Research map / Video / Video segmentation and tracking
SciVid: Cross-Domain Evaluation of Video Models in Scientific Applications
Yana Hasson, Pauline Luc, Liliane Momeni, Maks Ovsjanikov, Guillaume Le Moing, Alina Kuznetsova, Ira Ktena, Jennifer J. Sun, Skanda Koppula, Dilara Gokay, Joseph Heyward, Etienne Pot and 1 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 cross-attention readout instead of linear readout (ablation) | frozen 4DS-e · FlyVsFly | mAP: 0.568 -> 0.894 | Improved |
| use cross-attention readout instead of linear readout (ablation) | frozen 4DS-e · CalMS21 | mAP: 0.525 -> 0.817 | Improved |
| use cross-attention readout instead of linear readout (ablation) | frozen 4DS-e · Digital Typhoon | RMSE: 7.45 -> 4.23 | Improved |
| finetune the 4DS-e backbone instead of frozen (ablation) | 4DS-e readout · STIR | accuracy: 51.3 -> 59.3 | Improved |
| use LoRA rank 32 instead of frozen backbone (ablation) | 4DS-e readout · STIR | accuracy: 51.3 -> 58.4 | Improved |
| increase LoRA rank from 4 to 32 (ablation) | 4DS-e readout · STIR | accuracy: 57.4 -> 58.4 | Improved |
| finetune 4DS-e backbone instead of frozen | 4DS-e readout · STIR test | accuracy: 57.8 -> 69.2 | Improved |
| train readout longer (40k to 400k steps) (ablation) | VideoPrism-g · FlyVsFly | mAP: 87.2 -> 92.5 | Improved |
| train readout longer (40k to 400k steps) (ablation) | VideoPrism-g · CalMS21 | mAP: 92.0 -> 90.9 | Hurt |
| use DPT readout instead of attention readout (ablation) | frozen VideoMAEv2-g · WeatherBench 2 | Z500 wRMSE: 892 -> 594 | Improved |
| add conditioning input (ablation) | finetuned 4DS-e · WeatherBench 2 | Z500 wRMSE: 531 -> 549 | Hurt |
| add conditioning input (ablation) | finetuned 4DS-e · WeatherBench 2 | Q700 wRMSE: 1.55e-03 -> 1.54e-03 | Mixed |
Closest papers by meaning
- SciVideoBench: Benchmarking Scientific Video Reasoning in Large Multimodal Models (ICML 2026 · similarity 0.88)
- Specialized Foundation Models Struggle to Beat Supervised Baselines (ICLR 2025 · similarity 0.87)
- Apollo: An Exploration of Video Understanding in Large Multimodal Models (CVPR 2025 · similarity 0.86)
- Empowering World Models with Reflection for Embodied Video Prediction (ICML 2025 · similarity 0.86)
- Animal-Bench: Benchmarking Multimodal Video Models for Animal-centric Video Understanding (NeurIPS 2024 · similarity 0.86)
- TOMATO: Assessing Visual Temporal Reasoning Capabilities in Multimodal Foundation Models (ICLR 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.
