GaitAnalysisVLM

Enhancing Gait Video Analysis in Neurodegenerative Diseases by Knowledge Augmentation in Vision Language Model

Learned Numerical Text Embedding Space

We visualize the numerical text embeddings projected by MLPs learned through cross-modal training. We leverage Uniform Manifold Approximation and Projection (UMAP) to reduce the embedding dimension from 64 to 3.
Clink the links for interactive graphs.

Per-class Clinical Gait Notions

We employ specific clinical gait notions to develop per-class learnable prompts for prompt tuning. These notions have been generated using ChatGPT-4, then subsequently filtered, modified, and validated by a neurologist.

Per-class Automatic Prompts

We extract keywords from clinical gait notions to make per-class automatic prompts.