Comprehensive and Precise Benchmark for Evaluating Biases in Text-to-Image Generation Models
FAIntbench provides a holistic and precise benchmark for evaluating various types of biases in text-to-image generation models, along with a specific bias definition system and a comprehensive dataset.
Fairness Implications of Low-Rank Adaptation of Large Models Across Vision and Language Domains
Low-rank adaptation (LoRA) can match the performance of full fine-tuning of large models while being computationally more efficient, but its fairness implications across subgroups are not well understood.
High-Quality 3D Reconstruction from Multi-View Images using Regularized Dipole Sums
The authors introduce a new point-based representation called the regularized dipole sum, which generalizes the winding number to enable efficient and robust 3D reconstruction from multi-view images through inverse rendering.
4개의 뷰에서 가우시안 스플래팅을 통한 고품질 3D 객체 재구성
희소한 4개의 뷰에서도 가우시안 스플래팅을 통해 고품질의 3D 객체 재구성이 가능하다.
High-Quality 3D Object Reconstruction from Four Views Using Gaussian Splatting
GaussianObject, a framework that can reconstruct high-quality 3D objects from only 4 input images using Gaussian splatting, significantly outperforms previous state-of-the-art methods.
Hyper-STTN: A Hypergraph-based Spatial-Temporal Transformer Network for Accurate Human Trajectory Prediction in Crowded Scenarios
Hyper-STTN leverages a hypergraph-based spatial-temporal transformer network to effectively capture both group-wise and pair-wise social interactions for accurate human trajectory prediction in crowded scenarios.
비디오 표현을 위한 망각 없는 승리 하위 네트워크
연속 학습 문제를 해결하기 위해 Fourier 하위 신경 연산자(FSO)를 제안하여 복잡한 주기 신호를 효율적으로 표현할 수 있다.
Goal-Guided Diffusion Model with Tree Sampling for Accurate and Efficient Multi-Modal Pedestrian Trajectory Prediction
The proposed GDTS framework integrates goal estimation and a novel two-stage tree sampling diffusion model to generate accurate and diverse multi-modal pedestrian trajectory predictions in real-time.
利用 Sentinel-2 衛星影像高精度偵測建築物和道路
本研究展示如何利用多張 10 公尺解析度的 Sentinel-2 衛星影像,生成 50 公分解析度的建築物和道路分割遮罩。這是透過訓練一個「學生」模型,使其能複製「老師」模型(使用高解析度影像)的預測結果而實現的。雖然預測結果沒有老師模型的細節,但仍能保留大部分的性能:建築物分割的平均交集比例(mIoU)為 79.0%,而高解析度老師模型的準確率為 85.5% mIoU。本研究還描述了兩種其他使用 Sentinel-2 影像的相關方法:一種是建築物計數,可達到 R2 = 0.91 的準確度;另一種是建築物高度預測,平均絕對誤差為 1.5 公尺。這項工作為使用免費的 Sentinel-2 影像進行各種以往只能使用高解析度衛星影像才能完成的任務開啟了新的可能性。
저해상도 Sentinel-2 위성 영상을 활용한 고해상도 건물 및 도로 탐지
저해상도 Sentinel-2 위성 영상을 활용하여 고해상도 건물 및 도로 탐지 모델을 개발하였으며, 이를 통해 기존에 고해상도 위성 영상이 필요했던 다양한 응용 분야에서 활용할 수 있게 되었다.
© 2024 by Linnk AI