instruction stringclasses 95
values | input stringlengths 141 265k | output stringlengths 15 1.61k |
|---|---|---|
Generate 2 bullet points for the "Background / Related Work" section of an academic poster. | ## Paper Content
# Paper Title
Monocular Occupancy Prediction for Scalable Indoor Scenes
## Introduction
3D scene understanding is a crucial task in computer vision, becoming increasingly important for applications such as robotic navigation [10], augmented reality [2], and autonomous driving [47]. While humans pos... | • Monocular 3D Semantic Scene Completion (SSC) reconstructs full 3D geometry and semantics from a single RGB image.
• Goal: Predict voxel-wise occupancy without additional 3D inputs. |
Generate 2 bullet points for the "Qualitative Results / Visualization" section of an academic poster. | ## Paper Content
# Paper Title
Osprey: Pixel Understanding with Visual Instruction Tuning
## Abstract
Multimodal large language models (MLLMs) have recently achieved impressive general-purpose vision-language capabilities through visual instruction tuning. However, current MLLMs primarily focus on image-level or bo... | • Box-level referring inherits irrelevant background features and leads to inexact region-text pair alignment on LLM.
• Osprey (mask-level referring) enables precise regional representation, higher resolution for input images, and understanding regions with various granularity. |
Generate 3 bullet points for the "Core Method / Technical Approach" section of an academic poster. | ## Paper Content
# Paper Title
Self-Supervised Audio-Visual Soundscape Stylization
## Method
We propose a self-supervised task that trains a model to stylize input sounds, using audio-only, visual-only, or audio-visual conditioning.
We describe our conditional soundscape stylization model $\mathcal{F}_{\theta}$ ... | • Uses a latent diffusion model (Liu et al., 2023) conditioned on audio-visual clip
• Input spectrogram and target spectrogram are encoded, concatenated, and diffused
• HF-GAN vocoder converts stylized spectrogram to audio |
Generate 2 bullet points for the "Research Motivation / Problem Background" section of an academic poster. | ## Paper Content
# Paper Title
Mind the Gap: Confidence Discrepancy Can Guide Federated Semi-Supervised Learning Across Pseudo-Mismatch
## Abstract
Federated Semi-Supervised Learning (FSSL) aims to leverage unlabeled data across clients with limited labeled data to train a global model with strong generalization ab... | • Observation 1: As heterogeneity intensifies, the pseudo-label predictions of the local model grow more confident, while those of the global model become more conservative.
• Observation 2: The local model exhibits a higher utilization rate of unlabeled data in the early training stages compared to the global model. |
Generate 2 bullet points for the "Experimental Results / Performance Analysis" section of an academic poster. | ## Paper Content
# Paper Title
LEAP-VO: Long-term Effective Any Point Tracking for Visual Odometry
## Abstract
Visual odometry estimates the motion of a moving camera based on visual input. Existing methods, mostly focusing on two-view point tracking, often ignore the rich temporal context in the image sequence, th... | • Quantitative and qualitative comparisons on Replica, MPI Sintel and TartanAir Shibuya datasets
• Our method shows better camera tracking performance |
Generate 3 bullet points for the "Method Overview / Framework" section of an academic poster. | ## Paper Content
## Abstract
Recently, diffusion models have achieved great success in mono-channel audio generation. However, when it comes to stereo audio generation, the soundscapes often have a complex scene of multiple objects and directions. Controlling stereo audio with spatial contexts remains challenging due... | • Introduces a novel framework for generating spoken audio conditioned on natural language descriptions.
• Aims to bridge the gap between text-to-speech and speech-to-text paradigms.
• Proposes a dual-encoder-decoder architecture with cross-modal attention. |
Generate 1 bullet points for the "Experimental Results / Performance Analysis" section of an academic poster. | ## Paper Content
# Paper Title
Diffusion-ES: Gradient-free Planning with Diffusion for Autonomous and Instruction-guided Driving
## Abstract
Diffusion models excel at modeling complex and multimodal trajectory distributions for decision-making and control. Reward-gradient guided denoising has been recently proposed... | • When combined with a reward function similar to PDM-Closed, Diffusion-ES achieves SOTA planning performance in nuPlan |
Generate 2 bullet points for the "Experimental Results / Performance Analysis" section of an academic poster. | ## Paper Content
# Paper Title
Loopy-SLAM: Dense Neural SLAM with Loop Closures
## Abstract
Neural RGBD SLAM techniques have shown promise in dense Simultaneous Localization And Mapping (SLAM), yet face challenges such as error accumulation during camera tracking resulting in distorted maps. In response, we introdu... | • Tracking: Quantitative comparison on ScanNet Dataset across multiple sequences
• Reconstruction: Visual comparison of reconstructed scenes from ESLAM, GO-SLAM, Point-SLAM, Ours, and Ground Truth |
Generate 2 bullet points for the "Core Method / Technical Approach" section of an academic poster. | ## Paper Content
# Paper Title
TO TRUST OR NOT TO TRUST? ENHANCING LARGE LANGUAGE MODELS' SITUATED FAITHFULNESS TO EXTERNAL CONTEXTS
## Abstract
Large Language Models (LLMs) are often augmented with external contexts, such as those used in retrieval-augmented generation (RAG). However, these contexts can be inaccur... | • Rule-based Confidence Reasoning (RCR): extracts explicit confidence signals from the LLM and determines the final answer using predefined rules.
• Self-Guided Confidence Reasoning (SCR): prompts the LLM to self-assess the confidence of external information relative to their own internal knowledge to produce the most ... |
Generate 4 bullet points for the "Experimental Results / Performance Analysis" section of an academic poster. | ## Paper Content
# Paper Title
Evaluating the Adversarial Robustness of Semantic Segmentation: Trying Harder Pays Off
## Experiments
The internal attacks applied in adversarial training used the $\ell_{\infty}$ -norm neighborhood $\Delta = \{\delta : \| \delta \|_{\infty} \leq \epsilon\}$ with $\epsilon = 0.03 ... | • Shows mIoU distributions for SEA models on PASCAL VOC 2012 with background pixels ignored.
• Compares clean vs. robust settings for Tiny and Small object categories.
• Vertical lines indicate CmIoU and NmIoU values.
• Confirms that size-bias persists even when background is excluded. |
Generate 4 bullet points for the "Conclusion / Future Work" section of an academic poster. | ## Paper Content
# Paper Title
Improving Hyperbolic Representations via Gromov-Wasserstein Regularization
## Conclusion
In this paper, we have delved into the integration of the GW distance as a novel regularization term within the realms of hyperbolic neural networks, with a par-
Table 8: F1 (%) result of the nod... | • Integrated GW distance as a novel regularization mechanism within hyperbolic neural networks.
• Focused on leveraging the GM formulation to preserve geometric structures.
• Method validated across diverse datasets including image data for few-shot classification and graph data for link prediction and node classificat... |
Generate 3 bullet points for the "Core Method / Technical Approach" section of an academic poster. | ## Paper Content
# Paper Title
LatentEditor: Text Driven Local Editing of 3D Scenes
## Method
InstructPix2Pix. IP2P [2] edits an input image $I$ based on a textual editing instruction $C_e$ . Leveraging latent diffusion techniques [38] and a Variational Autoencoder (VAE) with an encoder $\mathcal{E}$ and decod... | • Overall pipeline: Initialize NeRF in latent space, refine iteratively with edited latents, and use Delta Module to interpret prompts and produce masks.
