Paper Content
# Paper Title
HyperDreamBooth: HyperNetworks for Fast Personalization of Text-to-Image Models
## Abstract
Personalization has emerged as a prominent aspect within the field of generative AI, enabling the synthesis of individuals in diverse contexts and styles, while retaining high-fidelity to their... | • Phase 2 - Fast Fine-Tuning: Our approach uses a HyperNetwork prediction step, and then fast finetuning which takes around 20 seconds in order to capture the fine-grained details of a person's identity. |
Generate 4 bullet points for the "Core Method / Technical Approach" 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... | • Proposes a new learning framework subsuming inductive inference as a special case.
• Protocol: Nature selects ground-truth h* ∈ H; learner predicts yt; Nature reveals true label h*(xt).
• Goal: error bound can depend on h*.
• Definition: H is non-uniform online learnable if there exists a deterministic algorithm A wi... |
Generate 2 bullet points for the "Experimental Results / Performance Analysis" section of an academic poster. | ## Paper Content
# Paper Title
Tree- $D$ Fusion: Simulation-Ready Tree Dataset from Single Images with Diffusion Priors
## Experiments
Fig. 5: The input image (a) is reconstructed into a digital twin (b-c) that responds to the envi... | • Tree-D Fusion achieves an average realism score improvement of 44.85%±25.0% over baselines.
• Outperforms all compared methods across all genera in perceived realism. |
Generate 3 bullet points for the "Experimental Results / Performance Analysis" section of an academic poster. | ## Paper Content
# Paper Title
GFlowVLM: Enhancing Multi-step Reasoning in Vision-Language Models with Generative Flow Networks
## Abstract
Vision-Language Models (VLMs) have recently shown promising advancements in sequential decision-making tasks through task-specific fine-tuning. However, common fine-tuning meth... | • Compares GFlowVLM variants against SFT, RL4VLM, and PPO baselines.
• Reports metrics including NL, NL-OOD, BJ, and ablation results.
• Shows GFlowVLM w/ SubTB achieves 100.0 on NL and 7.0 on BJ. |
Generate 4 bullet points for the "Implementation Details" section of an academic poster. | ## Paper Content
# Paper Title
FTBC: Forward Temporal Bias Correction for Optimizing ANN-SNN Conversion
## Method
Spiking Neural Networks (SNNs) often employ integrate-and-fire (IF) neurons, a popular choice in converting Artificial Neural Networks (ANNs) to SNNs. At each time step, these neurons integrate weighted... | • FTBC is evaluated on CIFAR-10/100 and ImageNet datasets.
• VGG16 model pre-trained on CIFAR100 is used for calibration experiments.
• Hyperparameter α = 0.5 is used for bias calibration.
• Accuracy is measured at various timesteps during batch iteration to track convergence. |
Generate 4 bullet points for the "Core Method / Technical Approach" section of an academic poster. | ## Paper Content
# Paper Title
Monocular Occupancy Prediction for Scalable Indoor Scenes
## Method
Problem definition. We focus on the problem of monocular 3D Occupancy Prediction. Specifically, this task takes a single RGB image $\mathbf{I}^{RGB}$ as input and output a voxel-wise occupancy along with semantic ca... | • Uses 2D UNet for feature extraction.
• Depth Branch refines depth using pre-trained model and BCE loss.
• D-FLoSP projects 3D voxel centroids to 2D and multiplies with features for 3D occupancy prediction.
• 3D UNet processes voxel features for final occupancy output. |
Generate 3 bullet points for the "Ablation Study" section of an academic poster. | ## Paper Content
# Paper Title
Sampling Innovation-Based Adaptive Compressive Sensing
## Abstract
Scene-aware Adaptive Compressive Sensing (ACS) has attracted significant interest due to its promising capability for efficient and high-fidelity acquisition of scene images. ACS typically prescribes adaptive sampling ... | • Evaluates adaptive sampling (AS) vs. uniform sampling (US) on BSD68 and Urban100.
• AS consistently outperforms US across all sampling ratios.
• For example, AS achieves 29.54/0.8401 vs. 28.38/0.8207 at SR=0.10 on BSD68. |
Generate 2 bullet points for the "Background / Related Work" section of an academic poster. | ## Paper Content
# Paper Title
TimeCraft: Navigate Weakly-Supervised Temporal Grounded Video Question Answering via Bi-directional Reasoning
## Introduction
Video Question Answering (VQA) has emerged as a critical method for assessing the capabilities of multi-modal models [15, 24, 29], aiming to understand
![](im... | • Lack of supervision for temporal grounding in video QA tasks.
