instruction stringclasses 95
values | input stringlengths 141 265k | output stringlengths 15 1.61k |
|---|---|---|
Generate 4 bullet points for the "Research Motivation / Problem Background" section of an academic poster. | ## Paper Content
# Paper Title
DreamDrone: Text-to-Image Diffusion Models are Zero-shot Perpetual View Generators
## Introduction
Recent advances in vision and graphics have enabled the synthesis of multi-view consistent 3D scenes along extended camera trajectories [3,7,18,20]. This emerging task, termed perpetual ... | • Asks whether perpetual scene images can be generated via warping operations.
• Highlights that naive warping + inpainting suffers from blank holes, blurriness, and distortion.
• Proposes warping intermediate latent codes of pre-trained diffusion models as an advanced solution.
• Raises key questions: How to guarantee... |
Generate 3 bullet points for the "Other Content" section of an academic poster. | ## Paper Content
# Paper Title
ProReflow: Progressive Reflow with Decomposed Velocity
## Abstract
Diffusion models have achieved significant progress in both image and video generation while still suffering from huge computation costs. As an effective solution, rectified flow aims to rectify the diffusion process o... | • I am a master's student in my second year at Tsinghua University and focus on effectimt visual generation model. If you have any question about this paper, please freely concat me.
• Email: kl23@mails.tsinghua.edu.cn
• WeChat: yssy_zqbx |
Generate 2 bullet points for the "Qualitative Results / Visualization" 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... | • Illustrates interpretation of detection failures, including misclassified and undetected instances.
• Shows attribution maps and searched regions for wrong/true classes and undetected objects with object scores and insertion metrics. |
Generate 3 bullet points for the "Core Method / Technical Approach" section of an academic poster. | ## Paper Content
# Paper Title
PointNeRF++: A multi-scale, point-based Neural Radiance Field
## Method
An overview of our method is shown in Fig. 2. We build a representation starting from an input point cloud, which we then use to volume-render [30] a scene. Specifically, given an input point cloud $\mathbf{P}_{\... | • Points are explicit, better efficiency, controllability etc.
• Challenges: False negative when querying happens due to holes existing in incomplete/sparse point cloud.
• Existing works assume input point cloud of high quality. Or they rely on hand-crafted heuristics. |
Generate 4 bullet points for the "Core Method / Technical Approach" section of an academic poster. | ## Paper Content
# Paper Title
Rethinking Deep Unrolled Model for Accelerated MRI Reconstruction
## Method
Drawing inspiration from adaptive gradient and momentum-accelerated algorithms in gradient descent optimization, we dynamically incorporate prior knowledge in the data flow, to enable more informative and fast... | • DUM truncates and unrolls iterations, replacing hand-crafted R(x) with deep neural networks.
• Update formula: x_{t+1} = x_t - η_t A^H(Ax_t - y) + CNN(x_t).
• DUM shows superior performance and interpretability.
• Question: Has the potential of DUM been fully exploited? |
Generate 2 bullet points for the "Core Method / Technical Approach" section of an academic poster. | ## Paper Content
# Paper Title
Neuro-Symbolic Evaluation of Text-to-Video Models using Formal Verification
## Abstract
Recent advancements in text-to-video models such as Sora, Gen-3, MovieGen, and CogVideoX are pushing the boundaries of synthetic video generation, with adoption seen in fields like robotics, autono... | • NeuS-V evaluates videos across four distinct modes: Overall Consistency, Object Existence, Spatial Relationship, and Object Action Alignment.
• Each mode is formalized using temporal logic to capture spatio-temporal semantics from the prompt. |
Generate 3 bullet points for the "Experimental Results / Performance Analysis" section of an academic poster. | ## Paper Content
# Paper Title
PiTe: Pixel-Temporal Alignment for Large Video-Language Model
## Experiments
PiTe is composed of a vision encoder to encode frames from video implemented as ViT [8], a vision adapter to project visual feature to semantic space of LLMs implemented as a linear projection layer, a LLM Vi... | • PiTe (Ours) achieves 71.6 Accuracy on MSVD-QA, 57.7 Accuracy on MSRVTT-QA, and 42.2 Accuracy on ActivityNet-QA.
• PiTe outperforms baselines including Video-ChatGPT and PG-Video-LLaVA.
• On Temporal Grounding and Dense Captioning, PiTe achieves 30.4 R@0.3↑ and 26.5 METEOR↑ respectively. |
Generate 2 bullet points for the "Experimental Results / Performance Analysis" section of an academic poster. | ## Paper Content
# Paper Title
MMEarth: Exploring Multi-Modal Pretext Tasks For Geospatial Representation Learning
## Experiments
In this section, we describe the proposed Multi-Pretext Masked Autoencoder (MP-MAE) approach as illustrated in Fig. 2. It builds on the promising results of MIM with the ConvNeXt V2 arch... | • Shows performance gains in low-data regimes (1% to 100% training set size).
• MMEarth64 consistently outperforms ImageNet and MMEarth4-S2 across So2Sat20k, EuroSat2k, and BigEarth20k. |
Generate 3 bullet points for the "Core Method / Technical Approach" section of an academic poster. | ## Paper Content
# Paper Title
OpenMIBOOD: Open Medical Imaging Benchmarks for Out-Of-Distribution Detection
## Abstract
The growing reliance on Artificial Intelligence (AI) in critical domains such as healthcare demands robust mechanisms to ensure the trustworthiness of these systems, especially when faced with un... | • Suite of 14 datasets across three medical imaging domains: MIDOG (Histopathology), PhaKIR (Endoscopy), OASIS-3 (MRI).
• Evaluation of 24 Out-Of-Distribution detection methods: six feature-based, nine classification-based, nine hybrid methods.
• Includes covariate-shifted ID, near-OOD, and far-OOD categories for compr... |
Generate 4 bullet points for the "Core Method / Technical Approach" section of an academic poster. | ## Paper Content
# Paper Title
Universal Segmentation at Arbitrary Granularity with Language Instruction
## Abstract
This paper aims to achieve universal segmentation of arbitrary semantic level. Despite significant progress in recent years, specialist segmentation approaches are limited to specific tasks and data ... | • Proposes a bottom-up interaction framework with sufficient visual-linguistic alignment.
• Uses coarse-grained and fine-grained language instructions as input.
• Core components: Pre-Fusion, Symmetric Transformer, Multi-Modal Transformer, Segmentation Decoder, and FPN.
• Pipeline integrates visual-linguistic decoding ... |
Generate 3 bullet points for the "Core Method / Technical Approach" section of an academic poster. | ## Paper Content
# Paper Title
LayerDiff: Exploring Text-guided Multi-layered Composable Image Synthesis via Layer-Collaborative Diffusion Model
## Method
In this section, we first define the data format of the multi-layered composable image and the task formulation for achieving multi-layered composable image synt... | • LayerDiff performs inter-layer and intra-layer interaction.
• Cooperates with global and local text condition guidance.
• Uses image encoder, text encoder, and layer-specific prompt enhancer. |
Generate 3 bullet points for the "Core Method / Technical Approach" section of an academic poster. | ## Paper Content
# Paper Title
See Further When Clear: Curriculum Consistency Model
## Abstract
Significant advances have been made in the sampling efficiency of diffusion and flow matching models, driven by Consistency Distillation (CD), which trains a student model to mimic the output of a teacher model at a late... | • Presents the overall framework with curriculum consistency and multi-step iterative generation.
• Defines KDC as 100 - PSNR between student and teacher outputs.
