import numpy as np
import pandas as pd
import motion_capture
import plotly.offline as py

Load evaluation results

dataframe = pd.read_csv("/dataset/chenk/release/23.csv", index_col=0)

dataframe.describe()

	error_2d	error_3d	error_abs
count	10821.000000	10821.000000	10821.000000
mean	0.031684	0.030930	0.014521
std	0.016641	0.020535	0.009980
min	0.003362	0.003906	0.001655
25%	0.021332	0.019316	0.008863
50%	0.028137	0.026260	0.012095
75%	0.038144	0.036871	0.016940
max	0.322908	0.446010	0.200418

Visualization of predicted 3D shape v.s. ground truth.

data = np.load("/dataset/chenk/release/23.npz")

index = 20
py.init_notebook_mode()
fig = motion_capture.get_figure3d(data["shapes"][index], data["groundtruth"][index])
py.iplot(fig)