Research map / Vision-language and multimodal / Human and embodied modeling
YouTubePD: A Multimodal Benchmark for Parkinson’s Disease Analysis
Andy Zhou, Samuel Li, Pranav Sriram, Xiang Li, Jiahua Dong, Ansh Sharma, Yuanyi Zhong, Shirui Luo, Maria Jaromin, Volodymyr Kindratenko, Joerg Heintz, Christopher Zallek 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 spatial-temporal region attention with hierarchy loss instead of VGGFace linear probing | ResNet50 facial classifier · YouTubePD multiclass | AUROC: 0.68 -> 0.74 | Improved |
| same, binary classification | ResNet50 facial classifier · YouTubePD binary | AUROC: 0.86 -> 0.92 | Improved |
| same, multiclass top-1 accuracy | ResNet50 facial classifier · YouTubePD multiclass | Top-1 accuracy: 78.20 -> 77.03 | Hurt |
| add logit-fusion with landmark and audio modalities to facial video | facial expression model · YouTubePD multiclass | F1: 0.23 -> 0.28 | Improved |
| use Kinetics-pretrained action recognition backbones instead of VGGFace (ablation) | linear-probed video models · YouTubePD multiclass | accuracy: 78.20 -> 45.64 | Hurt |
| use AffectNet-8 pretrained backbone instead of VGGFace (ablation) | ResNet50 linear probing · YouTubePD binary | F1: 0.56 -> 0.45 | Hurt |
| use only temporal attention (no spatial) (ablation) | proposed model · YouTubePD binary | Top-1 accuracy: 88.00 -> 77.14 | Hurt |
| use only spatial attention (no temporal) (ablation) | proposed model · YouTubePD binary | Top-1 accuracy: 88.00 -> 82.86 | Hurt |
| fuse modalities by frame concatenation instead of frame voting (ablation) | multimodal model · YouTubePD multiclass | F1: 0.28 -> 0.21 | Hurt |
| replace MAE audio features with wav2vec 2.0 (ablation) | audio modality · YouTubePD multiclass | F1: 0.16 -> 0.14 | No clear effect |
| replace MAE audio features with MFCC (ablation) | audio modality · YouTubePD multiclass | F1: 0.16 -> 0.11 | Hurt |
| leave-one-subject-out split instead of default split | proposed vs VGGFace · YouTubePD multiclass | AUROC: 0.66 -> 0.71 | Improved |
Closest papers by meaning
- Accessible, At-Home Detection of Parkinson’s Disease via Multi-Task Video Analysis (AAAI 2025 · similarity 0.91)
- TULIP: Multi-camera 3D Precision Assessment of Parkinson's Disease (CVPR 2024 · similarity 0.89)
- Care-PD: A Multi-Site Anonymized Clinical Dataset for Parkinson’s Disease Gait Assessment (NeurIPS 2025 · similarity 0.88)
- Breaking Data Silos in Parkinson’s Disease Diagnosis: An Adaptive Federated Learning Approach for Privacy-Preserving Facial Expression Analysis (AAAI 2025 · similarity 0.86)
- CLIMB: Data Foundations for Large Scale Multimodal Clinical Foundation Models (ICML 2025 · similarity 0.83)
- PathView-Bench: Can Multimodal Large Language Models Achieve Fine-grained Multiscale Understanding of Pathology Images? (NeurIPS 2026 · similarity 0.82)
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.
