【Tanks and Temples】视频帧提取-->用于三维重建

三维重建第一步：从视频中提取若干图像帧用于后续的三维重建，以Tanks and Temples数据集Training Set中的Ignatius场景为例进行说明

1. Prepare Images

• 图像序列：直接到Step 2
• 视频: 需要将视频转为图像帧，具体如下步骤：
  

Ignatius.mp4视频信息

• 视频帧率：29.97FPS
• 图像分辨率：3840x2160 (4K分辨率)
• 帧数：~8000帧

Get jpg images from the video:
> ffmpeg -i Ignatius.mp4 -q:v 1 -vf fps=1 Ignatius/%06d.jpg

Resize:
> mogrify -quality 100 -resize 50% -path Ignatius2/ Ignatius/*.jpg

Write EXIF data:
> exiftool -overwrite_original -FocalLength="21" -focallengthin35mmformat="21" -make="Sony" -model="A7SM2" Ignatius2/*.jpg

ffmpeg命令行中fps=1表示每秒取一帧图像，输出图像共263张图像；-q:v 1表示输出的图片质量，一般是1到5之间（1 为质量最高）quality

2. Make a Reconstruction

• colmap cmd 命令行重建：https://www.cnblogs.com/Todd-Qi/p/10792685.html
• colmap GUI 图形化界面重建：https://www.cnblogs.com/Todd-Qi/p/17250204.html

3. Evaluation

详见官网

4. Tanks and Temples数据集

Tanks and Temples是Intel Labs于2017年在SIGGRAPH上发表的文章

+── Tanks and Temples benchmark
│   +── training data
│   +── testing data
│   │   +── intermediate
│   │   +── advanced

📃 Tanks and Temples

This dataset provides both indoor and outdoor scenes. This dataset is divided into training data and testing data. The testing datasets are organized into two groups: **intermediate** and **advances**. The intermediate group contains sculptures, large vehicles, and house-scale building with outside-looking-in camera trajectories. Additionally, for training datasets, ground-truth geometry is provided for training purposes.

参考链接

• https://www.tanksandtemples.org/tutorial/
• https://exiftool.org/
• https://imagemagick.org/script/mogrify.php