• Uses dynamic adjustment of learning objectives based on KDC thresholds. |
Generate 3 bullet points for the "Research Motivation / Problem Background" section of an academic poster. | ## Paper Content
# Paper Title
LED: A Large-scale Real-world Paired Dataset for Event Camera Denoising
## Abstract
Event camera has significant advantages in capturing dynamic scene information while being prone to noise interference, particularly in challenging conditions like low threshold and low illumination. H... | • Event cameras are prone to noise interference, particularly in challenging conditions like low threshold and low illumination;
• Existing denoising methods mainly focus on such gentle scenes, so the denoising accuracy is limited under the above challenges;
• The scarcity of the real-world paired event denoising datas... |
Generate 2 bullet points for the "Ablation Study" section of an academic poster. | ## Paper Content
# Paper Title
Monocular and Generalizable Gaussian Talking Head Animation
## Abstract
In this work, we introduce Monocular and Generalizable Gaussian Talking Head Animation (MGGTalk), which requires monocular datasets and generalizes to unseen identities without personalized re-training. Compared w... | • DSGR Module: Without mirroring, geometry for occluded regions (e.g., the left ear) is missing due to monocular depth limitations. Without Gaussian Filter, direct mirroring without filtering causes overlaps and loss of detail.
• SGP Module: Replacing geometry with expression features weakens structure awareness, espec... |
Generate 3 bullet points for the "Research Motivation / Problem Background" section of an academic poster. | ## Paper Content
# Paper Title
CAP-Net: A Unified Network for 6D Pose and Size Estimation of Categorical Articulated Parts from a Single RGB-D Image
## Abstract
This paper tackles category-level pose estimation of articulated objects in robotic manipulation tasks and introduces a new benchmark dataset. While recent... | • Prior approaches adopt a "segment-then-pose" strategy, leading to error accumulation from part segmentation and loss of global context.
• Existing point-cloud-based methods neglect dense semantic features from RGB images, resulting in suboptimal accuracy for small parts.
• Existing datasets lack photorealistic RGB im... |
Generate 5 bullet points for the "Core Method / Technical Approach" section of an academic poster. | ## Paper Content
# Paper Title
EIDT-V: Exploiting Intersections in Diffusion Trajectories for Model-Agnostic, Zero-Shot, Training-Free Text-to-Video Generation
## Abstract
Zero-shot, training-free, image-based text-to-video generation is an emerging area that aims to generate videos using existing image-based diffu... | • Pipeline Steps: LLM converts user text into frame-wise prompts and variation cues.
• Change Detection: LLM compares consecutive prompts to pinpoint semantics that should change.
• Mask Computation: A CLIP segmentation model turns detected differences into pixel-precise spatial–temporal masks highlighting dynamic regi... |
Generate 3 bullet points for the "Ablation Study" section of an academic poster. | ## Paper Content
# Paper Title
DISCOVERYWORLD: A Virtual Environment for Developing and Evaluating Automated Scientific Discovery Agents
## Abstract
Automated scientific discovery promises to accelerate progress across scientific domains. However, developing and evaluating an AI agent's capacity for end-to-end scie... | • Includes 10 unit tests showing agents are skilled at environment interaction/embodiment.
• Poor agent performance is driven by poor scientific ability, not navigation or interaction.
• Table shows unit test performance for ReAct, Plan+Execute, and Hypothesizer agents. |
Generate 3 bullet points for the "Experimental Results / Performance Analysis" section of an academic poster. | ## Paper Content
# Paper Title
SnapGen: Taming High-Resolution Text-to-Image Models for Mobile Devices with Efficient Architectures and Training
## Abstract
Existing text-to-image (T2I) diffusion models face several limitations, including large model sizes, slow runtime, and low-quality generation on mobile devices... | • SnapGen achieves 0.66 GenEval score, 84.8 CLIP-T, and 0.532 Image Reward
• Outperforms SD3.5-Large in throughput (2.6B vs 8.1B parameters) while maintaining competitive metrics
• Shows trade-offs between model size, speed, and quality across different architectures |
Generate 3 bullet points for the "Background / Related Work" section of an academic poster. | ## Paper Content
# Paper Title
MapEval: A Map-Based Evaluation of Geo-Spatial Reasoning in Foundation Models
## Abstract
Recent advancements in foundation models have improved autonomous tool usage and reasoning, but their capabilities in map-based reasoning remain underexplored. To address this, we introduce MapEv... | • Textual: Answer MCQs by reasoning over mixed structured and unstructured map-related textual data.
• API: Generate and interpret map API calls to retrieve and reason over structured data for answering MCQs.
• Visual: Analyze map snapshots and perform spatial reasoning to answer MCQs based on visual map elements. |
Generate 5 bullet points for the "Method Overview / Framework" section of an academic poster. | ## Paper Content
# Paper Title
VGGT: Visual Geometry Grounded Transformer
## Abstract
We present VGGT, a feed-forward neural network that directly infers all key 3D attributes of a scene, including camera parameters, point maps, depth maps, and 3D point tracks, from one, a few, or hundreds of its views. This approa... | • Feed-forward Reconstruction in One Go: No Optimization Required
• Simple and Efficient: runs in just seconds
• Comprehensive Outputs: camera, depth, points, and tracks
• SoTA: excels on core 3D tasks & benefits downstream tasks
• Strong Generalization: Handles diverse inputs: internet photos, cartoons, and generated ... |
Generate 3 bullet points for the "Core Method / Technical Approach" section of an academic poster. | ## Paper Content
# Paper Title
Animatable Gaussians: Learning Pose-dependent Gaussian Maps for High-fidelity Human Avatar Modeling
## Abstract
Modeling animatable human avatars from RGB videos is a long-standing and challenging problem. Recent works usually adopt MLP-based neural radiance fields (NeRF) to represent... | • Uses posed position maps and 2D CNNs for powerful representation ability
• Generates detailed avatars by combining Gaussian maps with pose conditioning
• Avoids blurring by replacing implicit MLPs with explicit Gaussian splatting |
Generate 3 bullet points for the "Research Motivation / Problem Background" section of an academic poster. | ## Paper Content
# Paper Title
Dual-Enhanced Coreset Selection with Class-wise Collaboration for Online Blurry Class Incremental Learning
## Abstract
Traditional online class incremental learning assumes class sets in different tasks are disjoint. However, recent works have shifted towards a more realistic scenario... | • Traditional online class incremental learning assumes disjoint class sets, but real-world tasks often share classes and have blurred boundaries (OBCIL setting).
• Existing rehearsal-based methods struggle with data imbalance and varying class-wise data volumes, complicating coreset selection.
• DECO is introduced to ... |
Generate 3 bullet points for the "Core Method / Technical Approach" section of an academic poster. | ## Paper Content
# Paper Title
Augmenting Multimodal LLMs with Self-Reflective Tokens for Knowledge-based Visual Question Answering
## Abstract
Multimodal LLMs (MLLMs) are the natural extension of large language models to handle multimodal inputs, combining text and image data. They have recently garnered attention... | • <NORET>: Answer is provided directly without retrieval information.
• <RET>: Initiate retrieval pipeline; retrieved passages are labeled as <REL> or <NOREL>.
• All <REL> passages are concatenated and given to the model for the final answer. |
Generate 4 bullet points for the "Core Method / Technical Approach" section of an academic poster. | ## Paper Content
# Paper Title
Looking Similar, Sounding Different: Leveraging Counterfactual Cross-Modal Pairs for Audiovisual Representation Learning
## Abstract
Audiovisual representation learning typically relies on the correspondence between sight and sound. However, there are often multiple audio tracks that ... | • We train Multiscale Vision Transformers (MViTs) for video and audio, using augmented and temporally jittered samples.
• We rely mainly on cross-modal training, with little to no within-modal contrast (weights for within-modal term are 0–0.2 depending on the model variant).
