Research map / 3D vision / 3D generation
SegVGGT: Joint 3D Reconstruction and Instance Segmentation from Multi-View Images
Jinyuan Qu, Hongyang Li, Lei Zhang
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add FADA attention-alignment loss (ablation) | SegVGGT · ScanNet200 val | mAP: 26.7 -> 31.1 | Improved |
| add FADA JS term to Hungarian matching cost (ablation) | SegVGGT · ScanNet200 val | mAP: 31.1 -> 31.9 | Improved |
| insert query interaction in all 24 layers instead of early 12 (ablation) | SegVGGT · ScanNet200 val | mAP: 23.8 -> 31.9 | Improved |
| insert query interaction in all 24 layers instead of late 12 (ablation) | SegVGGT · ScanNet200 val | mAP: 30.5 -> 31.9 | Improved |
| insert query interaction in all 24 layers instead of interleaved 12 (ablation) | SegVGGT · ScanNet200 val | mAP: 30.4 -> 31.9 | Improved |
| jointly fine-tune frame/global attention with LoRA instead of freezing (ablation) | SegVGGT · ScanNet200 val | mAP: 23.4 -> 31.9 | Improved |
| use unified architecture instead of separate 12-layer Mask2Former decoder on VGGT (ablation) | SegVGGT · ScanNet200 val | mAP: 23.4 -> 31.9 | Improved |
| increase separate mask decoder depth from 12 to 18 layers (ablation) | VGGT(FT)+Mask Decoder · ScanNet200 val | mAP: 23.4 -> 16.7 | Hurt |
| two-stage VGGT reconstruction then OneFormer3D instead of point cloud input | OneFormer3D · ScanNet200 val | mAP: 30.2 -> 2.5 | Hurt |
| joint fine-tuning with instance segmentation vs geometry-only fine-tuning | SegVGGT · ScanNet depth (10 scenes) | Abs Rel: 0.0311 -> 0.0307 | No clear effect |
| disable PanSt3R QUBO post-processing | PanSt3R · 100 input frames | total time (s): 381.96 -> 59.63 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- IGGT — IGGT: Instance-Grounded Geometry Transformer for Semantic 3D Reconstruction (ICLR 2026) · better on 4 of 4 shared results
- PanSt3R — PanSt3R: Multi-view Consistent Panoptic Segmentation (ICCV 2025) · better on 4 of 4 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- VGGT: Visual Geometry Grounded Transformer (CVPR 2025 · similarity 0.89)
- TurboVGGT: Fast Visual Geometry Reconstruction with Adaptive Alternating Attention (NeurIPS 2026 · similarity 0.88)
- IGGT4D: Streaming 4D Instance-Grounded Geometry Transformer (NeurIPS 2026 · similarity 0.87)
- LiteVGGT: Boosting Vanilla VGGT via Geometry-aware Cached Token Merging (CVPR 2026 · similarity 0.87)
- SA3DIP: Segment Any 3D Instance with Potential 3D Priors (NeurIPS 2024 · similarity 0.87)
- ODIN: A Single Model for 2D and 3D Segmentation (CVPR 2024 · similarity 0.86)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
