Adding distortions

GenerarSinteticos_3coord_denak.py

@@ -0,0 +1,88 @@
+import numpy as np
+import pandas as pd
+from scipy.stats import multivariate_normal
+import matplotlib.pyplot as plt
+from mpl_toolkits.mplot3d import Axes3D
+import aaaaa
+
+# checked; it is ok
+# To generate distorsed covariance matrixes
+def genS(sds, R, distor=1):
+    # sds: standard deviations
+    # R: correlation matrix
+    # distor: level of distorsion
+    sdsdis = sds * distor
+    S = R * np.outer(sdsdis, sdsdis)
+    np.fill_diagonal(S, sdsdis ** 2)
+    return S
+
+original_data_path = "/home/scpmaotj/Github/hcr/reference_data/"
+synthetic_data_path = "/home/scpmaotj/Github/hcr/synthetic_data/"
+plots_path = "/home/scpmaotj/Github/hcr/plots/"
+nkey = 21
+
+#Nivel distorsion
+# 1 - Distorsionamos las desviaciones sin distorsionar la estructura de correlacion
+# Funcion genS()
+d = [1, 1.01, 1.05, 1.1, 1.2, 1.5, 1.7, 2]
+# 2 - Distorsionamos las desviaciones y la estrucctura de correlacion
+# Funcion???
+
+# Intento sacar una mano al azar de una normal multivariante.
+# Sacaremos m=100 observaciones
+m = 100
+#gesto <- 1
+
+#for gesto in range(1, 43):
+for gesto in range(1, 2):
+    print(gesto)
+    datXYZ = aaaaa.get_datXYZ(gesto, original_data_path)
+    # Hacemos un grafico con ndib= 16 lecturas, no las 16 primeras sino desplazadas pos=4 posiciones
+    ndib = 16
+    pos = 4
+    hands = datXYZ[pos:pos+ndib].reshape(ndib, nkey, 3)
+    fig, axes = plt.subplots(4, 4, subplot_kw={'projection': '3d'})
+    for i, ax in enumerate(axes.flat):
+        ax.scatter(hands[i, :, 0], hands[i, :, 1], hands[i, :, 2], c='b', marker='o')
+        ax.set_title(f"Original {i+pos}")
+    plt.savefig(plots_path + f"{gesto}orig.pdf")
+    plt.close()
+
+    obsmean = np.mean(datXYZ, axis=0)
+    obssd = np.std(datXYZ, axis=0)
+    R = np.corrcoef(datXYZ, rowvar=False)
+
+    id = len(d)
+    np.random.seed(3663)
+
+    for i in range(id):
+        Sberria = genS(obssd, R, distor=d[i])
+        z = multivariate_normal.rvs(mean=obsmean, cov=Sberria, size=m)
+        np.savetxt(synthetic_data_path + f"gesto{gesto}d{i-1}.csv", z, delimiter=',', fmt='%.6f')
+        simhands = z[pos:pos+ndib].reshape(ndib, nkey, 3).transpose(0, 2, 1)
+        fig, axes = plt.subplots(4, 4, subplot_kw={'projection': '3d'})
+        for j, ax in enumerate(axes.flat):
+            ax.scatter(simhands[j, 0, :], simhands[j, 1, :], simhands[j, 2, :], c='b', marker='o')
+            ax.set_title(f"Simulated, d={i-1}, It. {j+4}")
+        plt.savefig(plots_path + f"gesto{gesto}_3coord_d{i-1}.pdf")
+        plt.close()
+
+# Selecciono un gesto, y miro algunas configuraciones simuladas segun grado de distorsion
+gesto = 4
+gd = 0
+mirar = synthetic_data_path + f"gesto{gesto}d{gd}.csv"
+z = np.loadtxt(mirar, delimiter=',')
+m = z.shape[0]
+simhands = np.zeros((nkey, 3, m))
+for i in range(m):
+    simhands[:, :, i] = z[i].reshape(nkey, 3)
+simhands = simhands.transpose(2, 0, 1)
+simhands = np.swapaxes(simhands, 1, 2)
+fig, axes = plt.subplots(4, 4, subplot_kw={'projection': '3d'})
+for i, ax in enumerate(axes.flat):
+    ax.scatter(simhands[i, 0, :], simhands[i, 1, :], simhands[i, 2, :], c='b', marker='o')
+    ax.set_title(str(i+1))
+plt.show()
+
+
+

aaaaa.py

@@ -2,6 +2,16 @@ import csv
 import numpy as np
 import plotly.graph_objects as go
 
+# To generate distorsed covariance matrixes
+def genS(sds, R, distor=1):
+    # sds: standard deviations
+    # R: correlation matrix
+    # distor: level of distorsion
+    sdsdis = sds * distor
+    S = R * np.outer(sdsdis, sdsdis)
+    np.fill_diagonal(S, sdsdis ** 2)
+    return S
+
 def get_datXYZ(gesto, data_path):
     file_path = data_path + f"config{gesto}.csv"
     with open(file_path, 'r') as file:
@@ -15,7 +25,7 @@ def get_datXYZ(gesto, data_path):
             coordinates[i, j] = [x, y, z]
     return coordinates
 