• Overall, we produce 11 model variants to i... |
Generate 3 bullet points for the "Other Content" section of an academic poster. | ## Paper Content
# Paper Title
CogAgent: A Visual Language Model for GUI Agents
## Abstract
People are spending an enormous amount of time on digital devices through graphical user interfaces (GUIs), e.g., computer or smartphone screens. Large language models (LLMs) such as ChatGPT can assist people in tasks like w... | • CogAgent is open-sourced under the CogVLM2 series.
• GitHub: github.com/THUDM/CogVLM
• Huggingface: huggingface.co/THUDM/cogagent-chat-hf |
Generate 3 bullet points for the "Core Method / Technical Approach" section of an academic poster. | ## Paper Content
# Paper Title
Improved monocular depth prediction using distance transform over pre-semantic contours with self-supervised neural networks
## Abstract
Monocular depth estimation (MDE) with self-supervised training approaches struggles in low-texture areas, where photometric losses may lead to ambig... | • Self-supervised training reconstructs target view from source view using depth and camera pose networks.
• Reprojection formula: I_t^s,o(p) = I_s(K T_{t→s} D_t(p) K^-1 p).
• Ambiguity in homogeneous areas leads to ill-posed problem; goal is to introduce reproducible structure. |
Generate 3 bullet points for the "Background / Related Work" section of an academic poster. | ## Paper Content
# Paper Title
MultiOOD: Scaling Out-of-Distribution Detection for Multiple Modalities
## Abstract
Detecting out-of-distribution (OOD) samples is important for deploying machine learning models in safety-critical applications such as autonomous driving and robot-assisted surgery. Existing research h... | • Introduces 4x Multimodal Near-OOD and 2x Multimodal Far-OOD benchmarks.
• Covers 5 datasets: EPIC-Kitchens, HMDB51, UCF101, Kinetics-600, HAC.
• Includes 229 classes and 85,000 videos across 3 modalities (Video, Optical Flow, Audio). |
Generate 3 bullet points for the "Experimental Results / Performance Analysis" section of an academic poster. | ## Paper Content
# Paper Title
Visual Objectification in Films: Towards a New AI Task for Video Interpretation
## Abstract
In film gender studies, the concept of "male gaze" refers to the way the characters are portrayed on-screen as objects of desire rather than subjects. In this article, we introduce a novel vide... | • Task accuracy: EN vs. S, HN vs. S, (EN U HN) vs. S scores for models including ViViT-B/16, X-CLIP, Random, All positive, PCBM-DT, PCBM-LR
• Weighted F1-scores show Hard Neg annotations help
• Error Analysis: Logistic regression coefficients show which concepts (e.g., Body, Type of shot) are poorly captured |
Generate 3 bullet points for the "Core Method / Technical Approach" section of an academic poster. | ## Paper Content
# Paper Title
Scene Map-based Prompt Tuning for Navigation Instruction Generation
## Abstract
Navigation instruction generation (NIG), which provides interactive feedback and guidance to humans along a trajectory, is vital for developing embodied agents capable of human-machine communication and co... | • The scene representation encoding module captures fine-grained 3D scene features, including perspective embedding and 3D scene representation.
• The map prompt tuning module aggregates environment connectivity and visual cues across viewpoints, feeding topological representations into the LLM.
• The landmark uncertai... |
Generate 4 bullet points for the "Core Method / Technical Approach" section of an academic poster. | ## Paper Content
# Paper Title
From Variance to Veracity: Unbundling and Mitigating Gradient Variance in Differentiable Bundle Adjustment Layers
## Abstract
Various pose estimation and tracking problems in robotics can be decomposed into a correspondence estimation problem (often computed using a deep network) foll... | • Bundle Adjustment updates p̂jk = p̄jk,i + δ using Gauss-Newton steps.
• Flow Loss Interference: Outliers corrupt gradients for all depths projected on that frame.
• Linearization errors in BA: High variance in T, d leads to high variance Jacobian estimates.
• Weight-residual dependence: Weight gradients are proportio... |
Generate 2 bullet points for the "Experimental Results / Performance Analysis" 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... | • Improvements upon STEGO and Hidden Positives across benchmarks.
• Our method achieves 81.6 U. Acc. and 23.1 U. mIoU on COCOStuff-27 with VIT-B/8. |
Generate 3 bullet points for the "Method Overview / Framework" section of an academic poster. | ## Paper Content
# Paper Title
DreamDiffusion: High-Quality EEG-to-Image Generation with Temporal Masked Signal Modeling and CLIP Alignment
## Introduction
Image generation [4,15,21] has made great strides in recent years, especially after breakthroughs in text-to-image generation [1, 12, 29, 30, 33]. The recent te... | • Task: Generating high-quality images directly from brain electroencephalogram (EEG) signals.
• Challenge: EEG signals are captured non-invasively and thus are inherently noisy. In addition, EEG data are limited, and individual differences cannot be ignored.
• Solution: Temporal masked signal modeling for effective an... |
Generate 2 bullet points for the "Experimental Results / Performance Analysis" section of an academic poster. | ## Paper Content
# Paper Title
VISTA: Enhancing Long-Duration and High-Resolution Video Understanding by VVideo SpatioTemporal Augmentation
## Abstract
Current large multimodal models (LMMs) face significant challenges in processing and comprehending long-duration or high-resolution videos, which is mainly due to t... | • Created VISTA-400K: high-quality synthetic video instruction-following dataset with 400K video QA pairs
• VISTA-finetuned models achieve 3.3% avg gain on long-video benchmarks and 6.5% on HRVideoBench vs vanilla models |
Generate 1 bullet points for the "Implementation Details" section of an academic poster. | ## Paper Content
# Paper Title
On the Generalization of Handwritten Text Recognition Models
## Abstract
Recent advances in Handwritten Text Recognition (HTR) have led to significant reductions in transcription errors on standard benchmarks under the i.i.d. assumption, thus focusing on minimizing in-distribution (ID... | • We study 8 SOTA HTR models (3 alignments); 7 datasets (in 5 languages) |
Generate 3 bullet points for the "Core Method / Technical Approach" section of an academic poster. | ## Paper Content
# Paper Title
Text2HOI: Text-guided 3D Motion Generation for Hand-Object Interaction
## Abstract
This paper introduces the first text-guided work for generating the sequence of hand-object interaction in 3D. The main challenge arises from the lack of labeled data where existing ground-truth dataset... | • Three main components: contact map generation, motion generation, and refinement.
• Input text prompt guides generation of canonical object mesh, contact maps, and denoised outputs.
• Final refined outputs show realistic hand-object interactions. |
Generate 1 bullet points for the "Method Overview / Framework" section of an academic poster. | ## Paper Content
# Paper Title
One-Shot Open Affordance Learning with Foundation Models
## Abstract
We introduce One-shot Open Affordance Learning (OOAL), where a model is trained with just one example per base object category, but is expected to identify novel objects and affordances. While vision-language models ... | • We utilize foundation models to perform affordance learning with only one-shot sample per object class during training and can generalize to unseen object and affordance categories. |
Generate 4 bullet points for the "Experimental Results / Performance Analysis" section of an academic poster. | ## Paper Content
# Paper Title
AdMiT: Adaptive Multi-Source Tuning in Dynamic Environments
## Abstract
Incorporating transformer models into edge devices poses a significant challenge due to the computational demands of adapting these large models across diverse applications. Parameter-efficient tuning (PET) method... | • Datasets: Digits-Five, ImageNet-C, CIFAR-100C, Cityscapes → ACDC.
• Metrics: Accuracy (%), mIoU (%).
• Baselines: Single-Source TTA (TENT, SAR, BECOTTA), Multi-Source (SESM, π-tuning, CONTRAST, Model Soup).