• Local Editing: Delta module outputs mask M; noisy latent is generated and edited using DDIM; iterative denoising yields final edited latent.
• Architecture: Include... |
Generate 3 bullet points for the "Core Method / Technical Approach" section of an academic poster. | ## Paper Content
# Paper Title
GRABS: GENERATIVE EMBodied AGENT FOR 3D OBJECT SEGMENTATION WITHOUT SCENE SUPERVISION
## Abstract
We study the hard problem of 3D object segmentation in complex point clouds without requiring human labels of 3D scenes for supervision. By relying on the similarity of pretrained 2D feat... | • Combines Object-centric Network and Multi-object Estimation Network.
• Uses Generative Embodied Agent with Object Discovery and Object Segmentation Branches.
• Leverages Reinforcement Learning and Pseudo Masks for training. |
Generate 3 bullet points for the "Experimental Results / Performance Analysis" section of an academic poster. | ## Paper Content
# Paper Title
DaReNeRF: Direction-aware Representation for Dynamic Scenes
## Abstract
Addressing the intricate challenge of modeling and re- rendering dynamic scenes, most recent approaches have sought to simplify these complexities using plane-based explicit representations, overcoming the slow tr... | • Quantitative comparison on Plenoptic Video 4D dataset.
• Metrics: PSNR, D-SSIM, LPIPS, Training Time, Model Size.
• DaReNeRF-S and DaReNeRF achieve top PSNR (32.10 and 32.26) with reasonable training time and model size. |
Generate 2 bullet points for the "Experimental Results / Performance Analysis" section of an academic poster. | ## Paper Content
# Paper Title
3D Face Reconstruction with the Geometric Guidance of Facial Part Segmentation
## Abstract
3D Morphable Models (3DMMs) provide promising 3D face reconstructions in various applications. However, existing methods struggle to reconstruct faces with extreme expressions due to deficiencie... | • Quantitative comparison of methods on frontal-view and side-view metrics including nose, mouth, forehead, cheek, and average error.
• Visual comparison of reconstructed faces from different methods against input images. |
Generate 2 bullet points for the "Experimental Results / Performance Analysis" section of an academic poster. | ## Paper Content
# Paper Title
The Gaussian Discriminant Variational Autoencoder (GdVAE): A Self-Explainable Model with Counterfactual Explanations
## Experiments
The empirical evaluation aims to validate the performance of our model, focusing on two components: the predictive performance of the GdVAE and the quali... | • Smile classifier shows reduced performance and increased uncertainty in classifying males.
• Highlights potential gender bias in the classifier. |
Generate 2 bullet points for the "Experimental Results / Performance Analysis" section of an academic poster. | ## Paper Content
# Paper Title
RAM-Avatar: Real-time Photo-Realistic Avatar from Monocular Videos with Full-body Control
## Abstract
This paper focuses on advancing the applicability of human avatar learning methods by proposing RAM-Avatar, which learns a Real-time, photo-realistic Avatar that supports full-body co... | • Comparisons for novel in-distribution pose animation against ANR, HFMT, and InstantAvatar.
• Our method generates more realistic appearance and outperforms others for novel in-distribution poses. |
Generate 2 bullet points for the "Research Motivation / Problem Background" section of an academic poster. | ## Paper Content
# Paper Title
Attribute-Missing Multi-view Graph Clustering
## Abstract
The success of existing deep multi-view graph clustering methods is based on the assumption that node attributes are fully available across all views. However, in practical scenarios, node attributes are frequently missing due ... | • Existing methods are often not tailored specifically for clustering tasks and struggle to address missing attributes effectively.
• They tend to ignore the relational dependencies between nodes and their neighboring nodes. |
Generate 3 bullet points for the "Method Overview / Framework" section of an academic poster. | ## Paper Content
# Paper Title
SeMoLi: What Moves Together Belongs Together
## Abstract
We tackle semi-supervised object detection based on motion cues. Recent results suggest that heuristic-based clustering methods in conjunction with object trackers can be used to pseudo-label instances of moving objects and use ... | • Class-agnostically pseudo-label moving objects in unlabeled LiDAR streams for 3D detection.
• Cluster points in LiDAR point cloud based on motion patterns using message passing networks (MPNs).
• Extract 3D bounding boxes to train an off-the-shelf 3D object detector. |
Generate 2 bullet points for the "Core Method / Technical Approach" section of an academic poster. | ## Paper Content
# Paper Title
Active Generalized Category Discovery
## Abstract
Generalized Category Discovery (GCD) is a pragmatic and challenging open-world task, which endeavors to cluster unlabeled samples from both novel and old classes, leveraging some labeled data of old classes. Given that knowledge learne... | • Labeled loss combines contrastive and classification losses for labeled data.
• Unlabeled loss includes contrastive and entropy regularization terms for unlabeled data, with temperature scaling parameters. |
Generate 3 bullet points for the "Method Overview / Framework" section of an academic poster. | ## Paper Content
# Paper Title
Self-training Room Layout Estimation via Geometry-aware Ray-casting
## Introduction
While significant progress has been made in room layout estimation, current state-of-the-art solutions predominantly rely on supervised frameworks, utilizing either monocular panoramic images [9,21,22,... | • Propose a multi-cycle ray-casting algorithm that optimizes noisy layout estimations towards consistent geometry for geometry-aware pseudo-labels.
• Pseudo-labels adapt any layout model to a new data domain without human intervention, using only multi-view images.
• Pseudo-label optimization handles any room geometry ... |
Generate 3 bullet points for the "Core Method / Technical Approach" section of an academic poster. | ## Paper Content
# Paper Title
RENO: Real-Time Neural Compression for 3D LiDAR Point Clouds
## Abstract
Despite the substantial advancements demonstrated by learning-based neural models in the LiDAR Point Cloud Compression (LPCC) task, realizing real-time compression—an indispensable criterion for numerous industri... | • Identifies real-time bottlenecks in preprocessing and neural inference.
• Introduces Fast Occupancy Generator (FOG) and Fast Coordinate Generator (FCG) to minimize preprocessing delays.
• Uses Target Occupancy Predictor (TOP) to optimize neural inference via efficient feature embedding. |
Generate 3 bullet points for the "Experimental Results / Performance Analysis" section of an academic poster. | ## Paper Content
# Paper Title
Adversarily Robust Distillation by Reducing the Student-Teacher Variance Gap
## Experiments
In this section, we provide our experimental settings and compare our STAR-SHIP method with other adversarially robust knowledge distillation approaches.
Datasets. We conduct all the experimen... | • Compares STARSHIP and Ada-STARSHIP against baselines on CIFAR-10 and CIFAR-100 under clean, PGD, CW, and AA attacks.
• STARSHIP achieves state-of-the-art robust accuracy, e.g., 86.47% on CIFAR-10 under PGD attack.
• Ada-STARSHIP further improves performance, especially under strong attacks. |
Generate 2 bullet points for the "Experimental Results / Performance Analysis" section of an academic poster. | ## Paper Content
# Paper Title
Parametric Point Cloud Completion for Polygonal Surface Reconstruction
## Abstract
Existing polygonal surface reconstruction methods heavily depend on input completeness and struggle with incomplete point clouds. We argue that while current point cloud completion techniques may recove... | • PaCo outperforms QEM and RoLoPM in terms of CD, HD, NC metrics while using fewer faces.
