Docker容器中使用Open3D保存点云可视化图像问题求助

问题背景 #

• 已通过Open3D实现点云可视化，Docker容器内可正常显示，但保存图像时全黑，报错error: XDG_RUNTIME_DIR not set in the environment.
• 需求：生成与现有Open3D代码效果一致的可视化图并保存，不限使用库，需保持原视角。

原Open3D代码：

import open3d as o3d
import numpy as np
import time

pcd = o3d.geometry.PointCloud()
pcd.points = o3d.utility.Vector3dVector(points)
# Create coordinate frame for reference
coordinate_frame = o3d.geometry.TriangleMesh.create_coordinate_frame(size=0.3)

# Create visualization window
vis = o3d.visualization.Visualizer()
vis.create_window(width=1024, height=768)
vis.add_geometry(pcd)
vis.add_geometry(coordinate_frame)

# Set the camera position for a better view
ctr = vis.get_view_control()
ctr.set_front([0, 0, -1])   # Look toward the scene (forward)
ctr.set_up([0, -1, 0])      # Y is up
ctr.set_zoom(0.7)

# Improve the rendering quality
opt = vis.get_render_option()
opt.point_size = 3.0
opt.background_color = np.array([0.1, 0.1, 0.1])  # Dark gray background

# Render the visualization
vis.poll_events()
vis.update_renderer()

# Display for a while
time.sleep(20)

vis.capture_screen_image(save_path)

# Clean up
vis.destroy_window()

Docker容器配置：

default_test:
  privileged: true
  ipc: host # for dds shared memory transport
  environment:
    - QT_X11_NO_MITSHM=1
    - QT_DEBUG_PLUGINS=1
  volumes:
    - /tmp/.X11-unix:/tmp/.X11-unix
    - /dev/shm:/dev/shm 
  network_mode: host
  runtime: nvidia
  deploy:
    resources:
      reservations:
        devices:
          - driver: nvidia
            count: 1
            capabilities: [ gpu ]

解决方案 #

方法1：Open3D离线无窗口渲染（推荐） #

完全绕过X11窗口依赖，直接后台渲染图像，视角与原代码完全匹配：

import open3d as o3d
import numpy as np

pcd = o3d.geometry.PointCloud()
pcd.points = o3d.utility.Vector3dVector(points)
coordinate_frame = o3d.geometry.TriangleMesh.create_coordinate_frame(size=0.3)

# 配置渲染参数（与原代码一致）
render_option = o3d.visualization.RenderOption()
render_option.point_size = 3.0
render_option.background_color = np.array([0.1, 0.1, 0.1])

# 配置相机参数，匹配原视角
camera_params = o3d.camera.PinholeCameraParameters()
# 设置内参：对应1024x768分辨率，默认焦距500
camera_params.intrinsic.set_intrinsics(1024, 768, 500, 500, 512, 384)
# 根据原代码的front、up、zoom计算相机姿态
camera_params.extrinsic = o3d.camera.PinholeCameraParameters.create_from_look_at(
    eye=[0, 0, 1/0.7],  # zoom=0.7对应相机到原点的距离为1/0.7
    center=[0, 0, 0],    # 注视点为原点
    up=[0, -1, 0]        # 向上方向为Y轴负方向
)

# 离线渲染图像
img = o3d.visualization.rendering.render_to_image(
    geometry_list=[pcd, coordinate_frame],
    camera_param=camera_params,
    render_option=render_option,
    width=1024,
    height=768
)

# 保存图像
o3d.io.write_image(save_path, img)

方法2：修复Docker环境变量与代码渲染逻辑 #

如果坚持使用原可视化窗口方式，补充环境变量并调整渲染逻辑：

1. 修改Docker-compose配置，添加缺失的环境变量：

environment:
  - QT_X11_NO_MITSHM=1
  - QT_DEBUG_PLUGINS=1
  - XDG_RUNTIME_DIR=/tmp/runtime-user
  - DISPLAY=$DISPLAY

2. 容器启动时创建并授权XDG_RUNTIME_DIR目录：

mkdir -p /tmp/runtime-user && chmod 700 /tmp/runtime-user && chown $(id -u):$(id -g) /tmp/runtime-user

3. 修改原代码，替换sleep为强制渲染刷新：

# 替换原poll_events、update_renderer和sleep部分
vis.poll_events()
vis.update_renderer()
# 强制多次刷新确保渲染完成
for _ in range(10):
    vis.poll_events()
    vis.update_renderer()

# 添加do_render=True强制触发渲染
vis.capture_screen_image(save_path, do_render=True)

方法3：替代库Matplotlib实现 #

若Open3D仍有问题，可使用Matplotlib绘制点云，手动匹配原视角：

import matplotlib.pyplot as plt
from mpl_toolkits.mplot3d import Axes3D
import numpy as np

# 创建画布，匹配原分辨率
fig = plt.figure(figsize=(1024/100, 768/100), dpi=100)
ax = fig.add_subplot(111, projection='3d')

# 绘制点云
ax.scatter(points[:, 0], points[:, 1], points[:, 2], s=3, c='white', alpha=1.0)

# 匹配原视角：front=[0,0,-1]（沿-Z方向看）、up=[0,-1,0]（Y轴向下为上）
ax.view_init(elev=90, azim=0)
# 设置坐标轴范围，适配点云大小（可根据实际点云调整）
ax.set_xlim3d(-1, 1)
ax.set_ylim3d(-1, 1)
ax.set_zlim3d(-1, 1)
# 设置背景色
ax.set_facecolor((0.1, 0.1, 0.1))
fig.patch.set_facecolor((0.1, 0.1, 0.1))

# 添加参考坐标系
ax.quiver(0, 0, 0, 0.3, 0, 0, color='r', linewidth=2)  # X轴红色
ax.quiver(0, 0, 0, 0, 0.3, 0, color='g', linewidth=2)  # Y轴绿色
ax.quiver(0, 0, 0, 0, 0, 0.3, color='b', linewidth=2)  # Z轴蓝色

# 保存图像，避免裁剪
plt.savefig(save_path, facecolor=(0.1, 0.1, 0.1), bbox_inches='tight', pad_inches=0)
plt.close()

内容的提问来源于stack exchange，提问作者Frank