• Findings: Static D_T: AdMiT > Baselines (Avg. Acc). Dynamic D_T: AdMiT > Baselines (Avg. Acc), shows less for... |
Generate 1 bullet points for the "Other Content" section of an academic poster. | ## Paper Content
# Paper Title
EXPLORING THE LOSS LANDSCAPE OF REGULARIZED NEURAL NETWORKS VIA CONVEX DUALITY
## Abstract
We discuss several aspects of the loss landscape of regularized neural networks: the structure of stationary points, connectivity of optimal solutions, path with nonincreasing loss to arbitrary ... | • Lists references including Pilanci and Ergen (2020), Mishkin and Pilanci (2023), Bourrier and Flammarion (2023), Haefele and Vidal (2017), and Nguyen (2021). |
Generate 3 bullet points for the "Experimental Results / Performance Analysis" section of an academic poster. | ## Paper Content
# Paper Title
Pose Adapted Shape Learning for Large-Pose Face Reenactment
## Abstract
We propose the Pose Adapted Shape Learning (PASL) for large-posed face reenactment. The PASL framework consists of three modules, namely the Pose-Adapted face Encoder (PAE), the Cycle-consistent Shape Generator (C... | • PAE outperforms other face encoders for face verification on all six pose-pair sets.
• Tables show quantitative metrics (FID, CSIM, ARD, LPIPS) for Self Reenactment and Cross Reenactment on MPIE-LP, VoxCeleb1, and VoxCeleb2-LP datasets.
• PAE achieves the best scores across most metrics, e.g., 17.5 FID and 0.47 CSIM ... |
Generate 2 bullet points for the "Experimental Results / Performance Analysis" section of an academic poster. | ## Paper Content
# Paper Title
Improving Neural Surface Reconstruction with Feature Priors from Multi-View Images
## Experiments
Datasets. The DTU dataset tested in NSR comprises 15 scenes with 49 or 64 multi-view images collected in an indoor environment with fixed camera poses following [52]. The Chamfer Distance... | • Shows qualitative reconstructions for Scan24, Scan37, Scan40, and Scan110 using Neus, NeuralWarp, Match-NeuS, and MVS-NeuS.
• Includes quantitative evaluation across 15 scans with metrics including COLMAP, MVSDF, IDR, NeRF, NeuS, VolSDF, NeuralWarp, GeoNeuS*, D-NeuS, and PET-NeuS. |
Generate 2 bullet points for the "Experimental Results / Performance Analysis" section of an academic poster. | ## Paper Content
# Paper Title
Visual Program Distillation: Distilling Tools and Programmatic Reasoning into Vision-Language Models
## Abstract
Solving complex visual tasks such as "Who invented the musical instrument on the right?" involves a composition of skills: understanding space, recognizing instruments, and... | • According to our human study: VPD models have better accuracy and are more likely to explain their answers.
• Explanations are more factual and consistent. |
Generate 5 bullet points for the "Core Method / Technical Approach" section of an academic poster. | ## Paper Content
# Paper Title
LTA-PCS: Learnable Task-Agnostic Point Cloud Sampling
## Abstract
Recently, many approaches directly operate on point clouds for different tasks. These approaches become more computation and storage demanding when point cloud size is large. To reduce the required computation and stora... | • Sampling Network: Our model follows PointNet architecture.
• Projection: We project point clouds to multi-view depth maps and copy them three times to form input for the 2D feature extractor (e.g., CLIP).
• Overall Loss Function: L = L_geo + αL_sem.
• Geometric Loss: Combines L_a, L_w, and L_s terms to preserve geome... |
Generate 3 bullet points for the "Background / Related Work" section of an academic poster. | ## Paper Content
# Paper Title
Adventurer: Optimizing Vision Mamba Architecture Designs for Efficiency
## Abstract
In this work, we introduce the Adventurer series models where we treat images as sequences of patch tokens and employ uni-directional language models to learn visual representations. This modeling para... | • Causal modeling is sufficient for image encoding and can achieve competitive results with standard ViTs.
• Standard ViTs involve redundant computations; causal modeling can accelerate self-attention by ~50%.
• Visual backbones can be more efficient by incorporating RNN-like token mixers like Mamba, which scale linear... |
Generate 4 bullet points for the "Core Method / Technical Approach" section of an academic poster. | ## Paper Content
# Paper Title
Optimal Transport Aggregation for Visual Place Recognition
## Abstract
The task of Visual Place Recognition (VPR) aims to match a query image against references from an extensive database of images from different places, relying solely on visual cues. State-of-the-art pipelines focus ... | • Uses DINOV2 ViT as backbone for feature extraction
• Applies Optimal Transport-based SALAD aggregation to form global descriptor
• Incorporates global context and controls cluster size with learned projection
• Replaces per-row normalization with row & column normalization using Sinkhorn Algorithm |
Generate 3 bullet points for the "Research Motivation / Problem Background" section of an academic poster. | ## Paper Content
# Paper Title
PoseBH: Prototypical Multi-Dataset Training Beyond Human Pose Estimation
## Abstract
We study multi-dataset training (MDT) for pose estimation, where skeletal heterogeneity presents a unique challenge that existing methods have yet to address. In traditional domains, e.g. regression a... | • Pose estimation on multiple datasets
• Unify diverse skeletons without dataset-dependent learnable parameters for high transferability
• Supervise unlabeled skeletons without intense computational overhead (e.g. distillation or augmentation) |
Generate 3 bullet points for the "Experimental Results / Performance Analysis" section of an academic poster. | ## Paper Content
# Paper Title
Stabilizing and Accelerating Autofocus with Expert Trajectory Regularized Deep Reinforcement Learning
## Abstract
Autofocus is a crucial component of modern digital cameras. While recent learning-based methods achieve state-of-the-art in focus prediction accuracy, they unfortunately i... | • Compares multiple methods (DRL [6], DRL [54], RoI-PE & Len-PE [7], [7] with relative label, Expert Trajectory Ours) across steps 1–4.
• Metrics include MAE↓, RMSE↓, FHL↓, and step-wise accuracy (≤0, ≤1, ≤2, ≤4).
• Our method (Expert Trajectory Ours) achieves the best performance across most metrics, especially in red... |
Generate 3 bullet points for the "Core Method / Technical Approach" section of an academic poster. | ## Paper Content
# Paper Title
Binarized Low-light Raw Video Enhancement
## Abstract
Recently, deep neural networks have achieved excellent performance on low-light raw video enhancement. However, they often come with high computational complexity and large memory costs, which hinder their applications on resource-... | • The BRVE model uses multiple Binary U-Nets with Shift Binary U-Nets for spatial-temporal feature alignment.
• The architecture includes Binary Conv Blocks and Shift Encoder/Decoder modules for efficient processing.
• A Binary Fusion Block and Shift Encoder/Decoder integrate features across frames. |
Generate 3 bullet points for the "Experimental Results / Performance Analysis" section of an academic poster. | ## Paper Content
# Paper Title
Motions as Queries: One-Stage Multi-Person Holistic Human Motion Capture
## Abstract
Existing methods for capturing multi-person holistic human motions from a monocular video usually involve integrating the detector, the tracker, and the human pose & shape estimator into a cascaded sy... | • Ours outperforms baselines including Fast-RCNN, Multi-HMR, and ByteTrack.
• Achieves 91.51 ID%, 97.01 MOTA, 79.88 MPJPE.
• Inference time is 0.027s, significantly faster than baselines. |
Generate 5 bullet points for the "Core Method / Technical Approach" section of an academic poster. | ## Paper Content
# Paper Title
UNEM: UNrolled Generalized EM for Transductive Few-Shot Learning
## Abstract
Transductive few-shot learning has recently triggered wide attention in computer vision. Yet, current methods introduce key hyper-parameters, which control the prediction statistics of the test batches, such ... | • {z_n} are feature vectors extracted from a pre-training network.
• N total samples to be classified within K distinct classes.
• S and Q are indices of support and query samples.