• Line charts show PaCo maintains higher NC and lower CD/HD across varying face counts. |
Generate 2 bullet points for the "Method Overview / Framework" section of an academic poster. | ## Paper Content
# Paper Title
UniBind: LLM-Augmented Unified and Balanced Representation Space to Bind Them All
## Abstract
We present UniBind, a flexible and efficient approach that learns a unified representation space for seven diverse modalities - image, text, audio, point cloud, thermal, video, and event data... | • Learn a unified and balanced representation space for multi-modalities (more than three).
• Augment multi-modal learning via the LLM. |
Generate 3 bullet points for the "Qualitative Results / Visualization" section of an academic poster. | ## Paper Content
# Paper Title
PolyOculus: Simultaneous Multi-view Image-based Novel View Synthesis
## Experiments
Experimental Setup. We evaluate on the standard RealEstate10K [64] and Matterport3D [10] datasets. RealEstate10K consists of real-world multi-view image sets derived from videos. The scenes provide ric... | • Standard autoregressive sampling generates stereo views with temporal inconsistency.
• Our grouped method produces temporally consistent stereo pairs (left/right) over time.
• Demonstrated with image grids showing left and right views over time for both methods. |
Generate 1 bullet points for the "Experimental Results / Performance Analysis" section of an academic poster. | ## Paper Content
# Paper Title
Can Machines Understand Composition? Dataset and Benchmark for Photographic Image Composition Embedding and Understanding
## Abstract
With the rapid growth of social media and digital photography, visually appealing images have become essential for effective communication and emotiona... | • PICD achieves the largest dataset scale, greatest diversity in composition categories, broadest scene variety, and highest label quality compared to KUPCP, CADB, and LODB. |
Generate 3 bullet points for the "Experimental Results / Performance Analysis" section of an academic poster. | ## Paper Content
# Paper Title
Snuffy: Efficient Whole Slide Image Classifier
## Experiments
To avoid extensive training on large domain-specific datasets, we propose using continual few-shot self-supervised pre-training with AdaptFormer [7] on ViTs pre-trained on ImageNet-1K [47].
 in log scale.
• Snuffy variants (Snuffy DINO Adapter, Snuffy MAE Adapter) are positioned as efficient with high AUC.
• Includes qualitative WSI visualizations with color-coded tumor regions. |
Generate 3 bullet points for the "Method Overview / Framework" section of an academic poster. | ## Paper Content
## Introduction
Selective attention is an intrinsic property of human perception [15,44,47], enabling people to focus on task-relevant aspects of their surroundings while tuning out the rest [55]. For instance, it allows a driver to focus on traffic signals, road signs, and other vehicles, or an indi... | • Introduces SPARO as a method to enable selective attention in vision transformers.
• Uses human-inspired selective attention to isolate concepts (vehicle, background, person).
• SPARO structures encodings as separately attended concepts for compositional learning. |
Generate 2 bullet points for the "Other Content" section of an academic poster. | ## Paper Content
# Paper Title
Yo'Chameleon:
## Abstract
Large Multimodal Models (e.g., GPT-4, Gemini, Chameleon) have evolved into powerful tools with millions of users. However, they remain generic models and lack personalized knowledge of specific user concepts. Previous work has explored personalization for tex... | • QR code links to paper, project page, and code.
• Examples: Chameleon + Image Prompt, GPT-4o + Image Prompt, Yo'Chameleon descriptions. |
Generate 4 bullet points for the "Research Motivation / Problem Background" section of an academic poster. | ## Paper Content
# Paper Title
Upweighting Easy Samples in Fine-Tuning Mitigates Forgetting
## Abstract
Fine-tuning a pre-trained model on a downstream task often degrades its original capabilities, a phenomenon known as "catastrophic forgetting". This is especially an issue when one does not have access to the dat... | • Standard fine-tuning (SFT) may severely degrade pre-training capabilities.
• SFT on downstream tasks leads to significant drift from the pre-trained model.
• Data-oblivious setting: no access to pre-training data or recipe.
• Common strategy: keep fine-tuned model close to pre-trained model. |
Generate 4 bullet points for the "Implementation Details" section of an academic poster. | ## Paper Content
# Paper Title
Federated Learning with Domain Shift Eraser
## Abstract
Federated learning (FL) is emerging as a promising technique for collaborative learning without local data leaving their devices. However, clients' data originating from diverse domains may degrade model performance due to domain... | • Datasets: DomainNet, Office-Caltech10, PACS.
• Model: AlexNet.
• Metric: Top-1 Accuracy.
• Results: FDSE achieves SOTA performance across all three datasets and produces larger inter-class distances with smaller intra-class distances. |
Generate 2 bullet points for the "Core Method / Technical Approach" section of an academic poster. | ## Paper Content
# Paper Title
DiffBIR: Toward Blind Image Restoration with Generative Diffusion Prior
## Method
In this work, we aim to exploit a powerful generative prior to solve BIR problem. Generative diffusion prior has demonstrated its effectiveness in conditional image generation [70] through enabling condi... | • Low-frequency regions induce higher loss and are more influenced by the high-fidelity condition.
• High-frequency regions are less affected and can maintain more generated content during sampling. |
Generate 4 bullet points for the "Core Method / Technical Approach" section of an academic poster. | ## Paper Content
# Paper Title
Resolution Limit of Single-Photon LiDAR
## Abstract
Single-photon Light Detection and Ranging (LiDAR) systems are often equipped with an array of detectors for improved spatial resolution and sensing speed. However, given a fixed amount of flux produced by the laser transmitter across... | • The energy carried by the reflected signal is modeled as an integral over time.
• The probability of receiving M photons follows a Poisson distribution.
• MLE depth estimation is derived by maximizing the log-likelihood of timestamps.
• Single Pixel Mean Squared Error (MSE) is defined as the expected squared error. |
Generate 5 bullet points for the "Research Motivation / Problem Background" section of an academic poster. | ## Paper Content
# Paper Title
AV2AV: Direct Audio-Visual Speech to Audio-Visual Speech Translation with Unified Audio-Visual Speech Representation
## Abstract
This paper proposes a novel direct Audio-Visual Speech to Audio-Visual Speech Translation (AV2AV) framework, where the input and output of the system are mu... | • The main challenge in AV2AV translation is the absence of parallel AV translation data.
• Our strategy uses unified AV speech representation from the self-supervised AV-HuBERT model, which has modality-agnostic characteristics.
• We employ modality dropout during pre-training to obtain unified AV speech representatio... |
Generate 2 bullet points for the "Implementation Details" section of an academic poster. | ## Paper Content
# Paper Title
A Simple Baseline for Spoken Language to Sign Language Translation with 3D Avatars
## Method
This section details our methodology, which comprises three stages: dictionary construction (Section 3.1), 3D sign estimation (Section 3.2), and Spoken2Sign translation (Section 3.3). Addition... | • 3D keypoint augmentation improves robustness.
• Dual-view SLT enables front and side view generation for better spatial understanding. |
Generate 5 bullet points for the "Core Method / Technical Approach" section of an academic poster. | ## Paper Content
# Paper Title
MP5: A Multi-modal Open-ended Embodied System in Minecraft via Active Perception
## Abstract
It is a long-lasting goal to design an embodied system that can solve long-horizon open-world tasks in human-like ways. However, existing approaches usually struggle with compound difficulties... | • Parser: decomposes a long-horizon task into a sequence of sub-goals.
• Percipient: answers various questions about the observed images.
• Planner: schedules the actions of a sub-goal given the situation.
• Performer: executes the actions with frequent interaction with the env.
