bvh格式转换

AI算法网奇

已于 2024-01-23 01:53:00 修改

阅读量773

点赞数 7

分类专栏： aigc与数字人姿态检测文章标签：人工智能

于 2024-01-16 16:33:18 首次发布

本文链接：https://blog.csdn.net/jacke121/article/details/135628432

版权

aigc与数字人同时被 2 个专栏收录

91 篇文章 6 订阅

订阅专栏

姿态检测

34 篇文章 7 订阅

订阅专栏

BVH_reader和bvh-converter 一致

bvh格式介绍

[转载] BVH文件格式解析 - 知乎

3d点转bvh

https://github.com/Garfield-kh/PoseTriplet/blob/eb93132f99161bd776dafbcb713e9fb43e501c36/estimator/pose2bvh_debug.py

npz格式：

这个可能是标准的：

https://github.com/HW140701/VideoTo3dPoseAndBvh/blob/bb06a001a91482b6ece76b9cb672dd50b6e4dd3a/bvh_skeleton/smartbody_skeleton.py

保存：

 np.savez_compressed(output_npz, reconstruction=output_3d_all)

读取：

    prediction = np.load(r'D:\data\output_keypoints_3d.npz', allow_pickle=True)

    for key in prediction.keys():
        prediction = prediction[key]

    prediction = prediction.astype('float32')

bvh格式转换：

https://github.com/Garfield-kh/PoseTriplet/blob/eb93132f99161bd776dafbcb713e9fb43e501c36/estimator/bvh_skeleton/bvh_helper.py

bvh-converter ok

pip install bvh-converter

bvh-converter 文件名

or：python -m bvh-converter 文件名

import csv
import os
import bpy

objects = bpy.context.scene.objects

empties = []

for object in objects:
    if object.type == 'EMPTY':
        empties.append(object)
        
print(empties)

filename = 'csv_joined.csv'
directory = r'hmr/output/csv_joined'  # <-- if you have linux or osx

fullpath = os.path.join(directory, filename)

with open(fullpath, 'r', newline='') as csvfile:
    ofile = csv.reader(csvfile, delimiter=',')
    next(ofile) # <-- skip the x,y,z header

    for line in ofile:
        f, *pts = line

        # these things are still strings (that's how they get stored in the file)
        # here we recast them to integer and floats
        frame_num = int(f)
        fpts = [float(p) for p in pts]
        coordinates = [fpts[0:3], fpts[3:6], fpts[6:9], fpts[9:12],
                       fpts[12:15], fpts[15:18], fpts[18:21], fpts[21:24],
                       fpts[24:27], fpts[27:30], fpts[30:33], fpts[33:36],
                       fpts[36:39], fpts[39:42],fpts[42:45], fpts[45:48],
                       fpts[48:51], fpts[51:54], fpts[54:57], fpts[57:60]]
                       

        bpy.context.scene.frame_set(frame_num)
        bpy.data.scenes['Scene'].frame_end = frame_num + 1
        
        for ob, position in zip(empties, coordinates):
            ob.location = position
            ob.keyframe_insert(data_path="location", index=-1)

bpy.data.objects['rig'].select = True

target_file = 'hmr/output/bvh_animation/estimated_animation.bvh'

bpy.ops.export_anim.bvh(filepath=target_file)

pkl转换bvh

【姿态识别】PoseTriplet——3D转bvh-CSDN博客

代码地址：
https://github.com/Garfield-kh/PoseTriplet/blob/eb93132f99161bd776dafbcb713e9fb43e501c36/estimator/posegan_evaluate.py#L360

运行posegan_evaluate.py

环境安装
sklearn

pip install scikit-learn
pip install torchgeometry
問題解決
No such file or directory: ‘data/data_3d_h36m.npz’
下载地址：https://download.csdn.net/download/weixin_46870724/88584488

FileNotFoundError: [Errno 2] No such file or directory:
‘data/data_2d_h36m_gt.npz’
下载地址：https://download.csdn.net/download/weixin_46870724/88584804

FileNotFoundError: [Errno 2] No such file or directory: './data_cross/3dhp/3dhp_testset_bySub.pkl'
下载地址：https://download.csdn.net/download/weixin_46870724/88585024?spm=1001.2014.3001.5503

運行
运行pose2bvh_debug.py
代码中配置好生成3D的pkl文件即可

原文链接：https://blog.csdn.net/weixin_46870724/article/details/134719737