• Goal: identify classes of unlabeled samples by optimizing a general clustering objective via unrolling iterative block coordinate optimiz... |
Generate 2 bullet points for the "Research Motivation / Problem Background" section of an academic poster. | ## Paper Content
# Paper Title
KNOWING YOUR TARGET: TARGET-AWARE TRANSFORMER MAKES BETTER SPATIO-TEMPORAL VIDEO GROUNDING
## Abstract
Transformer has attracted increasing interest in spatio-temporal video grounding, or STVG, owing to its end-to-end pipeline and promising result. Existing Transformer-based STVG appr... | • Existing Transformer-based STVG methods use zero-initialized queries, which struggle with discriminative target localization in complex scenes due to lack of target-specific semantic cues.
• Our Target-Aware Transformer STVG Model uses target-aware queries that explore target-specific cues from the video-text pair to... |
Generate 4 bullet points for the "Core Method / Technical Approach" section of an academic poster. | ## Paper Content
# Paper Title
Repurposing Diffusion-Based Image Generators for Monocular Depth Estimation
## Abstract
Monocular depth estimation is a fundamental computer vision task. Recovering 3D depth from a single image is geometrically ill-posed and requires scene understanding, so it is not surprising that t... | • Encodes input RGB to latent space
• Initializes depth latent with standard noise
• Iteratively denoises depth latent using Latent Diffusion U-Net
• Decodes final depth latent to output depth map |
Generate 3 bullet points for the "Method Overview / Framework" section of an academic poster. | ## Paper Content
# Paper Title
DECENTRALIZED SPORADIC FEDERATED LEARNING: A UNIFIED ALGORTHMIC FRAMEWORK WITH CONVERGENCE GUARANTEES
## Abstract
Decentralized federated learning (DFL) captures FL settings where both (i) model updates and (ii) model aggregations are exclusively carried out by the clients without a c... | • Proposes a Sporadic DFL framework capturing resource heterogeneity and dynamics.
• Provides convergence analysis under mild assumptions for convex and non-convex settings.
• Conducts experiments in heterogeneous and time-varying DFL settings, referenced in Figures 2, 3 & 4. |
Generate 2 bullet points for the "Experimental Results / Performance Analysis" section of an academic poster. | ## Paper Content
# Paper Title
T2IShield: Defending Against Backdoors on Text-to-Image Diffusion Models
## Experiments
Attack Models. We consider two types of attack models in the experiment, where Rickrolling [43] leverages the vulnerability of the text encoder (i.e., CLIP [37]) and Villan Diffusion [8] leverages ... | • We mitigate the model by utilizing Refact and UCE.
• Shows infected samples, outputs before mitigation, benign outputs, and mitigated outputs by Refact and UCE. |
Generate 3 bullet points for the "Core Method / Technical Approach" section of an academic poster. | ## Paper Content
# Paper Title
Seamless Human Motion Composition with Blended Positional Encodings
## Abstract
Conditional human motion generation is an important topic with many applications in virtual reality, gaming, and robotics. While prior works have focused on generating motion guided by text, music, or scen... | • APE stage: each action generated independently; builds global dependencies matching condition.
• RPE stage: strong time-invariant motion prior emerges; preserves motion realism and smoothness.
• Training: randomly alternate APE/RPE modes. |
Generate 2 bullet points for the "Background / Related Work" section of an academic poster. | ## Paper Content
# Paper Title
UniMODE: Unified Monocular 3D Object Detection
## Abstract
Realizing unified monocular 3D object detection, including both indoor and outdoor scenes, holds great importance in applications like robot navigation. However, involving various scenarios of data to train models poses challe... | • Two-Stage Detection Architecture: Utilizing the first stage network to inform the second stage about rough target distribution, the training is stabilized.
• To address the grid size conflict between indoor and outdoor scenes, we propose the uneven BEV grid. |
Generate 3 bullet points for the "Research Motivation / Problem Background" 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... | • Mainstream image-to-3D methods learn 3D distribution but lack generality and control.
• Procedural Content Generation (PCG) offers high-quality, controllable outputs but is hard to control and not general.
• Inverse PCG combines strengths: uses PCG parameters as denoising variables for diffusion models. |
Generate 3 bullet points for the "Research Motivation / Problem Background" section of an academic poster. | ## Paper Content
# Paper Title
Event Camera Data Dense Pre-training
## Introduction
An event camera asynchronously records pixel-wise brightness changes of a scene [18]. In contrast to conventional RGB cameras that capture all pixel intensities at a fixed frame rate, event cameras offer a high dynamic range and mic... | • Objective: Pre-training a network to benefit diverse event camera data-based dense prediction tasks.
• Challenges: Event data is spatially sparse, patch feature learning suffers from event noise, and event data is prone to lead model collapse.
• Contributions: An event data dense pre-training framework; a context-lev... |
Generate 5 bullet points for the "Core Method / Technical Approach" section of an academic poster. | ## Paper Content
# Paper Title
DiffMorpher: Unleashing the Capability of Diffusion Models for Image Morphing
## Abstract
Diffusion models have achieved remarkable image generation quality surpassing previous generative models. However, a notable limitation of diffusion models, in comparison to GANs, is their diffic... | • Train LoRA on UNet with prompts for input images (e.g., 'A photo of a cat', 'A photo of a rabbit')
• Use DDIM inversion to obtain latent representations
• Perform LoRA interpolation between latent noise and adjust with AdaIN
• Apply self-attention interpolation & replacement to enhance textual smoothness
• Use resche... |
Generate 3 bullet points for the "Ablation Study" section of an academic poster. | ## Paper Content
# Paper Title
Learning to Highlight Audio by Watching Movies
## Abstract
Recent years have seen a significant increase in video content creation and consumption. Crafting engaging content requires the careful curation of both visual and audio elements. While visual cue curation, through techniques ... | • Ablation study comparing performance when using text captions versus video as conditioning signals.
• Results show that text outperforms video in some cases, suggesting flexibility in conditioning modalities.
• The table includes metrics for different numbers of players, parameters, and context types (video vs. text)... |
Generate 3 bullet points for the "Core Method / Technical Approach" section of an academic poster. | ## Paper Content
# Paper Title
MultiOOD: Scaling Out-of-Distribution Detection for Multiple Modalities
## Abstract
Detecting out-of-distribution (OOD) samples is important for deploying machine learning models in safety-critical applications such as autonomous driving and robot-assisted surgery. Existing research h... | • Introduces Agree-to-Disagree (A2D): modalities agree on ground-truth class, disagree on others by maximizing prediction distance.
• A2D training amplifies modality prediction discrepancy, enhancing OOD detection.
• Nearest Neighbor Prototype-based Mixup (NP-Mix) synthesizes outliers by leveraging nearest neighbor cla... |
Generate 3 bullet points for the "Method Overview / Framework" section of an academic poster. | ## Paper Content
# Paper Title
RISurConv: Rotation Invariant Surface Attention-Augmented Convolutions for 3D Point Cloud Classification and Segmentation
## Introduction
Point cloud has become the most promising 3D data representation for a wide range of immersive applications from robot navigation to autonomous dri... | • Highly expressive Rotation Invariant Surface Properties are constructed which can completely describe the local surface structure
• RISurConv: a novel attention-augmented convolution operator that is agnostic to both point cloud rotations and point orders
• Extensive experiments show supreme performance surpassing th... |
Generate 1 bullet points for the "Experimental Results / Performance Analysis" section of an academic poster. | ## Paper Content
# Paper Title
Image Neural Field Diffusion Models
## Abstract
Diffusion models have shown an impressive ability to model complex data distributions, with several key advantages over GANs, such as stable training, better coverage of the training distribution's modes, and the ability to solve inverse... | • Can learn from mixed-resolution training data and generate more details |
Generate 3 bullet points for the "Method Overview / Framework" section of an academic poster. | ## Paper Content
# Paper Title
All in One Framework for Multimodal Re-identification in the Wild
## Abstract
In Re-identification (ReID), recent advancements yield noteworthy progress in both unimodal and cross-modal retrieval tasks. However, the challenge persists in developing a unified framework that could effec... | • First framework could handle all four modalities
• Support any combination of multimodal inputs
• Competitive performance on both cross-modal and multimodal ReID tasks |
Generate 3 bullet points for the "Implementation Details" section of an academic poster. | ## Paper Content
# Paper Title
4Deform: Neural Surface Deformation for Robust Shape Interpolation
## Abstract
Generating realistic intermediate shapes between non-rigidly deformed shapes is a challenging task in computer vision, especially with unstructured data (e.g., point clouds) where temporal consistency acros... | • Compares training time of our method with baselines.
