3d关键点 可视化

目录

[open3d 可以展示](#open3d 可以展示)

[vedo 可以录屏，不能保存](#vedo 可以录屏，不能保存)

mayavi

---

open3d 可以展示

import pickle

import open3d as o3d
import numpy as np
import time

# 模拟关键点数据：50帧，每帧17个关节
n_frames = 50
n_joints = 17
# keypoints_3d = np.random.rand(n_frames, n_joints, 3)

pkl_path = r"D:\BaiduNetdiskDownload\pkl_mp4\pkl_mp4\0879d197-8b7a-43fc-97dd-f98bc5dc4214_1014.pkl"
video_masks = pickle.load(open(pkl_path, "rb"))
if isinstance(video_masks, tuple):
    video_masks = video_masks[0]

object_ids = [1]
video_mask = []
points = []

keypoints_3d=[]
for frame_id in video_masks.keys():
    mask_all = None
    print('frame_id',frame_id)
    object_dict = video_masks[frame_id]
    for object_id, object_item in object_dict.items():
        if hasattr(object_item, 'smpl') and object_item['smpl'] is not None:
            joints3d = object_item['smpl']['joints3d']

            keypoints_3d.append(joints3d[0])
keypoints_3d=np.array(keypoints_3d)

# 骨骼连接（按顺序连接，你可以自定义拓扑结构）
lines = [[i, i + 1] for i in range(n_joints - 1)]

# 初始点和线（从第0帧）
pts = keypoints_3d[0]

# 构建点云
points = o3d.geometry.PointCloud()
points.points = o3d.utility.Vector3dVector(pts)

# 构建骨架连线
lineset = o3d.geometry.LineSet()
lineset.points = o3d.utility.Vector3dVector(pts)
lineset.lines = o3d.utility.Vector2iVector(lines)
lineset.colors = o3d.utility.Vector3dVector([[0, 0, 1]] * len(lines))  # 蓝色

# 初始化可视化器
vis = o3d.visualization.Visualizer()
vis.create_window()
vis.add_geometry(points)
vis.add_geometry(lineset)

# 强制第一次渲染
vis.poll_events()
vis.update_renderer()

# 播放动画
for i in range(n_frames):
    pts = keypoints_3d[i]

    # 更新数据
    points.points = o3d.utility.Vector3dVector(pts)
    lineset.points = o3d.utility.Vector3dVector(pts)

    # 通知渲染器数据已变
    vis.update_geometry(points)
    vis.update_geometry(lineset)
    vis.poll_events()
    vis.update_renderer()

    time.sleep(0.1)  # 控制播放速度

vis.destroy_window()

vedo 可以录屏，不能保存

from vedo import Plotter, Points, Line
import numpy as np

# 生成测试数据
n_frames = 100
n_joints = 17
keypoints = np.random.rand(n_frames, n_joints, 3)

# 初始化场景
plt = Plotter(axes=1, interactive=False, offscreen=True)

# 创建初始对象时使用有效数据
initial_points = np.zeros((n_joints, 3))  # 初始化为零点的占位数据
points_obj = Points(initial_points, r=12, c='red')  # 重命名避免命名冲突

# 定义骨骼连接关系（COCO格式示例）
bones = [[0, 1], [1, 2], [2, 3],  # 头部
    [0, 4], [4, 5], [5, 6],  # 左臂
    [0, 7], [7, 8], [8, 9],  # 右臂
    [4, 10], [7, 11],  # 躯干连接
    [10, 11], [10, 12], [11, 13],  # 臀部
    [12, 14], [13, 15], [14, 16], [15, 16]  # 腿部
]

lines_obj = Line(initial_points, bones, lw=3, c='blue')

# 添加对象到场景
plt += [points_obj, lines_obj]

# 生成动画帧
for i in range(n_frames):
    current = keypoints[i]

    # 正确更新关键点坐标（使用赋值操作）
    points_obj.points = current  # 直接赋值而非调用方法

    # 更新骨骼连线
    lines_obj.points = current  # 同样使用赋值操作

    # 渲染并保存帧
    plt.show()
    # plt.screenshot(f"frame_{i:03d}.png")

plt.close()
print("视频合成命令: ffmpeg -framerate 30 -i frame_%03d.png -c:v libx264 -pix_fmt yuv420p output.mp4")

mayavi

import numpy as np
from mayavi import mlab

# 生成示例数据
n_frames = 20
n_joints = 17
keypoints_3d = np.random.rand(n_frames, n_joints, 3)

# 创建动画
@mlab.animate(delay=100)  # 每帧100ms
def animate():
    for i in range(n_frames):
        mlab.clf()  # 清空当前帧
        current_frame = keypoints_3d[i]
        
        # 绘制关键点
        pts = mlab.points3d(
            current_frame[:, 0], current_frame[:, 1], current_frame[:, 2],
            color=(1, 0, 0), scale_factor=0.05
        )
        
        # 绘制骨骼连线
        for j in range(n_joints - 1):
            mlab.plot3d(
                [current_frame[j, 0], current_frame[j+1, 0]],
                [current_frame[j, 1], current_frame[j+1, 1]],
                [current_frame[j, 2], current_frame[j+1, 2]],
                tube_radius=0.01, color=(0, 0, 1)
            )
        
        yield  # 切换到下一帧

animate()
mlab.show()