video_augmentation

Recopilation of tools for data augmentation with videos

Quickstart

Just download the video_augmentation.py file, copy it in your working directory and use:

import video_augmentation as va

Now everything is set to use all the functionalities of the repository!

Try, for example, the following code:

import video_augmentation as va
seed(1974)

## Folders
input_dir = "data/"
# input_file = "data/example.mp4" # TODO
input_format = "mp4"

output_dir = "data/augmented/"
output_format = "mp4" # Also accepts "avi"

## Booleans
save_video = False # Save the video in output_dir?
show_size = True # Show the size of the video in the title?
show_video = True # Show the video at all?
slow = True # Play the video at real rate or at opencv speed?

## Other parameters
seconds_before_action = 0 # Seconds before watching the video

### Execution
va.augmentate(input_dir, output_dir, input_format, output_format, show_video, save_video, slow, show_size, seconds_before_action, ["aff"])

What can you do with this repository?

For now, take a look at the Quickstart section or try running the example.

Status

• Of the repository:
  • Complete README.md
• Of the main function:
  • Improve verbosity
  • Include a progress bar
  • Make active use of the GPU to boost performance
  • Use the augment function to create a multi_augment function in order to allow a quick deployment of a new dataset
• Included transformations:
  • Affinity
    • Scaling
    • Homothety
    • Translation
    • Rotation
    • Random w/o control of the randomness
    • Random w/ control of the randomness
  • Salt & Pepper
    • Pseudorandom
    • Efficient pseudorandomization
    • Efficient randomization
    • For every size
  • Blur
    • Median-based blur
  • Upsampling & Downsampling
    • With custom probabilities
  • Darken & Lighten

Other resources

• Data augmentation in Spanish
• Project webpage

Used data

• example_different_size.mp4 taken from Pixabay
• example_small.jpg taken from Pixabay
• example.jpg taken from Pixabay
• example.mp4 taken from Pixabay
• example2.mp4 taken from Pixabay
• work_in_progress.png taken from Pixabay