Research map / Probabilistic and scientific modeling / Differential equations and PDEs
Deep Trajectory Supervision: Deep Supervision Strikes Back
Han Wang, Weijie Wang, Jiaqi Liu, Hilde Kuehne, Nicu Sebe
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 DTS depth-dependent soft-target auxiliary supervision | ViT-Small · CIFAR-100 (from scratch) | top-1 accuracy: 54.8 -> 57.0 | Improved |
| add DTS auxiliary supervision | ViT-Small · CIFAR-100 + RandAug | top-1 accuracy: 63.5 -> 64.1 | Improved |
| add DTS auxiliary supervision | ViT-Small · iNat19 + RandAug | top-1 accuracy: 46.4 -> 49.1 | Improved |
| add DTS auxiliary supervision | ViT-Small · Tiny-ImageNet | top-1 accuracy: 39.9 -> 41.9 | Improved |
| add DTS auxiliary supervision | ViT-Small · CIFAR-10 | top-1 accuracy: 82.2 -> 83.9 | Improved |
| add standard hard-label deep supervision | ViT-Small · CIFAR-100 + RandAug | top-1 accuracy: 63.5 -> 62.9 | Hurt |
| add standard hard-label deep supervision | ViT-Small · CIFAR-100 | top-1 accuracy: 54.8 -> 55.9 | Improved |
| add DTS to DeiT-III-T | DeiT-III-T · ImageNet-1K | top-1 accuracy: 72.7 -> 75.2 | Improved |
| add DTS to DeiT-III-B | DeiT-III-B · ImageNet-1K | top-1 accuracy: 82.3 -> 82.6 | Improved |
| use exponential target schedule instead of linear (ablation) | DTS (DeiT-III-S) · ImageNet-1K | top-1 accuracy: 80.9 -> 82.2 | Improved |
| use exponential target schedule instead of tanh (ablation) | DTS (DeiT-III-S) · ImageNet-1K | top-1 accuracy: 80.7 -> 82.2 | Improved |
| use exponential curvature beta=0.5 instead of linear trajectory (ablation) | DTS (DeiT-III-S) · ImageNet-1K (50% training) | top-1 accuracy: 73.60 -> 73.83 | No clear effect |
Closest papers by meaning
- ChainFlow-VLA: Causal Flow Planning with Vision-Language Models (NeurIPS 2026 · similarity 0.85)
- OneVL: One-Step Latent Reasoning and Planning with Vision-Language Explanation (NeurIPS 2026 · similarity 0.85)
- Grounding Driving VLA via Inverse Kinematics (NeurIPS 2026 · similarity 0.84)
- ProMotion: Prototypes As Motion Learners (CVPR 2024 · similarity 0.84)
- Normalizing Trajectory Models (NeurIPS 2026 · similarity 0.84)
- T4P: Test-Time Training of Trajectory Prediction via Masked Autoencoder and Actor-specific Token Memory (CVPR 2024 · similarity 0.84)
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.
