Research map / Architectures and efficiency / Attention and transformers
InvPT++: Inverted Pyramid Multi-Task Transformer for Visual Scene Understanding
Hanrong Ye, Dan Xu
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 UP-Transformer block (ablation) | InvPT++ Baseline (MT) · NYUD-v2 | Semseg mIoU: 41.06 -> 43.18 | Improved |
| add UP-Transformer block (ablation) | InvPT++ Baseline (MT) · NYUD-v2 | Depth RMSE: 0.6350 -> 0.5643 | Improved |
| add UP-Transformer block (ablation) | InvPT++ Baseline (MT) · NYUD-v2 | Normal mErr: 21.47 -> 21.05 | Improved |
| add Fusion Attention to UTB (ablation) | InvPT++ · PASCAL-Context (Swin-T) | Semseg mIoU: 72.34 -> 73.29 | Improved |
| add Encoder Feature Aggregation (ablation) | InvPT++ w/ UTB+FA · PASCAL-Context (Swin-T) | Semseg mIoU: 73.29 -> 73.93 | Improved |
| add Encoder Feature Aggregation (ablation) | InvPT++ w/ UTB+FA · NYUD-v2 (Swin-T) | Semseg mIoU: 43.64 -> 44.27 | Improved |
| add Encoder Feature Aggregation (ablation) | InvPT++ w/ UTB+FA · NYUD-v2 (Swin-T) | Boundary odsF: 76.10 -> 76.10 | No clear effect |
| multi-task learning versus single-task models | InvPT++ · NYUD-v2 | multi-task gain: 2.59 | Improved |
| multi-task learning versus single-task models | InvPT++ · PASCAL-Context | multi-task gain: 1.76 | Improved |
| use Selective Attention instead of Fusion Attention (ablation) | InvPT++ (ViT-L) · PASCAL-Context | Semseg mIoU: 79.03 -> 80.22 | Improved |
| use Selective Attention instead of Fusion Attention (ablation) | InvPT++ (ViT-L) · PASCAL-Context | decoder FLOPs: -22.51% | Improved |
| use Selective Attention instead of Fusion Attention (ablation) | InvPT++ (ViT-L) · NYUD-v2 | Depth RMSE: 0.5183 -> 0.5096 | Improved |
Later papers that compared with this method
1 later accepted paper reports InvPT as a baseline in their results tables; 1 of them report a better number on most of the results they share with it.
- Resolving Token-Space Gradient Conflicts: Token Space Manipulation for Transformer-Based Multi-Task Learning (ICCV 2025 · better on 2 of 2 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
- Going Beyond Multi-Task Dense Prediction with Synergy Embedding Models (CVPR 2024 · similarity 0.87)
- LSceneLLM: Enhancing Large 3D Scene Understanding Using Adaptive Visual Preferences (CVPR 2025 · similarity 0.87)
- VMT-Adapter: Parameter-Efficient Transfer Learning for Multi-Task Dense Scene Understanding (AAAI 2024 · similarity 0.86)
- Crab: A Unified Audio-Visual Scene Understanding Model with Explicit Cooperation (CVPR 2025 · similarity 0.86)
- CARE Transformer: Mobile-Friendly Linear Visual Transformer via Decoupled Dual Interaction (CVPR 2025 · similarity 0.86)
- Understanding Multi-View Transformers (NeurIPS 2026 · 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.