• Patroller: checks the responses and ve... |
Generate 2 bullet points for the "Ablation Study" section of an academic poster. | ## Paper Content
# Paper Title
Learning Anomalies with Normality Prior for Unsupervised Video Anomaly Detection
## Experiments
In this section, we first provide experimental details, then draw comparisons with the existing UVAD methods, and finally study different components of our method.
Evaluation Datasets. We ... | • Ablation study shows the contribution of normality propagation and loss re-weighting strategy.
• Pairwise similarity comparison and T/F ablation demonstrate the importance of each component. |
Generate 5 bullet points for the "Research Motivation / Problem Background" section of an academic poster. | ## Paper Content
# Paper Title
TKG-DM: Training-free Chroma Key Content Generation Diffusion Model
## Abstract
Diffusion models have enabled the generation of high-quality images with a strong focus on realism and textual fidelity. Yet, large-scale text-to-image models, such as Stable Diffusion, struggle to generat... | • Clean foreground-background separation is crucial for image editing and video production.
• Existing methods rely heavily on detailed text or annotations, making quick separation difficult.
• Our insight: Attention mechanisms in diffusion models naturally distinguish foreground objects from backgrounds.
• Our solutio... |
Generate 4 bullet points for the "Ablation Study" section of an academic poster. | ## Paper Content
# Paper Title
EntAugment: Entropy-Driven Adaptive Data Augmentation Framework for Image Classification
## Experiments
Comparison with state-of-the-arts We compare our methods with the 11 most representative and commonly used data augmentation methods, including HaS [34], Cutout [9], CutMix [45], Gr... | • Visualization using t-SNE shows EntAugment and EA+EL produce more separable and compact class clusters compared to baseline.
• Convergence analysis shows EntLoss leads to faster and more stable convergence.
• Tradeoff plot shows EntAugment achieves better accuracy-efficiency balance than other methods.
• Ablation on ... |
Generate 3 bullet points for the "Research Motivation / Problem Background" section of an academic poster. | ## Paper Content
# Paper Title
EBS-EKF: Accurate and High Frequency Event-based Star Tracking
## Abstract
Event-based sensors (EBS) are a promising new technology for star tracking due to their low latency and power efficiency, but prior work has thus far been evaluated exclusively in simulation with simplified sig... | • Star trackers determine 3D rotation using star images and typically employ active pixel sensors (APS).
• Event-based sensors (EBS) offer superior temporal resolution, dynamic range, and power efficiency, making their application to star tracking an active area of research.
• Our work proposes a new event-based star t... |
Generate 3 bullet points for the "Other Content" section of an academic poster. | ## Paper Content
# Paper Title
Pathways on the Image Manifold: Image Editing via Video Generation
## Abstract
Recent advances in image editing, driven by image diffusion models, have shown remarkable progress. However, significant challenges remain, as these models often struggle to follow complex edit instructions... | • Unintended camera motion.
• Struggles in out-of-distribution edits.
• Video generation is resource-heavy. |
Generate 3 bullet points for the "Research Motivation / Problem Background" section of an academic poster. | ## Paper Content
# Paper Title
Action-slot: Visual Action-centric Representations for Multi-label Atomic Activity Recognition in Traffic Scenes
## Abstract
In this paper, we study multi-label atomic activity recognition. Despite the notable progress in action recognition, it is still challenging to recognize atomic... | • Videl-level representations are not discriminative.
• Object representations fail due to imperfect detection and incomplete agent involvement.
• Proposes repurposing slot-attention to learn action-centric representations capturing motion and context. |
Generate 3 bullet points for the "Core Method / Technical Approach" section of an academic poster. | ## Paper Content
# Paper Title
IS-FUSION: Instance-Scene Collaborative Fusion for Multimodal 3D Object Detection
## Abstract
Bird's eye view (BEV) representation has emerged as a dominant solution for describing 3D space in autonomous driving scenarios. However, objects in the BEV representation typically exhibit s... | • HSF includes a Point-to-Grid Transformer and a Grid-to-Region Transformer.
• Point-to-Grid Transformer summarizes point features to BEV grid features with multi-head self-attention.
• Grid-to-Region Transformer groups grid features to region features, exploring cross-region interactions. |
Generate 3 bullet points for the "Research Motivation / Problem Background" section of an academic poster. | ## Paper Content
# Paper Title
Personalized Preference Fine-tuning of Diffusion Models
## Abstract
RLHF techniques like DPO can significantly improve the generation quality of text-to-image diffusion models. However, these methods optimize for a single reward that aligns model generation with population-level prefe... | • Alignment methods like DPO improve text-to-image diffusion models by optimizing for population-level preferences.
• However, such methods optimize for a single-reward and cannot capture individual users' preferences.
• We propose a multi-reward optimization objective that aligns diffusion models with personalized pre... |
Generate 4 bullet points for the "Research Motivation / Problem Background" section of an academic poster. | ## Paper Content
# Paper Title
Uncertainty-Instructed Structure Injection for Generalizable HD Map Construction
## Abstract
Reliable high-definition (HD) map construction is crucial for the driving safety of autonomous vehicles. Although recent studies demonstrate improved performance, their generalization capabili... | • The generalization ability across unfamiliar driving scenes of HD map construction remains unexplored.
• Models tend to memorize similar driving scenes rather than capturing the underlying road structures.
• Explicit structural information from PV spaces can be essential for model robustness.
• Map Uncertainty can se... |
Generate 3 bullet points for the "Core Method / Technical Approach" section of an academic poster. | ## Paper Content
# Paper Title
Feature 3DGS: Supercharging 3D Gaussian Splitting to Enable Distilled Feature Fields
## Abstract
3D scene representations have gained immense popularity in recent years. Methods that use Neural Radiance fields are versatile for traditional tasks such as novel view synthesis. In recent... | • Incorporates a semantic feature as an essential attribute on each 3D Gaussian.
• Key innovation is the Parallel N-dimensional Gaussian Rasterizer, complemented by an optional convolutional speed-up module.
• This approach efficiently renders arbitrary high-dimensional features without compromising downstream performa... |
Generate 2 bullet points for the "Background / Related Work" section of an academic poster. | ## Paper Content
# Paper Title
Unifying Top-down and Bottom-up Scanpath Prediction Using Transformers
## Abstract
Most models of visual attention aim at predicting either top-down or bottom-up control, as studied using different visual search and free-viewing tasks. In this paper we propose the Human Attention Tran... | • Predicting human fixation scanpaths requires spatio-temporal understanding of fixed image contents and their relationship to external goals.
• Requires capturing both low-level features and high-level semantics of the input image. |
Generate 4 bullet points for the "Core Method / Technical Approach" section of an academic poster. | ## Paper Content
# Paper Title
Coherent Temporal Synthesis for Incremental Action Segmentation
## Abstract
Data replay is a successful incremental learning technique for images. It prevents catastrophic forgetting by keeping a reservoir of previous data, original or synthesized, to ensure the model retains past kno... | • Temporally Coherent Action Modeling: conditional VAE conditioned on one-hot action label and temporal coherence variable
• Replay Data Generation: sample symbolic action sequences, generate temporally coherent segments, concatenate to form videos
• Incremental Training: construct replay data with past task generators... |
Generate 3 bullet points for the "Background / Related Work" section of an academic poster. | ## Paper Content
# Paper Title
The STVchrono Dataset: Towards Continuous Change Recognition in Time
## Abstract
Recognizing continuous changes offers valuable insights into past historical events, supports current trend analysis, and facilitates future planning. This knowledge is crucial for a variety of fields, su... | • Subjects include weather, building, tree, road, lawn/grassland, soil/sand, river, road fence, human, animal.