• Our method is significantly faster than implicit-based methods.
• Training time is under 100 minutes for our method. |
Generate 3 bullet points for the "Ablation Study" section of an academic poster. | ## Paper Content
# Paper Title
Rethinking Unsupervised Outlier Detection via Multiple Thresholding
## Experiments
In our approach, we employ LVAD [34], one of the UOD SOTAs, as the initial outlier score function. LVAD assumes each image feature $\mathbf{x}_i \in \mathbf{X}$ comes from one of $T$ high-dimensiona... | • Evaluates Multi-T integrated with various scoring functions (IF, ECOD, ABOD, etc.) on STL-10.
• Multi-T consistently improves AUC across all backbones (ResNet, CLIP) and scoring methods.
• Average improvement: 1.5–3.5% across methods, demonstrating generalizability. |
Generate 3 bullet points for the "Core Method / Technical Approach" section of an academic poster. | ## Paper Content
# Paper Title
Do Visual Imaginations Improve Vision-and-Language Navigation Agents?
## Abstract
Vision-and-Language Navigation (VLN) agents are tasked with navigating an unseen environment using natural language instructions. In this work, we study if visual representations of sub-goals implied by ... | • Instructions are segmented into sub-instructions, filtering out vague noun phrases like 'it' or 'right'.
• Valid sub-instructions are used as prompts to generate synthetic images via SDXL.
• Example: 'Go down the stairs and then turn right' generates staircase image; 'Go stand behind the easel by a unicycle' generate... |
Generate 1 bullet points for the "Research Motivation / Problem Background" section of an academic poster. | ## Paper Content
# Paper Title
SocialCircle: Learning the Angle-based Social Interaction Representation for Pedestrian Trajectory Prediction
## Abstract
Analyzing and forecasting trajectories of agents like pedestrians and cars in complex scenes has become more and more significant in many intelligent systems and a... | • Inspired by Marine Animals that localize others objects and communicate with their companions underwater through Echolocation, we build a new angle-based trainable social interaction representation, named SocialCircle, for continuously reflecting the context of social interactions at different Angular Orientations. |
Generate 3 bullet points for the "Ablation Study" section of an academic poster. | ## Paper Content
# Paper Title
Minimal Perspective Autocalibration
## Abstract
We introduce a new family of minimal problems for reconstruction from multiple views. Our primary focus is a novel approach to autocalibration, a long-standing problem in computer vision. Traditional approaches to this problem, such as t... | • Tests impact of vanishing point accuracy on calibration
• Evaluates performance with varying numbers of point correspondences
• Measures sensitivity to initial parameter guesses |
Generate 2 bullet points for the "Core Method / Technical Approach" section of an academic poster. | ## Paper Content
# Paper Title
CADDreamer: CAD Object Generation from Single-view Images
## Abstract
Diffusion-based 3D generation has made remarkable progress in recent years. However, existing 3D generative models often produce overly dense and unstructured meshes, which stand in stark contrast to the compact, st... | • Uses Wonder3D to generate multi-view normal and semantic maps.
• Compares Wonder3D (normal + image) with Ours (normal + semantic maps) for better segmentation. |
Generate 5 bullet points for the "Core Method / Technical Approach" section of an academic poster. | ## Paper Content
# Paper Title
Polynomial-Time Computation of Exact $\Phi$ -Equilibria in Polyhedral Games
## Abstract
It is a well-known fact that correlated equilibria can be computed in polynomial time in a large class of concisely represented games using the celebrated Ellipsoid Against Hope algorithm (Papadim... | • Minimax theorem: min_y max_x x^T A y = max_x min_y x^T A y = 0 (WLOG).
• Given y ∈ Y, find x ∈ X such that x^T A y ≥ 0 (Good-Enough-Response, GER).
• Computing best-response x' is NP-hard in general.
• Use GER for meta-game: compute fixed points x_p = φ_p(x_p) for tuple y = (φ_1, ..., φ_n).
• Final solution x_* is a ... |
Generate 3 bullet points for the "Research Motivation / Problem Background" section of an academic poster. | ## Paper Content
# Paper Title
All-directional Disparity Estimation for Real-world QPD Images
## Abstract
Quad Photodiode (QPD) sensors represent an evolution by providing four sub-views, whereas dual-pixel (DP) sensors are limited to two sub-views. In addition to enhancing autofocus performance, QPD sensors also e... | • QPD sensors advance upon DP technology by integrating four same-color pixels under one on-chip lens.
• We explore disparity estimation using QPD sensors, widely adopted in mobile devices like OV50A and OV50H.
• QPD sensors provide absolute depth with single camera and high precision within range. |
Generate 4 bullet points for the "Method Overview / Framework" section of an academic poster. | ## Paper Content
# Paper Title
SIGNATURE KERNEL CONDITIONAL INDEPENDENCE TESTS IN CAUSAL DISCOVERY FOR STOCHASTIC PROCESSES
## Abstract
Inferring the causal structure underlying stochastic dynamical systems from observational data holds great promise in domains ranging from science and health to finance. Such proce... | • Develop CI constraints on paths generated by SDEs to prove Markov with respect to acyclic dependence.
• Introduce efficient constraint-based causal discovery algorithm, proven sound and complete.
• Propose practical and consistent kernel-based CI test on path-space with signature.
• Benchmark test in isolation and as... |
Generate 4 bullet points for the "Background / Related Work" section of an academic poster. | ## Paper Content
# Paper Title
Dataset Distillation by Automatic Training Trajectories
## Introduction
Deep learning has showcased remarkable achievements across various computer vision problems [24, 43, 56]. Nevertheless, its success often hinges on extensive datasets, resulting in substantial computational demand... | • Dataset Distillation (DD): Condense large dataset into small dataset for training purpose
• Short-range matching methods
• Long-range matching methods (kernel ridge regression-based methods included)
• Long-Range Matching Dataset Distillation: [diagram reference] |
Generate 3 bullet points for the "Core Method / Technical Approach" section of an academic poster. | ## Paper Content
# Paper Title
Exploring the Deep Fusion of Large Language Models and Diffusion Transformers for Text-to-Image Synthesis
## Abstract
This paper does not describe a new method; instead, it provides a thorough exploration of an important yet understudied design space related to recent advances in text... | • Compares 1D-RoPE + APE, 1D + 2D-RoPE, and M-RoPE for input sequence encoding.
• 1D + 2D-RoPE achieves [2,2] for IMG1 and [2,3] for IMG2.
• M-RoPE uses [0,0] for TXT1 and [1,1] for TXT2. |
Generate 3 bullet points for the "Background / Related Work" section of an academic poster. | ## Paper Content
# Paper Title
Cloud-Device Collaborative Learning for Multimodal Large Language Models
## Abstract
The burgeoning field of Multimodal Large Language Models (MLLMs) has exhibited remarkable performance in diverse tasks such as captioning, commonsense reasoning, and visual scene understanding. Howeve... | • Multimodal Large Language Models (MLLMs) show strong performance in tasks like captioning and visual reasoning.