• Difficulty in aligning questions with precise temporal moments in videos. |
Generate 3 bullet points for the "Experimental Results / Performance Analysis" section of an academic poster. | ## Paper Content
# Paper Title
LEARNING AND ALIGNING SINGLE-NEURON INVARIANCE MANIFOLDS IN VISUAL CORTEX
## Abstract
Understanding how sensory neurons exhibit selectivity to certain features and invariance to others is central to uncovering the computational principles underlying robustness and generalization in vi... | • Validated LAMINR on simulated neurons with known ground truth invariance manifolds.
• Introduced intra-group differences via affine transformations.
• Learned activation matrix captured pairwise functional similarity and matched ground-truth groups. |
Generate 3 bullet points for the "Method Overview / Framework" section of an academic poster. | ## Paper Content
# Paper Title
Variance-Based Membership Inference Attacks Against Large-Scale Image Captioning Models
## Abstract
The proliferation of multi-modal generative models has introduced new privacy and security challenges, especially due to the risks of memorization and unintentional disclosure of sensit... | • Propose two black-box MIAs under different attacker knowledge settings using image-only inputs
• MV Metric: A new measure capturing variance in vector embeddings
• Confidence-based training: Refines pseudo-member selection using variance thresholds |
Generate 1 bullet points for the "Research Motivation / Problem Background" section of an academic poster. | ## Paper Content
# Paper Title
Re-HOLD: Video Hand Object Interaction Reenactment via adaptive Layout-instructed Diffusion Model
## Abstract
Current digital human studies focusing on lip-syncing and body movement are no longer sufficient to meet the growing industrial demand, while human video generation techniques... | • Our key insight is to pursue hand-object disentanglement by involving specialized layout representations for hands and objects, respectively. |
Generate 1 bullet points for the "Experimental Results / Performance Analysis" section of an academic poster. | ## Paper Content
# Paper Title
DiN: Diffusion Model for Robust Medical VQA with Semantic Noisy Labels
## Abstract
Medical Visual Question Answering (Med-VQA) systems benefit the interpretation of medical images containing critical clinical information. However, the challenge of noisy labels and limited high-quality... | • Performance comparison of DiN against baselines (MMBERT, Q2ATransformer, SimT, CoDo, SNL-C) under Clean, 10%-Semantic Noise, 20%-Semantic Noise, and 20%-Random Noise conditions on VQA-RAD and PathVQA datasets. |
Generate 3 bullet points for the "Core Method / Technical Approach" section of an academic poster. | ## Paper Content
# Paper Title
SDPT: Synchronous Dual Prompt Tuning for Fusion-based Visual-Language Pre-trained Models
## Method
This section begins with a preliminary of key notations in fusion-based VLPMs, before introducing our SDPT.
Fusion-based VLPMs. In this paper, we use the GLIP [22] models to illustrate ... | • Construct unified prototype tokens within the pre-established cross-attention space, i.e., the fusion space.
• Establish two inverse linear projections which require no training, enabling synchronous mapping of the unified prototype tokens back to the text and image latent spaces.
• Basis: Pre-trained mappings from d... |
Generate 3 bullet points for the "Core Method / Technical Approach" section of an academic poster. | ## Paper Content
# Paper Title
FedSOL: Stabilized Orthogonal Learning with Proximal Restrictions in Federated Learning
## Abstract
Federated Learning (FL) aggregates locally trained models from individual clients to construct a global model. While FL enables learning a model with data privacy, it often suffers from... | • Defines local loss combined with proximal loss: L^k(w_k) = L_local^k(w_k) + β * L_p^k(w_k; w_g).
• FedSOL seeks local gradient orthogonal to the proximal loss gradient.
• Visualizes gradient perturbation and orthogonality in 2D space. |
Generate 3 bullet points for the "Qualitative Results / Visualization" section of an academic poster. | ## Paper Content
# Paper Title
ONDA-Pose: Occlusion-Aware Neural Domain Adaptation for Self-Supervised 6D Object Pose Estimation
## Abstract
Self-supervised 6D object pose estimation has received increasing attention in computer vision recently. Some typical works in literature attempt to translate the synthetic im... | • Visualized results show estimated poses, translated images, and CAD models for various objects.
• Demonstrates effectiveness of ONDA-Pose in handling occlusions and generating accurate pose estimates.
• Includes examples for objects like ape, can, cat, driller, duck, eggbox, glue, and holepuncher. |
Generate 3 bullet points for the "Research Motivation / Problem Background" section of an academic poster. | ## Paper Content
# Paper Title
Relation Rectification in Diffusion Model
## Abstract
Despite their exceptional generative abilities, large T2I diffusion models, much like skilled but careless artists, often struggle with accurately depicting visual relationships between objects. This issue, as we uncover through ca... | • Even the best diffusion models sometimes make mistakes about the direction of relationships.
• Images generated by SDXL based on prompt "A tiger below the desk" often show incorrect spatial relationships.
• Goal: rectify the diffusion model so generated images have the desired direction of relationship. |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.