• Human annotators label change contents: distinction changes, tendency, superlative, similarity.
• Tasks: Continual Change Captioning (Image Pair), Continual Change Captioning (Image Sequence), Change-Aware S... |
Generate 3 bullet points for the "Qualitative Results / Visualization" section of an academic poster. | ## Paper Content
# Paper Title
UNION-OVER-INTERSECTIONS: OBJECT DETECTION BEYOND WINNER-TAKES-ALL
## Abstract
This paper revisits the problem of predicting box locations in object detection architectures. Typically, each box proposal or box query aims to directly maximize the intersection-over-union score with the ... | • Detection Task: Success — UoI corrects box stretch via union of intersecting boxes; Failure — in crowded scenes, merges separate instances.
• Segmentation Task: Works for segmentation as well; Fails when too much overlap between same class instances.
• Plug-and-play approach: UoI integrates easily with existing pipel... |
Generate 4 bullet points for the "Research Motivation / Problem Background" section of an academic poster. | ## Paper Content
# Paper Title
Pre-trained Model Guided Fine-Tuning for Zero-Shot Adversarial Robustness
## Abstract
Large-scale pre-trained vision-language models like CLIP have demonstrated impressive performance across various tasks, and exhibit remarkable zero-shot generalization capability, while they are also... | • Defines zero-shot adversarial robustness task on CLIP.
• Introduces notation for input image, text, CLIP encoders, and adversarial budget.
• Formulates adversarial attack generation using perturbation and loss function.
• Shows performance of current methods with bar charts comparing robustness. |
Generate 2 bullet points for the "Method Overview / Framework" section of an academic poster. | ## Paper Content
# Paper Title
PAV: Personalized Head Avatar from Unstructured Video Collection
## Introduction
Digital human avatars enabling the editing of facial expressions and head motion have broad applications across telepresence, animation, and digital content creation. The demands for personalized human av... | • Given monocular talking face videos of the same character, PAV learns a unified model for dynamic deformable neural radiance field (NeRF) controlled by face blendshapes and appearance embeddings.
• Contributions include a framework for learning controllable head avatars from unconstrained short video collections, syn... |
Generate 4 bullet points for the "Core Method / Technical Approach" section of an academic poster. | ## Paper Content
# Paper Title
SI-MIL: Taming Deep MIL for Self-Interpretability in Gigapixel Histopathology
## Abstract
Introducing interpretability and reasoning into Multiple Instance Learning (MIL) methods for Whole Slide Image (WSI) analysis is challenging, given the complexity of gigapixel slides. Traditional... | • Framework combines conventional MIL branch and self-interpretable branch.
• PathExpert extracts morphometric and graph features for nuclei-related properties.
• PAG Top-K patch selection differentially selects top patches using attention scores.
• Output combines weighted PathExpert features for WSI-level prediction. |
Generate 1 bullet points for the "Experimental Results / Performance Analysis" section of an academic poster. | ## Paper Content
# Paper Title
Instance-level Expert Knowledge and Aggregate Discriminative Attention for Radiology Report Generation
## Abstract
Automatic radiology report generation can provide substantial advantages to clinical physicians by effectively reducing their workload and improving efficiency. Despite t... | • Comparing the performance of our EKAGen with other state-of-the-art methods on IU X-Ray and MIMIC-CXR datasets, with the highest performing results highlighted in bold. |
Generate 3 bullet points for the "Ablation Study" section of an academic poster. | ## Paper Content
# Paper Title
Making Old Film Great Again: Degradation-aware State Space Model for Old Film Restoration
## Abstract
Unlike modern native digital videos, the restoration of old films requires addressing specific degradations inherent to analog sources. However, existing specialized methods still fal... | • Ablation shows Mamba module contributes significantly to performance.
• FMDA improves alignment and reduces temporal artifacts.
• DPB with N=8 achieves best results among tested configurations. |
Generate 3 bullet points for the "Qualitative Results / Visualization" section of an academic poster. | ## Paper Content
# Paper Title
One Diffusion to Generate Them All
## Abstract
We introduce OneDiffusion, a versatile, large-scale diffusion model that seamlessly supports bidirectional image synthesis and understanding across diverse tasks. It enables conditional generation from inputs such as text, depth, pose, la... | • Model demonstrates zero-shot generalization
• Composes learned tasks without additional training
• Examples include face editing, pose transfer, and scene composition |
Generate 3 bullet points for the "Method Overview / Framework" section of an academic poster. | ## Paper Content
# Paper Title
MoVE-KD: Knowledge Distillation for VLMs with Mixture of Visual Encoders
## Abstract
Visual encoders are fundamental components in vision-language models (VLMs), each showcasing unique strengths derived from various pre-trained visual foundation models. To leverage the various capabil... | • We propose the MoVE-KD framework for multi-vision encoder fusion, marking the first approach to integrate different encoders for large vision-language models from a knowledge distillation perspective.
• We introduce attention-guided KD regularization, which enhances the distillation of critical visual tokens and assi... |
Generate 2 bullet points for the "Background / Related Work" section of an academic poster. | ## Paper Content
# Paper Title
SecondPose: SE(3)-Consistent Dual-Stream Feature Fusion for Category-Level Pose Estimation
## Abstract
Category-level object pose estimation, aiming to predict the 6D pose and 3D size of objects from known categories, typically struggles with large intra-class shape variation. Existin... | • Compared to other methods, our method does not require a shape prior.
• Instead, it uses a 'Semantic Prior' from DINOv2 that is more informative and flexible. |
Generate 2 bullet points for the "Qualitative Results / Visualization" section of an academic poster. | ## Paper Content
# Paper Title
A noisy elephant in the room: Is your out-of-distribution detector robust to label noise?
## Abstract
The ability to detect unfamiliar or unexpected images is essential for safe deployment of computer vision systems. In the context of classification, the task of detecting images outsi... | • Classifier trained on clean CIFAR10 labels shows clear separation between correctly and incorrectly classified ID test sets.
• With ~9% and ~40% noisy labels, the separation degrades, indicating OOD detectors struggle to distinguish incorrect ID from true OOD. |
Generate 3 bullet points for the "Qualitative Results / Visualization" section of an academic poster. | ## Paper Content
# Paper Title
RegionGPT: Towards Region Understanding Vision Language Model
## Abstract
Vision language models (VLMs) have experienced rapid advancements through the integration of large language models (LLMs) with image-text pairs, yet they struggle with detailed regional visual understanding due ... | • User asks why a banana is placed on a phone.
• RegionGPT explains it as a humorous and unexpected scene, possibly a playful act or prank.
• GPT-4V provides a similar interpretation, highlighting the model's ability to reason about object juxtaposition and intent. |
Generate 3 bullet points for the "Core Method / Technical Approach" section of an academic poster. | ## Paper Content
# Paper Title
CoMBO: Conflict Mitigation via Branched Optimization for Class Incremental Segmentation
## Abstract
Effective Class Incremental Segmentation (CIS) requires simultaneously mitigating catastrophic forgetting and ensuring sufficient plasticity to integrate new classes. The inherent confl... | • Employs a split training strategy.
• Uses dual loss functions.