• Deploying large MLLMs on client devices is hindered by their extensive parameters, causing generalization decline when compressed.
• We introduce a Cloud-Device Collaborative Continual Adaptation framewor... |
Generate 4 bullet points for the "Method Overview / Framework" section of an academic poster. | ## Paper Content
# Paper Title
Human-centered Interactive Learning via MLLMs for Text-to-Image Person Re-identification
## Abstract
Despite remarkable advancements in text-to-image person re-identification (TIREID) facilitated by the breakthrough of cross-modal embedding models, existing methods often struggle to d... | • Text-to-image person re-identification (TIRelD) retrieves target person images using natural language queries.
• Existing methods struggle with distinguishing challenging candidates due to network architecture and data quality.
• We propose Interactive Cross-modal Learning (ICL) leveraging human-centered interaction ... |
Generate 3 bullet points for the "Core Method / Technical Approach" section of an academic poster. | ## Paper Content
# Paper Title
Flopping for FLOPs: Leveraging Equivalence for Computational Efficiency
## Abstract
Incorporating geometric invariance into neural networks enhances parameter efficiency but typically increases computational costs. This paper introduces new equivariant neural networks that preserve sy... | • Splits tokens into invariants and (-1)-equivariants.
• Linear layers split using Schur's lemma, reducing FLOPs and parameters by 1/2.
• Modifications to non-linearities are computationally cheap by transforming to spatial domain. |
Generate 2 bullet points for the "Implementation Details" section of an academic poster. | ## Paper Content
# Paper Title
Think Small, Act Big: Primitive Prompt Learning for Lifelong Robot Manipulation
## Abstract
Building a lifelong robot that can effectively leverage prior knowledge for continuous skill acquisition remains significantly challenging. Despite the success of experience replay and paramete... | • Shows real-world and simulation setups for tasks including Grasp the banana, Place the banana to the basket, Grasp the block, Place the block to the blanket, Push down the teapot handle, Place the croissant, Place the drink on cutting board.
• Demonstrates applicability in both simulated and real environments. |
Generate 3 bullet points for the "Core Method / Technical Approach" section of an academic poster. | ## Paper Content
# Paper Title
DriveGEN: Generalized and Robust 3D Detection in Driving via Controllable Text-to-Image Diffusion Generation
## Abstract
In autonomous driving, vision-centric 3D detection aims to identify 3D objects from images. However, high data collection costs and diverse real-world scenarios lim... | • Perform PCA to obtain semantic components {P_i}_{i=1}^{N_x} of self-attention features for g_sa.
• Further introduce shallow feature alignment g_sh to preserve fine-grained object details.
• Total scheme of DriveGEN: ε_t = (2 - s) ε_θ(z_t; t, c) - s ε_θ(z_t; t, c̄) + g_sa + g_sh. |
Generate 3 bullet points for the "Experimental Results / Performance Analysis" section of an academic poster. | ## Paper Content
# Paper Title
Bayesian Self-Training for Semi-Supervised 3D Segmentation
## Experiments
We evaluate our method on all three semi-supervised 3D semantic perception tasks: 3D semantic segmentation, 3D instance segmentation, and dense 3D visual grounding. The dataset and implementation details regardi... | • Evaluates visual grounding performance on ScanRefer.
• Shows qualitative examples of object localization based on natural language descriptions.
• Our method correctly grounds descriptions like 'the table is square' with surrounding chairs. |
Generate 3 bullet points for the "Research Motivation / Problem Background" section of an academic poster. | ## Paper Content
# Paper Title
Behind the Veil: Enhanced Indoor 3D Scene Reconstruction with Occluded Surfaces Completion
## Abstract
In this paper, we present a novel indoor 3D reconstruction method with occluded surface completion, given a sequence of depth readings. Prior state-of-the-art (SOTA) methods only foc... | • Most existing methods only focus on reconstructing visible areas in a scene, neglecting invisible areas due to occlusions.
• Input: a sequence of depth images and associated camera poses.
• Goal: reconstruct the 3D scene geometry with occluded surface completion. |
Generate 2 bullet points for the "Method Overview / Framework" section of an academic poster. | ## Paper Content
# Paper Title
DiET-GS: Diffusion Prior and Event Stream-Assisted Motion Deblurring 3D Gaussian Splatting
## Abstract
Reconstructing sharp 3D representations from blurry multi-view images is a long-standing problem in computer vision. Recent works attempt to enhance high-quality novel view synthesis... | • A novel framework to construct deblurring 3DGS by jointly leveraging event streams and the prior knowledge of a pretrained diffusion model.
• A two-stage training strategy to effectively utilize real-captured data and diffusion prior together. Once optimized, our method is capable of recovering well-defined details w... |
Generate 3 bullet points for the "Research Motivation / Problem Background" section of an academic poster. | ## Paper Content
# Paper Title
RepKPU: Point Cloud Upsampling with Kernel Point Representation and Deformation
## Abstract
In this work, we present RepKPU, an efficient network for point cloud upsampling. We propose to promote upsampling performance by exploiting better shape representation and point generation str... | • Abundant geometric information: different kernel points can represent distinct geometric patterns thanks to geometry-aware weight averaging and separated convolutional weights
• Flexible deformation: the positions of kernel points can adapt to local geometry, as revealed in deformable KPConv
• Inspired by this, we re... |
Generate 3 bullet points for the "Core Method / Technical Approach" section of an academic poster. | ## Paper Content
# Paper Title
Intriguing Properties of Diffusion Models: An Empirical Study of the Natural Attack Capability in Text-to-Image Generative Models
## Abstract
Denoising probabilistic diffusion models have shown breakthrough performance to generate more photo-realistic images or human-level illustratio... | • Covers 7 diffusion models (e.g., DALL-E 3 & Firefly) and 15 object classes.
• Demonstrates attacks on stop signs, fire hydrants, and horses by manipulating shape, color, text, and pattern features.
• Shows combinations of features (e.g., Shape & color, Text & pattern) that still fool object detectors. |
Generate 3 bullet points for the "Research Motivation / Problem Background" section of an academic poster. | ## Paper Content
# Paper Title
Masked AutoDecoder is Effective Multi-Task Vision Generalist
## Abstract
Inspired by the success of general-purpose models in NLP, recent studies attempt to unify different vision tasks in the same sequence format and employ autoregressive Transformers for sequence prediction. They ap... | • Recent Vision generalists unify different vision tasks in the same sequence format and employ Autoregressive Frameworks for sequence prediction.
• However, Vision task sequences (Object Detection, Instance Segmentation, etc.) don’t heavily exhibit Sequential Dependencies as language sequences, leading to inferior per... |
Generate 3 bullet points for the "Core Method / Technical Approach" 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 ... | • Uses principal component GD to compress feature domain dimensionality.
• Achieves high image fidelity with fewer operations.
• Includes PCPGD, CDPGD, and PAM modules for efficient reconstruction. |
Generate 2 bullet points for the "Core Method / Technical Approach" section of an academic poster. | ## Paper Content
# Paper Title
How Far Can We Compress Instant-NGP-Based NeRF?
## Abstract
In recent years, Neural Radiance Field (NeRF) has demonstrated remarkable capabilities in representing 3D scenes. To expedite the rendering process, learnable explicit representations have been introduced for combination with... | • Level-wise Context Models (Dashed blue box): we first find the vertex n_i of the feature vector θ_i using hash function and then estimate its distribution probability p_i using a Context Fuser C_p from the aggregated contexts from previously decoded levels.
• Level-wise Context Models (Dashed orange box): the last le... |
Generate 3 bullet points for the "Background / Related Work" section of an academic poster. | ## Paper Content
# Paper Title
Positive2Negative: Breaking the Information-Lossy Barrier in Self-Supervised Single Image Denoising
## Abstract
Image denoising enhances image quality, serving as a foundational technique across various computational photography applications. The obstacle to clean image acquisition in... | • The noise distribution is zero-mean and approximately symmetrical.
