Research map / Video / Video-language understanding
OphNet: A Large-Scale Video Benchmark for Ophthalmic Surgical Workflow Understanding
Ming Hu, Peng Xia, Lin Wang, Siyuan Yan, Feilong Tang, zhongxing xu, Yimin Luo, Kaimin Song, Jurgen Leitner, Xuelian Cheng, Jun Cheng, Chi Liu and 2 more
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 |
|---|---|---|---|
| use CLIP-based video models (X-CLIP, ViFi-CLIP) instead of I3D/SlowFast/X3D/MViT V2 | OphNet phase classification · OphNet phase classification (All) | Top-1 accuracy: 33.5 -> 68.4 | Improved |
| initialize from Kinetics-400 pretrained weights (ablation) | I3D/SlowFast/X3D/MViT V2 on phase classification · OphNet phase classification (All) | Top-1 accuracy: 25.7 -> 30.2 (I3D) | Improved |
| Kinetics-400 pretraining for MViT V2 on phase classification (ablation) | MViT V2 · OphNet phase classification (All) | Top-1 accuracy: 27.0 -> 28.3 (All); 26.2 -> 34.2 (Cataract); 21.0 -> 23.3 (Glaucoma) | Mixed |
| Kinetics-400 pretraining on primary surgery classification (ablation) | I3D/SlowFast/X3D/MViT V2 · OphNet primary surgery classification (All) | Top-1 accuracy: 29.8 -> 27.2 (I3D); 28.5 -> 35.4 (X3D) | Mixed |
| increase CLIP model input frames from 16 to 32 (ablation) | X-CLIP / ViFi-CLIP · OphNet phase and operation classification | Top-1 accuracy: 63.4 -> 62.7 (X-CLIP phase All); 66.1 -> 68.4 (ViFi-CLIP phase All) | Mixed |
| use TriDet with SwinViviT backbone instead of ActionFormer (ablation) | phase localization · OphNet phase localization | Avg. mAP: 46.4 -> 50.4 | Improved |
| use SlowFast instead of SwinViviT backbone for TriDet (ablation) | TriDet · OphNet phase localization | Avg. mAP: 50.4 -> 48.6 | Hurt |
| Kinetics-400 pretraining for phase anticipation (ablation) | I3D/SlowFast/MViT V2 · OphNet phase anticipation | Avg. Top-1 accuracy: 47.5 -> 48.2 (MViT V2) | Improved |
| increase observation ratio in phase anticipation (ablation) | anticipation models · OphNet phase anticipation | Top-1 accuracy: 27.8 -> 51.7 (0.1 -> 0.7, MViT V2*, approx per table) | Improved |
Closest papers by meaning
- CholecTrack20: A Multi-Perspective Tracking Dataset for Surgical Tools (CVPR 2025 · similarity 0.89)
- SurgCoT: Advancing Spatiotemporal Reasoning in Surgical Videos through a Chain-of-Thought Benchmark (CVPR 2026 · similarity 0.87)
- Bridging Vision and Language for Robust Context-Aware Surgical Point Tracking: The VL-SurgPT Dataset and Benchmark (AAAI 2026 · similarity 0.87)
- MM-OR: A Large Multimodal Operating Room Dataset for Semantic Understanding of High-Intensity Surgical Environments (CVPR 2025 · similarity 0.87)
- Towards Dynamic 3D Reconstruction of Hand-Instrument Interaction in Ophthalmic Surgery (NeurIPS 2025 · similarity 0.86)
- EgoExOR: An Ego-Exo-Centric Operating Room Dataset for Surgical Activity Understanding (NeurIPS 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.
