Attention map

[liulangwuyang]

Hi, how was the Figure 3 attention map drawn in the paper?

[ttttlh]

Hi, how was the Figure 3 attention map drawn in the paper?

Hi, you can refer to the following code for visualization.

import numpy as np
import open3d as o3d
from sklearn.manifold import TSNE
import matplotlib.pyplot as plt

# read your data,  including coordinates and features of point cloud
embedded_feat = TSNE(n_components=1, learning_rate='auto', init='pca', n_jobs=-1).fit_transform(feat)
feat_min = embedded_feat.min(0)
feat_max = embedded_feat.max(0)
norm_embedded_feat = (embedded_feat - feat_min) / (feat_max - feat_min)
norm_embedded_feat = norm_embedded_feat.reshape(-1)
colorMap = plt.get_cmap('jet_r')  # plt.get_cmap('jet')
color = colorMap(norm_embedded_feat)
color = np.array(color)[:, :3]

vis = o3d.visualization.Visualizer()
vis.create_window(width = 768, height = 768)

point_cloud = o3d.geometry.PointCloud()
point_cloud.points = o3d.utility.Vector3dVector(xyz)
point_cloud.colors = o3d.utility.Vector3dVector(color)

vis.add_geometry(point_cloud)
vis.get_render_option().point_size = 5.
vis.run()
vis.capture_screen_image('./')
vis.destroy_window()

[liulangwuyang]

Ok, thank you very much for your reply.

[liulangwuyang]

Hi, how was the Figure 3 attention map drawn in the paper?

Hi, you can refer to the following code for visualization.

import numpy as np
import open3d as o3d
from sklearn.manifold import TSNE
import matplotlib.pyplot as plt

# read your data,  including coordinates and features of point cloud
embedded_feat = TSNE(n_components=1, learning_rate='auto', init='pca', n_jobs=-1).fit_transform(feat)
feat_min = embedded_feat.min(0)
feat_max = embedded_feat.max(0)
norm_embedded_feat = (embedded_feat - feat_min) / (feat_max - feat_min)
norm_embedded_feat = norm_embedded_feat.reshape(-1)
colorMap = plt.get_cmap('jet_r')  # plt.get_cmap('jet')
color = colorMap(norm_embedded_feat)
color = np.array(color)[:, :3]

vis = o3d.visualization.Visualizer()
vis.create_window(width = 768, height = 768)

point_cloud = o3d.geometry.PointCloud()
point_cloud.points = o3d.utility.Vector3dVector(xyz)
point_cloud.colors = o3d.utility.Vector3dVector(color)

vis.add_geometry(point_cloud)
vis.get_render_option().point_size = 5.
vis.run()
vis.capture_screen_image('./')
vis.destroy_window()

By the way, what are the shapes of ‘xyz’ and ‘feat’? I have 1024 points, and the feature dimension is 64. Should i input [64,1024] feat to the TSNE function and use [1024,3] xyz to draw points? thank you!

[ttttlh]

Hi, how was the Figure 3 attention map drawn in the paper?

Hi, you can refer to the following code for visualization.

import numpy as np
import open3d as o3d
from sklearn.manifold import TSNE
import matplotlib.pyplot as plt

# read your data,  including coordinates and features of point cloud
embedded_feat = TSNE(n_components=1, learning_rate='auto', init='pca', n_jobs=-1).fit_transform(feat)
feat_min = embedded_feat.min(0)
feat_max = embedded_feat.max(0)
norm_embedded_feat = (embedded_feat - feat_min) / (feat_max - feat_min)
norm_embedded_feat = norm_embedded_feat.reshape(-1)
colorMap = plt.get_cmap('jet_r')  # plt.get_cmap('jet')
color = colorMap(norm_embedded_feat)
color = np.array(color)[:, :3]

vis = o3d.visualization.Visualizer()
vis.create_window(width = 768, height = 768)

point_cloud = o3d.geometry.PointCloud()
point_cloud.points = o3d.utility.Vector3dVector(xyz)
point_cloud.colors = o3d.utility.Vector3dVector(color)

vis.add_geometry(point_cloud)
vis.get_render_option().point_size = 5.
vis.run()
vis.capture_screen_image('./')
vis.destroy_window()

By the way, what are the shapes of ‘xyz’ and ‘feat’? I have 1024 points, and the feature dimension is 64. Should i input [64,1024] feat to the TSNE function and use [1024,3] xyz to draw points? thank you!

the shape of 'xyz' is Nx3 (i.e. [1024, 3]), the shape of 'feat' is NxF (i.e. [1024, 64]). N is the number of points, and F is the dimension of the feature.

[taibai123abc]

But the color doesn't represent the attention weight? How to address the problem?