Research map / Video / Video segmentation and tracking
Efficient Event Camera Data Pretraining with Adaptive Prompt Fusion
Quanmin Liang, Qiang Li, Shuai Liu, Xinzi Cao, Jinyi Lu, Feidiao Yang, Wei Zhang, Kai Huang, Yonghong Tian
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 overlapping patch embedding (overlap window) (ablation) | STP · N-ImageNet | finetuned top-1 accuracy: 67.04 -> 68.87 | Improved |
| add fine-grained conv after patch embedding (ablation) | STP · N-ImageNet | finetuned top-1 accuracy: 67.36 -> 68.87 | Improved |
| add both overlap window and conv (ablation) | STP · N-ImageNet | pretraining top-1 accuracy: 59.30 -> 66.01 | Improved |
| add local temporal aggregation (LTA) (ablation) | STP · N-ImageNet | finetuned top-1 accuracy: 68.12 -> 68.87 | Improved |
| use STP instead of E2VID reconstruction prompting (ablation) | image-pretrained ViT-S · N-ImageNet | finetuned top-1 accuracy: 64.93 -> 68.87 | Improved |
| use STP instead of vanilla patch-embedding prompt (ablation) | image-pretrained ViT-S · N-ImageNet | finetuned top-1 accuracy: 66.89 -> 68.87 | Improved |
| use TECM instead of ECM input (ablation) | STP · N-ImageNet | top-1 accuracy: 67.82 -> 68.87 | Improved |
| use TECM instead of Voxel Grid input (ablation) | STP · N-ImageNet | top-1 accuracy: 58.26 -> 68.87 | Improved |
| use TECM instead of EST input (ablation) | STP · N-ImageNet | top-1 accuracy: 68.07 -> 68.87 | Improved |
| scale frozen backbone from ViT-S/16 to ViT-L/16 (ablation) | STP · N-ImageNet | finetuned top-1 accuracy: 68.87 -> 78.06 | Improved |
| swap ViT-S/16 backbone for ConvNeXt-T (ablation) | STP · N-ImageNet | finetuned top-1 accuracy: 68.87 -> 71.90 | Improved |
| swap ViT-S/16 backbone for ResNet50 (ablation) | STP · N-ImageNet | finetuned top-1 accuracy: 68.87 -> 65.93 | Hurt |
Closest papers by meaning
- Event Camera Data Dense Pre-training (ECCV 2024 · similarity 0.89)
- EventZoom: A Progressive Approach to Event-Based Data Augmentation for Enhanced Neuromorphic Vision (AAAI 2025 · similarity 0.88)
- PASS: Path-selective State Space Model for Event-based Recognition (NeurIPS 2025 · similarity 0.87)
- Efficient Event-Based Semantic Segmentation via Exploiting Frame-Event Fusion: A Hybrid Neural Network Approach (AAAI 2025 · similarity 0.87)
- Bridge Frame and Event: Common Spatiotemporal Fusion for High-Dynamic Scene Optical Flow (CVPR 2025 · similarity 0.87)
- EventGPT: Event Stream Understanding with Multimodal Large Language Models (CVPR 2025 · 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.
