Ecosyste.ms: Awesome
An open API service indexing awesome lists of open source software.
https://github.com/MTCloudVision/mxnet-videoio
Mxnet video io reading performance optimization
https://github.com/MTCloudVision/mxnet-videoio
Last synced: 2 months ago
JSON representation
Mxnet video io reading performance optimization
- Host: GitHub
- URL: https://github.com/MTCloudVision/mxnet-videoio
- Owner: MTCloudVision
- License: bsd-3-clause
- Created: 2018-03-29T03:40:48.000Z (almost 7 years ago)
- Default Branch: master
- Last Pushed: 2018-03-29T04:41:33.000Z (almost 7 years ago)
- Last Synced: 2024-08-01T22:41:43.397Z (5 months ago)
- Language: Python
- Homepage:
- Size: 10.7 KB
- Stars: 43
- Watchers: 4
- Forks: 6
- Open Issues: 1
-
Metadata Files:
- Readme: README.md
- License: LICENSE
Awesome Lists containing this project
- Awesome-MXNet - mxnet-videoio
README
本工程封装MXNet Rec接口，实现视频I/O读取速度优化.
优化前后性能对比
-----
*mxnet*: 1.0.1\
*network*: BN-Inception\
*machine*: GTX 1080ti\
*data classes*: 101\
*number of frames*: 3f/video\
*frame resolution*: 256x320
| 读取方式 | 单个gpu | 4个gpu |
| --------------- | :----------------------------: | -------------------------------------- |
| cv2.imread，单线程 | 速度：23.44samples/sec 显存: 9491MB | 速度： 26.81samples/sec 显存: 12274MB |
| cv2.imread，多线程（10） | 速度：23.44samples/sec 显存: 9481MB | 速度： 26.81samples/sec 显存: 12274MB |
| 优化后 | 速度：120.50 samples/sec 显存: 9497MB | 速度： 464.67 samples/sec 显存: 12298MB |
接口使用姿势
-----
1:本工程未提供截帧程序，视频需要提前截帧，每个视频对应的帧图像存放在同一个文件夹
2:遍历文件夹，生成训练列表文件，文件每行描述一个视频信息，格式为：‘文件夹绝对路径 ，截取的帧图像数，视频所属类别’
3:运行video2rec.py读取txt文件，将图像打包成rec文件
4:运行dataloader.py读取rec文件，并打包成 VideoDataIter