[WIP] Upload folder using huggingface_hub (multi-commit 4c76aea3e91bf44420a5c3fb05cc61fc0821d64f4abf66f812fa5d643fa0826d)

.gitattributes

@@ -80,3 +80,4 @@ Library/PackageCache/com.unity.xr.oculus@3.3.0/Runtime/x86/OVRPlugin.dll filter=
 Library/ScriptAssemblies/Unity.InputSystem.dll filter=lfs diff=lfs merge=lfs -text
 Library/ScriptAssemblies/Unity.VisualScripting.Core.Editor.dll filter=lfs diff=lfs merge=lfs -text
 Library/SourceAssetDB filter=lfs diff=lfs merge=lfs -text
+demo_virtual.mp4 filter=lfs diff=lfs merge=lfs -text

DisplayPos.m

@@ -0,0 +1,10 @@
+function out = DisplayPos(obj)
+    pos = getJointAndFingerPos(obj);
+    out = ['End Effector Position - X: ' sprintf('%.2f', obj.EndEffectorPose(1)) ...
+      ', Y: ' sprintf('%.2f', obj.EndEffectorPose(2)) ...
+      ', Z: ' sprintf('%.2f', obj.EndEffectorPose(3)) ...
+      '. Twist Angle (in radians): ' sprintf('%.2f', pos(7)) ...
+      ', Finger 1 Strength: ' sprintf('%.2f', pos(8)) ...
+      ', Finger 2 Strength: ' sprintf('%.2f', pos(9)) ...
+      ', Finger 3 Strength: ' sprintf('%.2f', pos(10)) '.'];
+end

EEGClassification.m

@@ -0,0 +1,16 @@
+function step = EEGClassification(net, data, step)
+    
+    classNames = {'Left', 'Right', 'Grasp', 'Twist', 'Idle'};
+
+    gt = {'Right', 'Grasp', 'Left', 'Twist', 'Left'};
+    output = forward(net, dlarray(data, 'SSCB')); 
+    [~, predictedIndex] = max(output, [], 'all');
+    if strcmp(classNames{predictedIndex}, gt{step})
+        step = step + 1;
+        disp(['   디코딩 결과: ' classNames{predictedIndex} '. 맞습니다. 다음 단계로 갑니다.']);
+    else
+        step = 1;
+        disp(['   디코딩 결과: ' classNames{predictedIndex} '. 틀렸습니다. 단계 1로 갑니다.']);
+    end
+    pause(5);
+end

EEGInitialization.m

@@ -0,0 +1,24 @@
+function [inlet, fs, chan] = EEGInitialization()
+    % Load the LSL library
+    lib = lsl_loadlib();
+
+    % Find a stream of type 'EEG'
+    result = {};
+    maxAttempts = 1;  % Set a maximum number of attempts to find the stream
+    attempts = 0;
+
+    while isempty(result) && attempts < maxAttempts
+        result = lsl_resolve_byprop(lib, 'type', 'EEG');
+        pause(1);  % Optional: wait for a second before trying again
+        attempts = attempts + 1;
+    end
+
+    if isempty(result)
+        error('Failed to find an EEG stream');
+    end
+
+    % Connect to the first found stream
+    inlet = lsl_inlet(result{1});
+    fs = result{1}.nominal_srate();
+    chan = result{1}.channel_count();
+end

EEGNetModel.m

@@ -0,0 +1,60 @@
+function layers = EEGNetModel(in_chans, n_classes, varargin)
+    % EEGNetv4 creation function for MATLAB
+
+    % Default parameters
+    p = inputParser;
+    addRequired(p, 'in_chans');
+    addRequired(p, 'n_classes');
+    addRequired(p, 'input_window_samples');
+    addParameter(p, 'pool_mode', 'mean');
+    addParameter(p, 'F1', 8);
+    addParameter(p, 'D', 2);
+    addParameter(p, 'F2', 16);
+    addParameter(p, 'kernel_length', 64);
+    addParameter(p, 'third_kernel_size', [8, 4]);
+    addParameter(p, 'drop_prob', 0.25);
+    parse(p, in_chans, n_classes, varargin{:});
+
+    % Extract parameters from parsed input
+    params = p.Results;
+
+    % EEGNetv4 Layers
+
+    % First set of layers
+    layers = [
+        imageInputLayer([params.in_chans, params.input_window_samples, 1], 'Normalization', 'none')
+        convolution2dLayer([1, params.kernel_length], params.F1, 'Stride', [1, 1], 'Padding',[0, floor(params.kernel_length / 2)])
+        batchNormalizationLayer()
+        convolution2dLayer([params.in_chans, 1], params.F1*params.D, 'Stride', [1, 1], 'Padding', [0, 0])
+        batchNormalizationLayer()
+        reluLayer()
+        averagePooling2dLayer([1, 4], 'Stride', [1, 4])
+        dropoutLayer(params.drop_prob)
+    ];
+
+    % Second set of layers (Depthwise Separable Convolution)
+    layers = [
+        layers
+        convolution2dLayer([1, 16], params.F1*params.D, 'Stride', [1, 1], 'Padding', [0, 8])
+        convolution2dLayer([1, 1], params.F2, 'Stride', [1, 1], 'Padding', [0, 0])
+        batchNormalizationLayer()
+        reluLayer()
+        averagePooling2dLayer([1, 8], 'Stride', [1, 8])
+        dropoutLayer(params.drop_prob)
+    ];
+
+    % Third set of layers
+    layers = [
+        layers
+        convolution2dLayer([1, 23], params.n_classes)
+        softmaxLayer()
+        classificationLayer()
+    ];
+
+    % Convert layers to layerGraph
+%     lgraph = layerGraph(layers);
+    
+    % Convert layerGraph to dlnetwork
+%     net = dlnetwork(lgraph);
+end
+

EEGNetTraining.m

@@ -0,0 +1,16 @@
+function trainedNet = EEGNetTraining(data, labels, layers)
+   
+    options = trainingOptions('sgdm', ...
+    'InitialLearnRate',0.001, ...
+    'MaxEpochs',100, ...
+    'MiniBatchSize',1, ...
+    'Shuffle','every-epoch', ...
+    'Verbose',false, ...
+    'Plots','none');
+    
+%     data = dlarray(data, 'SSCB');
+    data = permute(data, [2 3 4 1]);
+    labels = categorical(labels)';
+    trainedNet = trainNetwork(data, labels, layers, options);
+
+end

demo_virtual.mp4

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:646b8844fbd1a002a02571abab9312d06d5392194dd45b0f64abdeb2a68a3b84
+size 151099245