The field of computer vision and 3D reconstruction has witnessed significant advancements in recent times, with a focus on developing innovative methods for generating high-quality images, videos, and 3D models. One notable trend is the use of diffusion models, which have shown great promise in image and video generation, as well as 3D reconstruction tasks. These models have been used to generate realistic images and videos, and have also been applied to tasks such as image inpainting, object removal, and novel view synthesis. Another area of research that has gained significant attention is the development of methods for 3D reconstruction from single or multi-view images. These methods have the potential to revolutionize fields such as architecture, product design, and robotics, where accurate 3D models are essential. Noteworthy papers in this area include TanDiT, which proposes a method for generating high-quality 360-degree panoramic images using a tangent-plane diffusion transformer, and AlignCVC, which introduces a framework for aligning cross-view consistency for single-image-to-3D generation. Additionally, methods such as MedDiff-FT and DiGA3D have been proposed for controllable medical image synthesis and versatile 3D inpainting, respectively. Overall, the field of computer vision and 3D reconstruction is rapidly evolving, with new methods and techniques being developed to address various challenges and applications. TanDiT is particularly noteworthy for its ability to generate high-quality panoramic images, while AlignCVC stands out for its ability to align cross-view consistency for single-image-to-3D generation. MedDiff-FT and DiGA3D are also notable for their applications in medical image synthesis and 3D inpainting, respectively.
Advances in Computer Vision and 3D Reconstruction
Sources
TanDiT: Tangent-Plane Diffusion Transformer for High-Quality 360{\deg} Panorama Generation
PrefPaint: Enhancing Image Inpainting through Expert Human Feedback
Quality Assessment and Distortion-aware Saliency Prediction for AI-Generated Omnidirectional Images
Single-Scanline Relative Pose Estimation for Rolling Shutter Cameras
MatChA: Cross-Algorithm Matching with Feature Augmentation
WarpRF: Multi-View Consistency for Training-Free Uncertainty Quantification and Applications in Radiance Fields
3D Shape Generation: A Survey
Single-Frame Point-Pixel Registration via Supervised Cross-Modal Feature Matching
Inpainting is All You Need: A Diffusion-based Augmentation Method for Semi-supervised Medical Image Segmentation
Unsupervised 3D Braided Hair Reconstruction from a Single-View Image
AlignCVC: Aligning Cross-View Consistency for Single-Image-to-3D Generation
MEMFOF: High-Resolution Training for Memory-Efficient Multi-Frame Optical Flow Estimation
PathDiff: Histopathology Image Synthesis with Unpaired Text and Mask Conditions
Time-variant Image Inpainting via Interactive Distribution Transition Estimation
GeoCD: A Differential Local Approximation for Geodesic Chamfer Distance
MTADiffusion: Mask Text Alignment Diffusion Model for Object Inpainting
ViewPoint: Panoramic Video Generation with Pretrained Diffusion Models
WAVE: Warp-Based View Guidance for Consistent Novel View Synthesis Using a Single Image
Towards Initialization-free Calibrated Bundle Adjustment
Refine Any Object in Any Scene
Puzzles: Unbounded Video-Depth Augmentation for Scalable End-to-End 3D Reconstruction
PriOr-Flow: Enhancing Primitive Panoramic Optical Flow with Orthogonal View
TextMesh4D: High-Quality Text-to-4D Mesh Generation
MVGBench: Comprehensive Benchmark for Multi-view Generation Models
MedDiff-FT: Data-Efficient Diffusion Model Fine-tuning with Structural Guidance for Controllable Medical Image Synthesis
Learning Dense Feature Matching via Lifting Single 2D Image to 3D Space
DiGA3D: Coarse-to-Fine Diffusional Propagation of Geometry and Appearance for Versatile 3D Inpainting
BEV-VAE: Multi-view Image Generation with Spatial Consistency for Autonomous Driving
TRACE: Temporally Reliable Anatomically-Conditioned 3D CT Generation with Enhanced Efficiency
Surgical Neural Radiance Fields from One Image
Semiautomatic Simplification
TurboReg: TurboClique for Robust and Efficient Point Cloud Registration
FreeMorph: Tuning-Free Generalized Image Morphing with Diffusion Model
DreamComposer++: Empowering Diffusion Models with Multi-View Conditions for 3D Content Generation
MAGIC: Mask-Guided Diffusion Inpainting with Multi-Level Perturbations and Context-Aware Alignment for Few-Shot Anomaly Generation
Mesh Silksong: Auto-Regressive Mesh Generation as Weaving Silk
MoGe-2: Accurate Monocular Geometry with Metric Scale and Sharp Details