• Enables unified learning of old and new classes. |
Generate 2 bullet points for the "Core Method / Technical Approach" section of an academic poster. | ## Paper Content
# Paper Title
Diffusion-Guided Weakly Supervised Semantic Segmentation
## Method
Denoising Diffusion Probabilistic Models. In our work, we employ diffusion model [21, 43] to provide well-structured high-level semantic information to the classifier that extracts the CAMs. DDPMs consist of a forward ... | • Locality Fusion Cross Attention (LFCA) transfers class semantics from ViT to diffusion features for semantic alignment
• Patch Affinity Consistency (PAC) uses denoised images as spatially and semantically consistent augmentation samples to guide patch affinity in ViT |
Generate 3 bullet points for the "Experimental Results / Performance Analysis" section of an academic poster. | ## Paper Content
# Paper Title
Dual Consolidation for Pre-Trained Model-Based Domain-Incremental Learning
## Abstract
Domain-Incremental Learning (DIL) involves the progressive adaptation of a model to new concepts across different domains. While recent advances in pre-trained models provide a solid foundation for ... | • Compares DUCT against 10 baselines across four datasets.
• Reports average accuracy (A) and best accuracy (A_B) for each method.
• DUCT achieves highest scores on all datasets, e.g., 86.27% on Office-Home A. |
Generate 3 bullet points for the "Research Motivation / Problem Background" section of an academic poster. | ## Paper Content
# Paper Title
Easy-editable Image Vectorization with Multi-layer Multi-scale Distributed Visual Feature Embedding
## Abstract
Current parameterized image representations embed visual information along the semantic boundaries and struggle to express the internal detailed texture structures of image ... | • Raster images suffer from excessive parameter redundancy and high coupling between image geometry and texture, resulting in challenging image editing.
• Image Vectorization aims to transform raster images into compact parametric representations: I = {s₁, s₂, ..., sₙ}; sᵢ means primitive vector.
• Current Image Vector... |
Generate 2 bullet points for the "Method Overview / Framework" section of an academic poster. | ## Paper Content
## Abstract
Recent 3D generation models typically rely on limited-scale 3D 'gold-labels' or 2D diffusion priors for 3D content creation. However, their performance is upper-bounded by constrained 3D priors due to the lack of scalable learning paradigms. In this work, we present See3D, a visual-condit... | • See3D is a scalable visual-conditional MVD model for open-world 3D creation from web-scale videos without camera pose annotations.
• Demonstrates capabilities including sparse multi-view to 3D, text-to-3D, 3D editing, and mesh reconstruction. |
Generate 3 bullet points for the "Research Motivation / Problem Background" section of an academic poster. | ## Paper Content
# Paper Title
Recurrence-Enhanced Vision-and-Language Transformers for Robust Multimodal Document Retrieval
## Abstract
Cross-modal retrieval is gaining increasing efficacy and interest from the research community, thanks to large-scale training, novel architectural and learning designs, and its ap... | • Unimodal retrievers fail to process multimodal queries or documents.
• Multimodal retrievers rely on global feature fusion from unimodal encoders.
• They miss fine-grained, local interactions between modalities. |
Generate 3 bullet points for the "Method Overview / Framework" section of an academic poster. | ## Paper Content
# Paper Title
REACTO: Reconstructing Articulated Objects from a Single Video
## Abstract
In this paper, we address the challenge of reconstructing general articulated 3D objects from a single video. Existing works employing dynamic neural radiance fields have advanced the modeling of articulated ob... | • Models articulated 3D objects using a shape/appearance model (Canonical NeRF) and a deformation model (Quasi-Rigid Blend Skinning).
• Transforms 3D points between observation and canonical space.
• Skinning weights (color-coded) assign each point to a bone. |
Generate 2 bullet points for the "Experimental Results / Performance Analysis" section of an academic poster. | ## Paper Content
# Paper Title
UNIMATCH: UNIVERSAL MATCHING FROM ATOM TO TASK FOR FEW-SHOT DRUG DISCOVERY
## Abstract
Drug discovery is crucial for identifying candidate drugs for various diseases. However, its low success rate often results in a scarcity of annotations, posing a few-shot learning problem. Existing... | • State-of-the-Art MPP Prediction on MoleculeNet and FS-Mol Benchmark.
• Validated in cross-domain experiment on Meta-MolNet Benchmark. |
Generate 3 bullet points for the "Core Method / Technical Approach" 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... | • Players play a regularized best response to their cumulative payoff vector.
• Follow The Regularized Leader: y(t) = v(x(t)), x(t) = Q(y(t)).
• Choice map Q: h-regularized (y,x)-argmax, with Archetype: Logit choice / Exponential Weights. |
Generate 2 bullet points for the "Method Overview / Framework" section of an academic poster. | ## Paper Content
# Paper Title
MaskINT: Video Editing via Interplicative Non-autoregressive Masked Transformers
## Abstract
Recent advances in generative AI have significantly enhanced image and video editing, particularly in the context of text prompt control. State-of-the-art approaches predominantly rely on diff... | • Non-autoregressive masked generative transformers achieve comparable performance to diffusion models in video editing.
• Significantly accelerates inference compared to diffusion-based approaches. |
Generate 2 bullet points for the "Experimental Results / Performance Analysis" section of an academic poster. | ## Paper Content
# Paper Title
Foundation Model Insights and a Multi-Model Approach for Superior Fine-Grained One-shot Subset Selection
## Abstract
One-shot subset selection serves as an effective tool to reduce deep learning training costs by identifying an informative data subset based on the information extracte... | • (1) State-of-the-art Subset Selection: Performance curves on Oxford-IIIT Pet, Food-101, and CUB-200-2011 showing superior results.
• (2) Multi-FM selection can outperform all single-FM baselines: Table comparing performance of different foundation models as information extractors, with D, C, E representing DINov2, CL... |
Generate 2 bullet points for the "Experimental Results / Performance Analysis" section of an academic poster. | ## Paper Content
# Paper Title
Learning to Solve Quadratic Unconstrained Binary Optimization in a Classification Way
## Abstract
The quadratic unconstrained binary optimization (QUBO) is a well-known NP-hard problem that takes an $n \times n$ matrix $Q$ as input and decides an $n$ -dimensional 0-1 vector $x$ ... | • Demonstrates GST’s efficiency in forming VCM.
• Shows OFV (%) over epochs for different VCM variants. |
Generate 1 bullet points for the "Experimental Results / Performance Analysis" section of an academic poster. | ## Paper Content
# Paper Title
Emotional Speech-driven 3D Body Animation via Disentangled Latent Diffusion
## Abstract
Existing methods for synthesizing 3D human gestures from speech have shown promising results, but they do not explicitly model the impact of emotions on the generated gestures. Instead, these metho... | • AMUSE is most preferred for gesture emotion (left) and speech synchronization (right) in the perceptual study. |
Generate 2 bullet points for the "Research Motivation / Problem Background" section of an academic poster. | ## Paper Content
# Paper Title
Structure-Aware Correspondence Learning for Relative Pose Estimation
## Abstract
Relative pose estimation provides a promising way for achieving object-agnostic pose estimation. Despite the success of existing 3D correspondence-based methods, the reliance on explicit feature matching ... | • Challenge 1: Estimate pose when two images have little or no overlap — solved by extracting structure-aware keypoints guided by image reconstruction loss.
• Challenge 2: Establish 3D correspondences without explicit matching — solved by using attention-based structure reasoning to directly regress 3D-3D correspondenc... |
Generate 3 bullet points for the "Ablation Study" section of an academic poster. | ## Paper Content
# Paper Title
Learning Group Activity Features Through Person Attribute Prediction
## Abstract
This paper proposes Group Activity Feature (GAF) learning in which features of multi-person activity are learned as a compact latent vector. Unlike prior work in which the manual annotation of group activ... | • Ablates GAF-L, GAF-L w/o F_loc, and Ours on Hit@1 metrics.
