AdaptiveGaitSegNet: An Innovative Model with Advanced Feature Extraction for Enhanced Parkinson's Disease Gait Recognition

📊 What's included

🚩Full pipeline code for:

• Dataset preprocessing (the process from the RGB videos input to the output of the normalized contour maps).
• Using Focal Convolution to realize progressive feature learning from coarse to fine.
• Edge-Aware Pooling + Frame-Dimension Max Pooling.
• Feature fusion and extraction.
• Joint loss function (Triplet Loss Function + Cross-Entropy Loss Function).
• Use early stopping during training. #Default: patience=100, delta=0.001

🪐Training details:

• cache: If cache is set as TRUE all the training data will be loaded at once before the training start.
• iter: Number of training rounds for the model

🌋Validation metrics:

• iter: Iteration of the checkpoint to load. #Default: 80000
• batch_size: Batch size of the parallel test. #Default: 8
• cache: If cache is set as TRUE all the test data will be loaded at once before the transforming start. This might accelerate the testing. #Default: FALSE

🥀Supplementary Information (SI):

• Accuracy: The proportion of correct predictions in all the forecasts.
• Precision: The proportion that are actually Parkinson's disease among all the samples predicted to be Parkinson's disease.
• Recall/Sensitivity: The proportion that was correctly identified by the model among all the actual Parkinson's disease patients.
• F1-Score: The harmonic mean of precision and recall rate
• Confusion Matrix: Normal and Parkinson.

---

🍀 Version highlight

• This v0.5 updates the new data processing method. The RGB video is transformed and segmented to obtain the contour map that can be preprocessed.