Research map / Vision-language and multimodal / Human and embodied modeling
Scaling Dense Event-Stream Pretraining from Visual Foundation Models
Zhiwen Chen, Junhui Hou, Zhiyu Zhu, Jinjian Wu, Guangming Shi
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 L1 cross-modal distillation from DINOv3 on image-event pairs (ablation) | ViT-L event encoder · DSEC-Semantic | mIoU: 64.31 -> 66.17 | Improved |
| add event activation mask to distillation (ablation) | ViT-L event encoder · DSEC-Semantic | mIoU: 66.17 -> 66.54 | Improved |
| add intra-modal structure loss (ablation) | ViT-L event encoder · DSEC-Semantic | mIoU: 66.54 -> 69.20 | Improved |
| add cross-modal structure loss alone (ablation) | ViT-L event encoder · DSEC-Semantic | mIoU: 66.54 -> 68.68 | Improved |
| add both structure losses vs intra-modal alone (ablation) | ViT-L event encoder · DSEC-Semantic | mIoU: 69.20 -> 69.65 | Improved |
| add both structure losses to masked distillation (ablation) | ViT-L event encoder · DSEC-Depth | RMSE: 4.025 -> 3.694 | Improved |
| add event pretraining vs image-domain initialization (ablation) | ViT-L event encoder · DDD17-Seg | mIoU: 59.60 -> 65.08 | Improved |
| use ScaleEvent ViT-S instead of DepthAnyEvent-R (same backbone) | Event depth model · DSEC-Depth | RMSE: 8.880 -> 4.564 | Improved |
| scale backbone from ViT-S to ViT-L | ScaleEvent · DSEC-Semantic | mIoU: 61.12 -> 69.65 | Improved |
| ScaleEvent ViT-S vs STP/ECDDP Swin-T on DDD17 semantic segmentation | ScaleEvent · DDD17-Seg | mIoU: 63.29 vs 59.64 | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- OpenESS — OpenESS: Event-based Semantic Scene Understanding with Open Vocabularies (CVPR 2024) · 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
- Segment Any Event Streams via Weighted Adaptation of Pivotal Tokens (CVPR 2024 · similarity 0.88)
- EventGPT: Event Stream Understanding with Multimodal Large Language Models (CVPR 2025 · similarity 0.87)
- EventBind: Learning a Unified Representation to Bind Them All for Event-based Open-world Understanding (ECCV 2024 · similarity 0.87)
- Event Camera Data Dense Pre-training (ECCV 2024 · similarity 0.87)
- EventLens: Event-Structure Reinforcement Learning for Video Understanding (NeurIPS 2026 · similarity 0.87)
- OpenESS: Event-based Semantic Scene Understanding with Open Vocabularies (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.