• Shows performance gains from location features and full model.
• Reports group activity accuracy improvements. |
Generate 3 bullet points for the "Experimental Results / Performance Analysis" section of an academic poster. | ## Paper Content
# Paper Title
A New Dataset and Framework for Real-World Blurred Images Super-Resolution
## Experiments
For comprehensive and reasonable comparative analysis, we conducted our PBaSR on three widely recognized BSR methods, namely Real-ESRGAN [65], FeMaSR [6], and SRFormer [84]. We compared them with... | • LPIPS scores on general and blur data show PbASR outperforms baselines.
• GAN loss analysis reveals PbASR reduces overfitting to blur regions.
• Performance varies with blur size and intensity, with PbASR maintaining robustness. |
Generate 4 bullet points for the "Method Overview / Framework" section of an academic poster. | ## Paper Content
# Paper Title
Learning Partonomic 3D Reconstruction from Image Collections
## Abstract
Reconstructing the 3D shape of an object from a single-view image is a fundamental task in computer vision. Recent advances in differentiable rendering have enabled 3D reconstruction from image collections using ... | • Reconstruct 3D overall shape and decompose meaningful parts based on a single 2D image
• Input: a single 2D image; Output: textured mesh along with their parts
• Motivation: objects are composed of parts defining functionality; previous methods rely on predefined primitives and are limited in expressiveness
• Two cha... |
Generate 4 bullet points for the "Method Overview / Framework" section of an academic poster. | ## Paper Content
# Paper Title
FACT: Frame-Action Cross-Attention Temporal Modeling for Efficient Action Segmentation
## Abstract
We study supervised action segmentation, whose goal is to predict framewise action labels of a video. To capture temporal dependencies over long horizons, prior works either improve fram... | • TASK: partition videos into action segments.
• CHALLENGE: better temporal model vs efficiency.
• Revisit synergy of Action and Frames: temporal modeling on both levels, capture complementary global/local info.
• Improves temporal modeling, refines features, improves efficiency (3x faster), leverages textual inputs (m... |
Generate 3 bullet points for the "Method Overview / Framework" section of an academic poster. | ## Paper Content
# Paper Title
DeRS: Towards Extremely Efficient Upcycled Mixture-of-Experts Models
## Abstract
Upcycled Mixture-of-Experts (MoE) models have shown great potential in various tasks by converting the original Feed-Forward Network (FFN) layers in pre-trained dense models into MoE layers. However, thes... | • We are the first to explore the unique redundancy mechanisms of experts in upcycled MoE models, and propose a novel DeRS paradigm that decomposes multiple experts into one expert-shared weight and multiple expert-specific weights to reduce parameter redundancy.
• Based on our DeRS paradigm, we further propose two app... |
Generate 4 bullet points for the "Experimental Results / Performance Analysis" section of an academic poster. | ## Paper Content
# Paper Title
Brain Decodes Deep Nets
## Abstract
We developed a tool for visualizing and analyzing large pre-trained vision models by mapping them onto the brain, thus exposing their hidden inside. Our innovation arises from a surprising usage of brain encoding: predicting brain fMRI measurements ... | • CLIP has the best hierarchical brain alignment.
• ImageNet and SAM’s last layer maps to middle-level brain tasks (classification/segmentation).
• MoCov3 last layer maps to 'where' (spatial) but not 'what' (semantic) part of brain.
• DiNOv2/MAE last layer does not map to any brain region; mask reconstruction is not pa... |
Generate 2 bullet points for the "Experimental Results / Performance Analysis" section of an academic poster. | ## Paper Content
# Paper Title
RePerformer: Immersive Human-centric Volumetric Videos from Playback to Photoreal Reperformance
## Abstract
Human-centric volumetric videos offer immersive free-viewpoint experiences, yet existing methods focus either on replaying general dynamic scenes or animating human avatars, lim... | • Qualitative comparison across methods: Ground Truth, Ours, NeuS2, Spacetime Gaussian, V^1, DualGS.
• Novel View and Novel Motion comparisons show Ours preserving detail and motion fidelity better than baselines. |
Generate 1 bullet points for the "Qualitative Results / Visualization" section of an academic poster. | ## Paper Content
# Paper Title
Dragin3D: Image Editing by Dragging in 3D Space
## Abstract
Interactive drag editing of images is a valuable task that has gained considerable attention for its precision and controllability. However, existing approaches have primarily focused on manipulating the shape or movement of ... | • More results of object rotation in real-world scenes demonstrate the strong generalization ability of our model. |
Generate 3 bullet points for the "Core Method / Technical Approach" section of an academic poster. | ## Paper Content
# Paper Title
Panorama Generation From NFoV Image Done Right
## Abstract
Generating 360-degree panoramas from narrow field of view (NFoV) image is a promising computer vision task for Virtual Reality (VR) applications. Existing methods mostly assess the generated panoramas with InceptionNet or CLIP... | • Defines distortion map using spherical coordinates: S(θ,φ,r) = I(i,j).
• Uses first-order Taylor expansion for distortion modeling.
• Maps pixel coordinates to spherical angles for accurate projection. |
Generate 5 bullet points for the "Conclusion / Future Work" section of an academic poster. | ## Paper Content
# Paper Title
Kermut: Composite kernel regression for protein variant effects
## Abstract
Reliable prediction of protein variant effects is crucial for both protein optimization and for advancing biological understanding. For practical use in protein engineering, it is important that we can also pr... | • Protein-specific — no learning across proteins and assays.
• Substitutions only — no support for insertions or deletions.
• Structure kernel models multi-mutants linearly — only epistasis via sequence embeddings.
• Extrapolation to higher order mutations is challenging and requires further analysis.
• GPUs scale cubi... |
Generate 3 bullet points for the "Method Overview / Framework" section of an academic poster. | ## Paper Content
# Paper Title
SIDA: Social Media Image Deepfake Detection, Localization and Explanation with Large Multimodal Model
## Abstract
The rapid advancement of generative models in creating highly realistic images poses substantial risks for misinformation dissemination. For instance, a synthetic image, w... | • We establish SID-Set, a comprehensive benchmark for detecting, localizing, and explaining deepfakes in social media images.
• We propose SIDA, a new image deepfake detection, localization, and explanation framework that not only detects images with high accuracy but also localizes and explains potential manipulations... |
Generate 3 bullet points for the "Background / Related Work" section of an academic poster. | ## Paper Content
# Paper Title
SOFTCVI: CONTRASTIVE VARIATIONAL INFERENCE WITH SELF-GENERATED SOFT LABELS
## Abstract
Estimating a distribution given access to its unnormalized density is pivotal in Bayesian inference, where the posterior is generally known only up to an unknown normalizing constant. Variational in... | • Classification problems share three ingredients: Samples, Labels, A classifier.
• Contrastive classification: Imagine we have K samples, containing exactly one sample from a positive distribution p⁺, and K−1 samples from the negative distribution p⁻. What is the optimal classifier?
• The optimal classifier assigns la... |
Generate 3 bullet points for the "Core Method / Technical Approach" section of an academic poster. | ## Paper Content
# Paper Title
TIME-MOE: BILLLION-SCALE TIME SERIES FOUNDATION MODELS WITH MIXTURE OF EXPERTS
## Abstract
Deep learning for time series forecasting has seen significant advancements over the past decades. However, despite the success of large-scale pre-training in language and vision domains, pre-tr... | • Tokenizes input time series into data points, encoded via N-stacked backbone layers.
