ScanNetv2
PulseAugur coverage of ScanNetv2 — every cluster mentioning ScanNetv2 across labs, papers, and developer communities, ranked by signal.
1 day(s) with sentiment data
-
New PrePARE method slashes memory use in multi-view geometry transformers
Researchers have developed PrePARE, a novel method for optimizing multi-view geometry transformers by pruning patch tokens before they enter the alternating-attention (AA) stack. This approach significantly reduces memo…
-
New AQ3D method advances 3D instance segmentation with adaptive queries
Researchers have introduced AQ3D, a novel approach to 3D instance segmentation designed to adapt to scenes of varying sizes. Unlike previous methods that use a fixed number of queries, AQ3D instantiates queries based on…
-
New framework enhances 3D object detection with LLM-guided alignment
Researchers have developed a new framework for open-vocabulary 3D object detection, aiming to improve the accuracy of identifying unseen objects in 3D scenes. The proposed method enhances novel object discovery through …
-
LiteMVS model enhances real-time 3D perception with distilled knowledge
Researchers have developed LiteMVS, a lightweight multi-view stereo model designed for efficient real-time 3D perception. This model integrates geometric reasoning with strong monocular semantic and structural priors, d…
-
New UTok3D tokenizer adapts CLIP for 3D understanding tasks
Researchers have developed UTok3D, a novel parameter-efficient framework designed to adapt CLIP, a vision-language model, for 3D understanding tasks. This tokenizer addresses the challenge of applying CLIP, which is tra…
-
New LDMR framework tackles model forgetting in 3D object detection
Researchers have developed a new framework called Learning-Dynamics-driven Memory and Review (LDMR) to address the problem of model forgetting in incremental 3D object detection. This framework aims to improve how detec…
-
New framework RelGraphOV enhances 3D scene understanding with object relationships
Researchers have introduced RelGraphOV, a novel framework designed to improve open-vocabulary 3D scene understanding by incorporating object relationships. This approach utilizes 3D scene graphs to infer and refine sema…
-
New distillation method boosts 3D semantic segmentation performance
Researchers have developed a novel knowledge distillation technique called Heterogeneous and Adept Snapshot Distillation (HAS-KD) to improve 3D semantic segmentation performance. This method transfers knowledge from mul…
-
New Gaussian-Voxel Duet improves 3D surface reconstruction
Researchers have developed a novel hybrid representation called Gaussian-Voxel Duet to improve monocular surface reconstruction. This method combines 3D Gaussian Splatting with a sparse voxel scaffold, confining Gaussia…