/vid2img

Image dataset from videos

Primary LanguageLua

vid2img

A video to image extrator converted from Atcold's original version of MATLAB.

Annotation

vid2img runs with assigned videos saved in Videos folder. Videos should be saved in corresponding categories. It can be run as follows.

cd Codes
qlua vid2img.lua

After images have been annotated with vid2img, patches can be extracted with crop.

th crop.lua

For a list of available parameters, run th crop.lua --help.

Directory structure

.
├── Codes
├── GUI
├── images
├── labels
├── torch-data
├── Patches
├── README.md
└── Videos

Main folders (uppercase)

  • Codes: where the vid2img scripts in Lua are saved.
  • GUI: where the graphical user interfaces designed in Qt for vid2img are saved.
  • Videos: videos are dropped in this folder with a specific category as the folder name. Inside the folder there is already a folder named exampleFolder containing 4 videos, named 01 - 04 for this particular case.

Local temporary folders (lowercase)

  • images: is generated by vid2img Lua script, which contains all the frames from chosen videos. The images are saved in the correponding folder as specified by the file names.
  • labels: contains the coordinates of the selected object within each (not skipped) frame.
  • patches: contains the extracted patches

License

MIT