PKU MMD and NTU RGBD are large skeleton datasets widely used in human action recognition. There are plenty of codes available to process the NTU RGBD dataset. But i cannot find any code available to process PKU MMD data set and convert it to the same format as in NTU RGBD. if anybody knows the preprocessing steps and code to preprocess PKU MMD, please share it.
Thank you.
Dear Shanaka Ramesh Gunasekara ,
The PKU-MMD dataset is a large skeleton-based action detection dataset. It contains 1076 long untrimmed video sequences performed by 66 subjects in three camera views. 51 action categories are annotated, resulting almost 20,000 action instances and 5.4 million frames in total. Similar to NTU RGB+D, there are also two recommended evaluate protocols, i.e. cross-subject and cross-view.
https://paperswithcode.com/dataset/pku-mmd
Regards,
Shafagat
onverting PKU-MMD dataset to the NTU RGB+D format involves several steps, including data preprocessing and formatting. Here are the steps to convert PKU-MMD to NTU RGB+D:
Here's a code snippet that demonstrates the conversion process:
import numpy as np
import os
# Load the joint positions
data_path = "/path/to/pku-mmd/dataset"
sample_name = "S001C001P001R001A001.skeleton"
joint_data = np.load(os.path.join(data_path, sample_name))
# Normalize the joint positions
mean_joint = np.mean(joint_data, axis=(0, 1))
std_joint = np.std(joint_data, axis=(0, 1))
normalized_data = (joint_data - mean_joint) / std_joint
# Format the data to the NTU RGB+D format
num_frames = 60
num_joints = 25
formatted_data = np.zeros((num_frames, num_joints, 3))
num_available_frames = min(normalized_data.shape[0], num_frames)
formatted_data[:num_available_frames, :, :] = normalized_data[:num_available_frames, :, :]
formatted_data = formatted_data.reshape((1, num_frames, num_joints, 3))
# Save the formatted data to a new file
output_path = "/path/to/ntu-rgbd/dataset"
output_name = "S001C001P001R001A001.npy"
np.save(os.path.join(output_path, output_name), formatted_data)
Note that you'll need to repeat the above steps for each sample in the PKU-MMD dataset to convert the entire dataset to the NTU RGB+D format.
- Badges
- Science topic
- Similar topics
- Mathematical Sciences
- Graphs