Data from "Learning a Hand Model from Dynamic Movements using High-density EMG and Convolutional Neural Networks"

FormatThis dataset contains the ground truths and predictions of all deep learning models from Sîmpetru et al (https://doi.org/10.1101/2022.07.29.502064).
Fig. 2-4The data is formatted as follows:
The data used for Fig. 2 is stored under 3D_points_{random, testset} depending on whether the AI was tested with the random or the test set.The data used for Fig. 3 is stored under angle_{random, test set} depending on whether the AI was tested with the random or the test set.The data used for Fig. 4 is stored under force.Each of these folders contains the predicted and expected data for each of the 13 subjects stored as Subject{1, ..., 13}_{predicted, expected}.pkl.Fig. 5The data for Fig. 5 is stored under comparison.

To distinguish between the EMG forms used, we use:

5 Hz low-pass filtered as 3D_points_55 Hz low-pass filtered and rectified as 3D_points_5_rectified20 Hz low-pass filtered as 3D_points_2020 Hz low-pass filtered and rectified as 3D_points_20_rectifiedraw as 3D_points_rawraw and rectified as 3D_points_raw_rectifiedraw and 20 Hz low-pass filtered combined as 3D_points_raw20
Each of these folders contains the predicted and expected data for each of the 4 subjects, stored as Subject{1, ..., 4}_{predicted, expected}.pkl.

General file formatThe saved .pkl files are tables where the columns represent the joint labels (Fig. 1B) and the rows represent the bin number.

Each cell of the table contains a series of 3 values representing the x, y and z coordinates of the respective joint label at a given time.

LoadingThe stored files are pandas (high-level python library for tables manipulation) dataframes stored using pickle (python object serialisation library).

To read such a file, we give the following minimal example:

import pandas as pd

dataframe = pd.read_pickle("PATH_TO_FILE.pkl")

# display head (top 5 rows) of the dataframe
print(dataframe.head())