• The opposite noise −n also follows the same distribution as that of the original noise n.
• Shows noisy image, green circle distribution, and red circle distribution plots. |
Generate 4 bullet points for the "Experimental Results / Performance Analysis" section of an academic poster. | ## Paper Content
# Paper Title
Open-World Objectness Modeling Unifies Novel Object Detection
## Abstract
The challenge in open-world object detection, similarly to few- and zero-shot learning, is to generalize beyond the class distribution of the training data. In this paper, we propose a general class-agnostic obj... | • Comparison on M-OWODB and S-OWODB using U-Recall and mAP@0.5.
• Results on OV-LVIS benchmark with extra data and pre-train models.
• Results on FSOD across varying K shots.
• t-SNE visualization of query embeddings from OWOBJ variants. |
Generate 3 bullet points for the "Background / Related Work" section of an academic poster. | ## Paper Content
# Paper Title
TABRED: ANALYZING PITFALS AND FILLING THE GAPS IN TABULAR DEEP LEARNING BENCHMARKS
## Abstract
Advances in machine learning research drive progress in real-world applications. To ensure this progress, it is important to understand the potential pitfalls on the way from a novel method'... | • Research on Tabular Deep Learning is making steady progress with Transformers, MLP baselines (2021), GBDTs (2014–2017), and Numerical Embeddings like TabPFN (2022).
• Tabular data is very diverse; validating that a new SoTA works on real-world issues (not just academic benchmarks) is critical.
• ML engineers face iss... |
Generate 3 bullet points for the "Experimental Results / Performance Analysis" section of an academic poster. | ## Paper Content
# Paper Title
Noisy-Correspondence Learning for Text-to-Image Person Re-identification
## Abstract
Text-to-image person re-identification (TIREID) is a compelling topic in the cross-modal community, which aims to retrieve the target person based on a textual query. Although numerous TIREID methods ... | • Evaluates on CUHK-PEDES, ICFG-PEDES, and RSTPReid datasets.
• Reports R-1, R-5, R-10, mAP, and mINP under 0%, 20%, and 50% noise levels.
• Our method consistently outperforms baselines across all metrics and noise conditions. |
Generate 3 bullet points for the "Ablation Study" section of an academic poster. | ## Paper Content
# Paper Title
SV3D: Novel Multi-view Synthesis and 3D Generation from a Single Image using Latent Video Diffusion
## Experiments
Reconstruction via Photometric Losses. Intuitively, we can treat the SV3D-generated images as multi-view pseudo-ground truth, and apply 2D reconstruction losses to train ... | • We learn to reconstruct the albedo instead of the radiance
• Explicit Spherical Gaussian (SG) Illumination model during reconstruction
• Helps remove low-frequency shading effects |
Generate 3 bullet points for the "Background / Related Work" section of an academic poster. | ## Paper Content
# Paper Title
Flexible Group Count Enables Hassle-Free Structured Pruning
## Abstract
Densely structured pruning methods — which generate pruned models in a fully dense format, allowing immediate compression benefits without additional demands — are evolving due to their practical significance. Tra... | • TMI-GKP allows dynamic clustering schemes per layer.
• Some schemes (e.g., k-PCA) are expensive to run.
• Requires training snapshots or checkpoints, making it not user-friendly in practice. |
Generate 2 bullet points for the "Ablation Study" section of an academic poster. | ## Paper Content
# Paper Title
FreeInit: Bridging Initialization Gap in Video Diffusion Models
## Task
Section Title: Influence of Low-frequency Components at Inference
Generate poster content for this section. | • Low-frequencies of initial noise are crucial for the generated content and quality.
• Removing increasing percentages (20% to 100%) of high frequencies degrades frame quality progressively. |
Generate 2 bullet points for the "Method Overview / Framework" section of an academic poster. | ## Paper Content
# Paper Title
UniHuman: A Unified Model For Editing Human Images in the Wild
## Abstract
Human image editing includes tasks like changing a person's pose, their clothing, or editing the image according to a text prompt. However, prior work often tackles these tasks separately, overlooking the benef... | • Goal: Change a person’s pose, clothing, or edit human image via text prompt.
• Demonstrates three editing types: Reposing, Virtual Try-on, Text Manipulation. |
Generate 3 bullet points for the "Core Method / Technical Approach" section of an academic poster. | ## Paper Content
# Paper Title
R-SCoRe: Revisiting Scene Coordinate Regression for Robust Large-Scale Visual Localization
## Abstract
Learning-based visual localization methods that use scene coordinate regression (SCR) offer the advantage of smaller map sizes. However, on datasets with complex illumination changes... | • Defines e2(x,y) as L2 norm of reprojection error.
• Introduces e3(x,y) as depth-adjusted error using sigma_o.
• Sigma_o combines depth uncertainty and fixed noise. |
Generate 1 bullet points for the "Other Content" section of an academic poster. | ## Paper Content
# Paper Title
Modality-agnostic Domain Generalizable Medical Image Segmentation by Multi-Frequency in Multi-Scale Attention
## Abstract
Generalizability in deep neural networks plays a pivotal role in medical image segmentation. However, deep learning-based medical image analyses tend to overlook t... | • LinkedIn and GitHub QR codes for further information and code access. |
Generate 3 bullet points for the "Core Method / Technical Approach" section of an academic poster. | ## Paper Content
# Paper Title
TIGHT LOWER BOUNDS UNDER ASYMMETRIC HIGH-ORDER HÖLDER SMOOTHNESS AND UNIFORM CONVEXITY
## Abstract
In this paper, we provide tight lower bounds for the oracle complexity of minimizing high-order Hölder smooth and uniformly convex functions. Specifically, for a function whose $p^{th}$... | • Orthogonal basis: v₁ ⊥ x₁, ..., xₜ and v₁, ..., vᵢ₊₁ ⊥ vᵢ, ∀ i ∈ [T̃].
• [Arjevani et al., 2019] q = 2, p = 2, ν = 1.
• [Our Work] f(x) = (H/2^(p+ν+1)(p+ν-1)!) * (1/(p+ν) * Σᵢ₌₁^p |⟨vᵢ, x⟩ - ⟨vᵢ₊₁, x⟩|^(p+ν) - γ⟨v₁, x⟩) + (σ/q)‖x‖^q. |
Generate 5 bullet points for the "Experimental Results / Performance Analysis" section of an academic poster. | ## Paper Content
# Paper Title
Make Me a BNN: A Simple Strategy for Estimating Bayesian Uncertainty from Pre-trained Models
## Abstract
Deep Neural Networks (DNNs) are powerful tools for various computer vision tasks, yet they often struggle with reliable uncertainty quantification — a critical requirement for real... | • ABNN improves uncertainty quantification with small computational overhead.
• Most gains are linked to improved epistemic uncertainty (measured by OOD detection).
• ABNN scales to larger datasets like ImageNet and across architectures.
• ABNN strikes a good balance between accuracy and uncertainty quantification perf... |
Generate 3 bullet points for the "Research Motivation / Problem Background" section of an academic poster. | ## Paper Content
# Paper Title
PhyT2V: LLM-Guided Iterative Self-Refinement for Physics-Grounded Text-to-Video Generation
## Abstract
Text-to-video (T2V) generation has been recently enabled by transformer-based diffusion models, but current T2V models lack capabilities in adhering to the real-world common knowledg... | • Generated T2V videos often contain physical illusions or artifacts due to model limitations.
• Examples include cream swirling into milk, eggs breaking, and objects defying gravity.
• These failures reflect the model’s inability to generate realistic content under unfamiliar conditions. |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.