-def plot_3d_points(points_21_3d):    
+def plot_3d_points(points_21_3d, title):    
     # Create a scatter plot
     colors = ['red', 'green', 'yellow', 'orange', 'blue'] 
     colors = ["purple"]
@@ -55,6 +65,7 @@ def plot_3d_points(points_21_3d):
             z=[start_point[2], end_point[2]],
             mode='lines',
             line=dict(color='black', width=2),
+            name ="",
             showlegend=False
         )
         
@@ -68,7 +79,7 @@ def plot_3d_points(points_21_3d):
             zaxis=dict(title='Z')
         ),
         dragmode = 'orbit',
-        title='3D Random Points'
+        title=title
     )
     
     # Return the figure

app.py

@@ -5,36 +5,65 @@ import matplotlib.pyplot as plt
 from mpl_toolkits.mplot3d import Axes3D
 from gradio.components import Slider, Number, Image, Dataframe, Textbox, Checkbox, Radio, Label, Plot
 import plotly.graph_objects as go
-import csv
-from aaaaa import get_datXYZ, plot_3d_points
+from aaaaa import get_datXYZ, plot_3d_points, genS
+from scipy.stats import multivariate_normal
 import PIL.Image
 
 data_path = "reference_data/"
 image_path = "images_configs/"
 
-def show_hand_3d(gesto, data_path, image_path):
+# To generate distorsed covariance matrixes
+def genS(sds, R, distor=1):
+    # sds: standard deviations
+    # R: correlation matrix
+    # distor: level of distorsion
+    sdsdis = sds * distor
+    S = R * np.outer(sdsdis, sdsdis)
+    np.fill_diagonal(S, sdsdis ** 2)
+    return S
+
+def show_hand_3d(gesto, attempt, distortion, mean, std):
     datXYZ = get_datXYZ(gesto, data_path)
     print(datXYZ.shape)
-    print(datXYZ)
-    first_row = datXYZ[0]  # Extract the first row of the array
+    # print(datXYZ)
+    obsmean = np.mean(datXYZ, axis=0)
+    print(obsmean.shape)
+    # change to 63 shape
+    obsmean = obsmean.reshape(63)
+    obssd = np.std(datXYZ, axis=0)
+    print(obssd.shape)
+    obssd = obssd.reshape(63)
+    datXYZ_reshaped = datXYZ.reshape(datXYZ.shape[0], -1)
+    R = np.corrcoef(datXYZ_reshaped, rowvar=False)
+    Sberria = genS(obssd, R, distor=distortion)
+    z = multivariate_normal.rvs(mean=obsmean, cov=Sberria, size=1)
+    print(z.shape)
+    # print(z)
+    # add here the code to transform z into an array eith the same shape as datXYZ
+    first_row = datXYZ[attempt]  # Extract the attempt row of the array
     reshaped_array = first_row[:63].reshape(21, 3)  # Reshape the first row to (21, 3)
-    hand_plot = plot_3d_points(reshaped_array)
+     # create noise from mean and std
+    noise = np.random.normal(mean, std, reshaped_array.shape)
+    reshaped_array_with_noise = reshaped_array + noise
+    distorted_hand = z.reshape(reshaped_array.shape)
+    hand_plot = plot_3d_points(reshaped_array, "Recorded Hand Configuration")
+    distorted_hand_plot = plot_3d_points(distorted_hand, "Distorted Hand Configuration (global)")
+    reshaped_array_with_noise_plot = plot_3d_points(reshaped_array_with_noise, "Distorted Hand Configuration (individual))")
     image_filename = image_path + str(int(gesto)) + ".png"
     print(image_filename)
     image = PIL.Image.open(image_filename)
-    width, height = image.size
-    newsize = (width // 2, height // 2)
-    image = image.resize(newsize)
     # image = Image(image_filename)  # Load the corresponding image
-    return image, hand_plot
+    return image, hand_plot, distorted_hand_plot, reshaped_array_with_noise_plot
 
 inputs = [
-    Number(value=42, precision=0, label="Number of Hand Configuration"),
-    Textbox(label="Data Path", value="reference_data/"),
-    Textbox(label="Image Path", value="images_configs/")
+    Number(value=42, precision=0, label="Number of Kind of Hand Configuration"),
+    Number(value=0, precision=0, label="Number of Concrete Hand Attempt"),
+    Number(value=1.0, label="Distortion Level"),
+    Number(value=0.0, label="Mean Distortion Level (one hand)"),
+    Number(value=0.006, label="Standard Deviation Distortion Level (one hand)")
 ]
 
-outputs = [Image(type="pil", shape=(50,50)), Plot()]
+outputs = [Image(type="pil"), Plot(), Plot(), Plot()]
 
 title = "Hand 3D Visualization"
 description = "Enter the gesture number (gesto) and the data path (data_path) to visualize the hand in 3D."