Depth Anything V2
PulseAugur coverage of Depth Anything V2 — every cluster mentioning Depth Anything V2 across labs, papers, and developer communities, ranked by signal.
2 day(s) with sentiment data
-
DA360 model enhances 360-degree depth estimation with scale invariance
Researchers have developed DA360, a panoramic adaptation of the Depth Anything V2 model, to improve 360-degree depth estimation. This new framework leverages the existing DAV2 model's zero-shot generalization capabiliti…
-
New methods PixelDense and Persistence Forcing boost diffusion model performance
Researchers have developed two novel techniques to enhance pixel-space diffusion models. PixelDense improves training by aligning semantic and geometric features separately, leading to better performance on tasks like i…
-
DriveZero autonomous driving system learns beyond human demonstrations
Researchers have introduced DriveZero, an end-to-end autonomous driving system that moves beyond imitating human driving data. DriveZero separates the driving task into a perception model (DriveVFM) and an action model …
-
New method distills reliable geometry for depth-free object detection
Researchers have developed a novel framework called \"method\" for RGB-D salient object detection that overcomes the limitations of unreliable sensor depth data. This approach uses a pre-trained Depth Anything V2 model …
-
AI model adapted for precise lunar surface height estimation
Researchers have adapted the Depth Anything V2 foundation model to improve lunar surface height estimation for future space missions. By fine-tuning the model with existing stereophotogrammetry-derived digital elevation…
-
2D vision models outperform 3D-aware counterparts in vehicle attribute recognition
A new paper evaluates 14 state-of-the-art 2D and 3D-aware vision foundation models for vehicle attribute recognition. The study found that standard 2D self-supervised models, particularly DINOv3, performed better than 3…
-
New AI model estimates hurricane debris volume from aerial images
Researchers have developed DebrisHeightNet, a novel system for estimating hurricane debris volume from single aerial RGB images. This method uses a lightweight network built upon frozen foundation models, regressing hei…
-
Sign language recognition models use synthetic depth images
Researchers have developed new models for sign language recognition using point clouds derived from depth images. The study compared classification accuracies using PointNet architectures with both original and syntheti…
-
New ControlNet method integrates pose and depth into FLUX.2 models
A new method has been developed to integrate pose and depth control into the FLUX.2 family of image generation models. This approach utilizes existing ControlNet models for pose, depth, or edge mapping, which are then f…
-
New methods enhance monocular depth estimation in challenging scenarios
Researchers have developed new methods to improve monocular depth estimation (MDE) in challenging visual scenarios. One approach, CapDepth, utilizes detailed long captions to guide depth decoding, achieving significant …
-
New "3D Mirage" failure mode identified in monocular depth models
Researchers have identified a new failure mode in monocular depth estimation models, termed the "3D Mirage." This phenomenon occurs when models hallucinate illusory 3D structures from ambiguous inputs, despite achieving…
-
ComfyUI node converts 2D images to 3D for music video creation
A user has created a music video using a new ComfyUI node that converts 2D images into 3D. This node, called ComfyUI-Y7-SBS-2Dto3D, utilizes depth-anything-v2 to generate depth estimations and then creates side-by-side …
-
Micro-UAVs navigate indoors using only monocular vision for search and rescue
Researchers have developed a novel indoor navigation system for micro-unmanned aerial vehicles (UAVs) designed for search-and-rescue operations. This system, named TRISTAR, exclusively uses monocular vision, eliminating…
-
New method improves sparse-view neural reconstruction with selective depth supervision
Researchers have developed a method to improve sparse-view neural reconstruction in outdoor driving scenes by selectively applying monocular depth supervision. The technique uses Depth Anything V2 to provide dense geome…
-
New Krea-2 LoRA model allows depth-controlled image generation
A new LoRA model, Patil/Krea-2-depth-controlnet, has been released, enabling users to maintain the 3D structure of an image while altering its content and style through text prompts. This control is achieved by extracti…
-
New LoRA enables Stable Diffusion composition control; users seek Krea 2 integration
A user has developed and shared a LoRA (Low-Rank Adaptation) model called krea2_controlnet_lora for Stable Diffusion, which allows for composition control using a Depth Anything V2 map and prompts. Another user is inqui…
-
New soccer view synthesis method uses depth guidance and ensembles
Researchers have developed DENSER, a novel approach for synthesizing new views of soccer matches using depth guidance and staged reconstruction. This method incorporates camera-height-based loss weighting, monocular dep…
-
New depth completion model uses sparse radar data for outdoor environments
Researchers have developed a novel depth completion model that can accurately estimate dense depth maps in challenging outdoor environments using extremely sparse depth measurements, such as those from low-cost radar. T…
-
Robots navigate using AI-powered depth estimation, ditching LiDAR
Researchers have developed a novel teacher-student framework for robot navigation that replaces traditional LiDAR sensors with vision-based monocular depth estimation. A teacher policy, trained with privileged LiDAR dat…