• Uses causal attention and sparse temporal mixture-of-expert layers.
• Dynamically schedules heads for flexible forecasting length. |
Generate 3 bullet points for the "Background / Related Work" section of an academic poster. | ## Paper Content
# Paper Title
Evidential Active Recognition: Intelligent and Prudent Open-World Embodied Perception
## Abstract
Active recognition enables robots to intelligently explore novel observations, thereby acquiring more information while circumventing undesired viewing conditions. Recent approaches favor... | • Active recognition addresses challenges unresolved by passive recognition.
• We built a dataset to facilitate evaluation in indoor simulators.
• Difficulty levels consider visibility, relative distance, and observed pixels, with examples like 'picture' (moderate) and 'chest of drawers' (hard). |
Generate 3 bullet points for the "Research Motivation / Problem Background" section of an academic poster. | ## Paper Content
# Paper Title
FRAME: Floor-aligned Representation for Avatar Motion from Egocentric Video
## Abstract
Egocentric motion capture with a head-mounted body-facing stereo camera is crucial for VR and AR applications but presents significant challenges such as heavy occlusions and limited annotated real... | • Record egocentric video via stereo camera
• Predict 6D pose of avatar
• Goal: generate realistic avatar motion from first-person view |
Generate 3 bullet points for the "Qualitative Results / Visualization" section of an academic poster. | ## Paper Content
# Paper Title
SceneDiffuser++: City-Scale Traffic Simulation via a Generative World Model
## Abstract
The goal of traffic simulation is to augment a potentially limited amount of manually-driven miles that is available for testing and validation, with a much larger amount of simulated synthetic mil... | • Shows long-term simulation at different steps (0, 100, 200, 600)
• Demonstrates stable agent behavior over time
• Includes route visualization and agent positions |
Generate 3 bullet points for the "Background / Related Work" section of an academic poster. | ## Paper Content
# Paper Title
All-Purpose Mean Estimation over $\mathbb{R}$ : Optimal Sub-Gaussianity with Outlier Robustness and Low Moments Performance
## Abstract
We consider the basic statistical challenge of designing an "all-purpose" mean estimation algorithm that is recommendable across a variety of settin... | • If p has finite variance, [LV22] achieves error bound matching Gaussian lower bound with tight constant.
• 'o(1)' is crucially independent of p.
• If p has 1+α moments for α < 1, median-of-means estimator is big-O optimal [BCL13, DLLO16]. |
Generate 4 bullet points for the "Experimental Results / Performance Analysis" section of an academic poster. | ## Paper Content
# Paper Title
Concept Arithmetics for Circumventing Concept Inhibition in Diffusion Models
## Experiments
We quantitatively evaluate the proposed attack implementations on the models that were inhibited for nudity in Section 4.1; object categories and recognizable figures in Section 4.2. Qualitativ... | • Tests nudity inhibition using ESD, UCE, SA methods.
• Concept presence metric: Nudity Detector Model (NudeNet).
• Prompts from I2P Dataset.
• Shows detection counts for female breast, female genitalia, male genitalia exposure. |
Generate 4 bullet points for the "Core Method / Technical Approach" section of an academic poster. | ## Paper Content
# Paper Title
Animating General Image with Large Visual Motion Model
## Abstract
We present the pioneering Large Visual Motion Model (LVMM), meticulously engineered to analyze the intrinsic dynamics encapsulated within real-world imagery. Our model, fortified with a wealth of prior knowledge extrac... | • Starting from a reference image I0, LVMM estimates a latent motion trajectory z1,…,K using a motion denoising model εθ.
• The trajectory is processed by the Motion Decoder D to generate optical flows and intention maps.
• The Neural Image Renderer R transforms I0 into a sequence of novel images I1,…,K guided by optic... |
Generate 3 bullet points for the "Other Content" section of an academic poster. | ## Paper Content
# Paper Title
HOISDF: Constraining 3D Hand-Object Pose Estimation with Global Signed Distance Fields
## Abstract
Human hands are highly articulated and versatile at handling objects. Jointly estimating the 3D poses of a hand and the object it manipulates from a monocular camera is challenging due t... | • The work was funded by EPFL and Microsoft Swiss Joint Research Center (H.Q., A.M.).
• We are grateful to the members of the Mathis Group and in particular Niels Poulsen for comments on an earlier version of this manuscript.
• We also sincerely thank Rong Wang, Wei Mao and Hongdong Li for sharing the hand-object rende... |
Generate 3 bullet points for the "Experimental Results / Performance Analysis" section of an academic poster. | ## Paper Content
# Paper Title
CorrBEV: Multi-View 3D Object Detection by Correlation Learning with Multi-modal Prototypes
## Abstract
Camera-only multi-view 3D object detection in autonomous driving has witnessed encouraging developments in recent years, largely attributed to the revolution of fundamental architec... | • Consistent improvements under different occlusion levels shown in performance table
• Reducing confusion among categories visualized via scatter plots
• Two typical occlusion scenarios demonstrate CorrBEV's positive effect with visual comparisons |
Generate 3 bullet points for the "Core Method / Technical Approach" section of an academic poster. | ## Paper Content
# Paper Title
Global and Local Prompts Cooperation via Optimal Transport for Federated Learning
## Abstract
Prompt learning in pretrained visual-language models has shown remarkable flexibility across various downstream tasks. Leveraging its inherent lightweight nature, recent research attempted to... | • Framework equips each client with a fixed pre-trained CLIP model and a prompt learner
• Global and local prompts are jointly updated via gradient descent and aggregated on the server
• Uses Unbalanced Optimal Transport to align global and local prompts by minimizing cost matrix based on cosine distance |
Generate 3 bullet points for the "Core Method / Technical Approach" section of an academic poster. | ## Paper Content
## Abstract
Vision-language models (VLMs) like CLIP have been widely used in various specific tasks. Parameter-efficient fine-tuning (PEFT) methods, such as prompt and adapter tuning, have become key techniques for adapting these models to specific domains. However, existing approaches rely on prior ... | • Step (a): Fine-tune the CLIP transformer block by sampling a batch of training data.
• Step (b): Select parameters to fine-tune by counting v_i' of each parameter in AdamW, using the formula θ_i = (θ_{i-1} - α·λ·θ_{i-1}) - (α/√(θ̂_i + ε))·m̂_i, where v_i' = Avg(1/√θ̂_i).
• Step (c): Fine-tune selected parameters usin... |
Generate 2 bullet points for the "Background / Related Work" section of an academic poster. | ## Paper Content
# Paper Title
LEOD: Label-Efficient Object Detection for Event Cameras
## Abstract
Object detection with event cameras benefits from the sensor's low latency and high dynamic range. However, it is costly to fully label event streams for supervised training due to their high temporal resolution. To ... | • Left: Weakly-supervised object detection (WSOD), where all event streams are sparsely annotated.
• Right: Semi-supervised object detection (SSOD), where some streams are densely labeled and others fully unlabeled. |
Generate 2 bullet points for the "Qualitative Results / Visualization" section of an academic poster. | ## Paper Content
# Paper Title
Navigation World Models
## Abstract
Navigation is a fundamental skill of agents with visual-motor capabilities. We introduce a Navigation World Model (NWM), a controllable video generation model that predicts future visual observations based on past observations and navigation actions... | • Shows input image, goal image, and three planned trajectories (P1, P2, P3) with loss values
• P2 (Min) has lowest loss (0.400) |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.