Cut-ViT: Task-Specific Model Pruning via Gram Anchoring Subspace Consistency

• URL: http://arxiv.org/abs/2608.28205v1
• Date: Fri, 28 Aug 2026 11:25:36 GMT
• Title: Cut-ViT: Task-Specific Model Pruning via Gram Anchoring Subspace Consistency
• Abstract summary: We propose a task-specific pruning pipeline, named Cut-ViT, for pruning visual foundation models.<n> Basis-agnostic and residual constraints are adopted to align the gram subspaces between the native and pruned DINOv3 models.<n>Experiments show that Cut-ViT requires approximately one minute on a single A100 GPU to obtainworks at various sparsity levels.
• Score: 71.81785123423516
• License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
• Abstract: Pruning visual foundation models has attracted considerable attention. However, existing methods focus on rigid point-to-point token alignment on a single dataset for pruning, suffering from two limitations: i) robustness degradation, and ii) task-specificity deficiency. To address these limitations, we propose a task-specific pruning pipeline, named Cut-ViT. Specifically, we first construct gram anchoring matrices from both spatial and semantic perspectives, and perform the subspace decomposition to extract the corresponding subspace bases. Basis-agnostic and residual constraints are then adopted to align the gram subspaces between the native and pruned DINOv3 models along spatial and channel dimensions, enabling subnetworks to inherit robust feature representations of native DINOv3. Furthermore, we design spectral entropy adaptation, which quantifies the information density of feature manifolds along spatial and channel dimensions, thereby adapting the pruning objective to specific downstream tasks. Experiments show that Cut-ViT requires approximately one minute on a single A100 GPU to obtain subnetworks at various sparsity levels, using only 20.9% of the time and 45.5% of the GPU memory compared with previous methods, while achieving SOTA performance on six tasks across nine datasets.

Related papers

• Scaling Parallel Sequence Models to Foundation-Scale Vision Encoders [123.58723804218151]
  Vision foundation models are bottlenecked by the quadratic cost of self-attention.<n>We present C-GSPN based on 2D spatial propagation propagation.
  arXiv  Detail & Related papers  (2026-05-30T14:29:43Z)
• SEED: Targeted Data Selection by Weighted Independent Set [76.68391670109433]
  We develop a robust and scalable data selection pipeline dubbed SEED.<n>SEED consistently outperforms state-of-the-art methods on instruction tuning, visual instruction tuning, and semantic segmentation.
  arXiv  Detail & Related papers  (2026-05-15T07:26:54Z)
• Holi-Spatial: Evolving Video Streams into Holistic 3D Spatial Intelligence [78.1406635199656]
  Holi-Spatial is the first fully automated, large-scale, spatially-aware multimodal dataset constructed from raw video inputs without human intervention.<n>Holi-Spatial-4M is the first large-scale, high-quality 3D semantic dataset, containing 12K optimized 3DGS scenes, 1.3M 2D masks, 320K 3D bounding boxes, 320K instance captions, 1.2M 3D grounding instances, and 1.2M spatial QA pairs.
  arXiv  Detail & Related papers  (2026-03-08T14:49:20Z)
• Through the Perspective of LiDAR: A Feature-Enriched and Uncertainty-Aware Annotation Pipeline for Terrestrial Point Cloud Segmentation [5.173182375745059]
  Accurate semantic segmentation of terrestrial laser scanning (TLS) point clouds is limited by costly manual annotation.<n>We propose a semi-automated, uncertainty-aware pipeline that integrates spherical projection, feature enrichment, ensemble learning, and targeted annotation.<n>Our approach projects 3D points to a 2D spherical grid, enriches pixels with multi-source features, and trains an ensemble of segmentation networks to produce pseudo-labels and uncertainty maps.
  arXiv  Detail & Related papers  (2025-10-08T02:25:59Z)
• X-PDNet: Accurate Joint Plane Instance Segmentation and Monocular Depth Estimation with Cross-Task Distillation and Boundary Correction [9.215384107659665]
  X-PDNet is a framework for the multitask learning of plane instance segmentation and depth estimation. We highlight the current limitations of using the ground truth boundary to develop boundary regression loss. We propose a novel method that exploits depth information to support precise boundary region segmentation.
  arXiv  Detail & Related papers  (2023-09-15T14:27:54Z)
• PointOcc: Cylindrical Tri-Perspective View for Point-based 3D Semantic Occupancy Prediction [72.75478398447396]
  We propose a cylindrical tri-perspective view to represent point clouds effectively and comprehensively. Considering the distance distribution of LiDAR point clouds, we construct the tri-perspective view in the cylindrical coordinate system. We employ spatial group pooling to maintain structural details during projection and adopt 2D backbones to efficiently process each TPV plane.
  arXiv  Detail & Related papers  (2023-08-31T17:57:17Z)
• TC-Net: Triple Context Network for Automated Stroke Lesion Segmentation [0.5482532589225552]
  We propose a new network, Triple Context Network (TC-Net), with the capture of spatial contextual information as the core. Our network is evaluated on the open dataset ATLAS, achieving the highest score of 0.594, Hausdorff distance of 27.005 mm, and average symmetry surface distance of 7.137 mm.
  arXiv  Detail & Related papers  (2022-02-28T11:12:16Z)
• Adaptive Context-Aware Multi-Modal Network for Depth Completion [107.15344488719322]
  We propose to adopt the graph propagation to capture the observed spatial contexts. We then apply the attention mechanism on the propagation, which encourages the network to model the contextual information adaptively. Finally, we introduce the symmetric gated fusion strategy to exploit the extracted multi-modal features effectively. Our model, named Adaptive Context-Aware Multi-Modal Network (ACMNet), achieves the state-of-the-art performance on two benchmarks.
  arXiv  Detail & Related papers  (2020-08-25T06:00:06Z)
• Scan-based Semantic Segmentation of LiDAR Point Clouds: An Experimental Study [2.6205925938720833]
  State of the art methods use deep neural networks to predict semantic classes for each point in a LiDAR scan. A powerful and efficient way to process LiDAR measurements is to use two-dimensional, image-like projections. We demonstrate various techniques to boost the performance and to improve runtime as well as memory constraints.
  arXiv  Detail & Related papers  (2020-04-06T11:08:12Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.