text updated and plot torus

README.md

@@ -1,5 +1,5 @@
 <div align="center">
-<h1> ManiPose: Manifold-Constrained Multi-Hypothesis 3D Human Pose Estimation </h1>
+<h1> ManiPose: Manifold-Constrained Multi-Hypothesis 3D Human Pose Estimation (NeurIPS 2024) </h1>
 <h3> Cédric Rommel, Victor Letzelter, Nermin Samet, Renaud Marlet, Matthieu Cord, Patrick Pérez, Eduardo Valle
 </h3>
 <!-- <h4> <i> International Conference on Computer Vision (ICCV), <br> Workshop Analysis and Modeling of Faces and Gestures, 2023 </i></h4> -->
@@ -177,7 +177,7 @@ python main_3dhp.py \
     train.batch_size_test=30
 ``` -->
 
-# 1D-to-2D toy experiment
+# toy experiments
 
 Figure 4 can be reproduced in ~2 minutes using a ~10GB GPU. Just run the following command:
 ```bash
@@ -187,11 +187,21 @@ cd toy_experiment ; python plotting_script.py
 You should find the corresponding figures in the `./figures` folder:
 <p align="center"> <img src="./assets/single_toy_picture_column_v2.png" width="80%"> </p>
 
-Table 1 can also be reproduced in ~2 minutes using the following command:
+Table 1 can be reproduced using the following command:
 ```bash
 bash toy_experiment/quantitative_comparison_toy2d.sh
 ```
 
+The Figure 8 from 2D-to-3D toy experiment can be reprocuced and saved in `toy_experiment/images` by running
+```python toy_experiment/toy-plot.py
+python 
+``` 
+
+Table 6 can also be reproced using the command:
+```bash
+bash toy_experiment/quantitative_comparison_toy3d.sh
+```
+
 # Acknowledgments
 
 The basis of the human pose lifting code was borrowed from [DiffHPE](https://github.com/valeoai/diffhpe), which builds on top of several other repositories, including:

toy_experiment/data/distributions.py

@@ -306,9 +306,14 @@ def __init__(
         for c, mu, kappa in zip(self.components, self.modes, self.dispersions):
             self.dist_list.append(BivariateVonMises(phi_loc=mu[0], psi_loc=mu[1],phi_concentration=kappa[0],psi_concentration=kappa[1],correlation=0.))
 
-    def sample(self, size: int) -> Tuple[np.array, np.array]:
+    def sample(self, size: int, output_components=False) -> Tuple[np.array, np.array]:
         angles = super().sample(size)
         cartesian_points = self.torusanglestocartesian(major_radius=self.major_radius,minor_radius=self.minor_radius,angles=angles)
         assert np.shape(np.stack((cartesian_points[:,0],cartesian_points[:,2]),axis=-1)) == (size,2)
         assert np.shape(cartesian_points) == (size,3)
-        return np.stack((cartesian_points[:,0],cartesian_points[:,2]),axis=-1), cartesian_points
+
+        if output_components is True:
+            return np.stack((cartesian_points[:,0],cartesian_points[:,2]),axis=-1), cartesian_points, self.picked_components
+
+        else : 
+            return np.stack((cartesian_points[:,0],cartesian_points[:,2]),axis=-1), cartesian_points

toy_experiment/tor-plot.py

@@ -0,0 +1,101 @@
+#%%
+
+import os 
+
+os.system('pip install pyro-ppl')
+os.system('pip install ipympl')
+os.system('pip install entrypoints')
+
+#%%
+
+import numpy as np
+from data import (
+    LiftingDist2Dto3D)
+import rootutils
+rootutils.setup_root(__file__, indicator=".project-root", project_root_env_var=True, pythonpath=True)
+
+major_radius = 2
+minor_radius = 1
+R = major_radius
+r = minor_radius
+N = 10
+x = np.linspace(-(R+2*r), (R+2*r), N)
+y = np.linspace(-(R+2*r), (R+2*r), N)
+z = np.linspace(-r, r, N)
+X_test = np.stack([x,z], axis=1) #(N,2)
+Y_test = np.stack([x,y,z], axis=1) #(N,3)
+
+distribution = LiftingDist2Dto3D(
+            major_radius=2,
+            minor_radius=1,
+            weights= [0.3, 0.4,0.2,0.1],
+            modes = [(-3.1415, 0), (0,3.1415/4),(0.5, -3.1415/4), (2*3.1415/3,3.1415/2)],
+           dispersions=[(2,2), (4,4),(3,3),(10,10)],
+            random_state=123)
+
+_, points, components = distribution.sample(size=1000, output_components=True)
+angles = distribution.torus_cartesian_to_angles_batch(major_radius=2,minor_radius=1,points = points)
+
+#%%
+
+import numpy as np
+import matplotlib.pyplot as plt
+from mpl_toolkits.mplot3d import Axes3D
+from matplotlib.lines import Line2D
+
+def plot_torus(angles, point_colors, R=2, r=1):
+    """
+    Visualize points on a torus given their angular coordinates.
+
+    :param angles: A (N, 2) shaped array of angular coordinates (u, v).
+    :param R: Major radius of the torus.
+    :param r: Minor radius of the torus.
+    """
+    # Convert angular coordinates to Cartesian coordinates
+    u, v = angles[:, 0], angles[:, 1]
+    x = (R + r * np.cos(v)) * np.cos(u)
+    y = (R + r * np.cos(v)) * np.sin(u)
+    z = r * np.sin(v)
+
+    # Create a 3D plot
+    fig = plt.figure()
+    ax = fig.add_subplot(111, projection='3d')
+
+    # Labeling the axes
+    ax.set_xlabel('X axis')
+    ax.set_ylabel('Y axis')
+    ax.set_zlabel('Z axis')
+
+    # Creating a mesh for the torus
+    u = np.linspace(0, 2 * np.pi, 50)
+    v = np.linspace(0, 2 * np.pi, 50)
+    u, v = np.meshgrid(u, v)
+    x_tor = (R + r * np.cos(v)) * np.cos(u)
+    y_tor = (R + r * np.cos(v)) * np.sin(u)
+    z_tor = r * np.sin(v)
+
+    # Plotting the torus mesh
+    ax.plot_wireframe(x_tor, y_tor, z_tor, color='gray', alpha=0.3,zorder=1)
+
+    ax.scatter(x, y, z, c=point_colors,zorder=2)
+
+    elevation = 60
+    azimuth = -45
+    ax.view_init(azim=azimuth, elev=elevation)
+
+    # plt.title('Distribution of Points on the Torus')
+
+    plt.show()
+
+colors = ['blue', 'green', 'red', 'purple']
+point_colors = [colors[comp] for comp in components]
+# Example usage
+plot_torus(angles, point_colors= point_colors)  # Visualize the distribution on a torus
+
+if not os.path.exists(os.path.join(os.environ['PROJECT_ROOT'],'toy_experiment/images')):
+    os.makedirs(os.path.join(os.environ['PROJECT_ROOT'],'toy_experiment/images'))
+
+filepath = os.path.join(os.environ['PROJECT_ROOT'],'toy_experiment/images/torus.pdf')
+
+plt.savefig(filepath)
+# %%