Summarizer icon indicating copy to clipboard operation
Summarizer copied to clipboard

hdf5 dataset generation

Open msaintja opened this issue 5 years ago • 1 comments
trafficstars

Allows the creation of an .h5 file from videos and .npy annotations (e.g. for the Twitch-LOL dataset).

May be extended later to generalize to other datasets (that don't have binary labels) or other formats than .npy for importing labels.

Possible improvements (could be added after merge):

  • [ ] Adding extra flags in generate_dataset.py to control the KTS parameters from utils/KTS/cpd_auto.py and utils/KTS/cpd_nonlin.pyx.
  • [ ] Handling non-binary score labels in addition to summary labels.
  • [ ] Handling other formats than .npy for importing labels.

msaintja avatar Jun 06 '20 16:06 msaintja

The SumMe and TVSum datasets have a 2-second uniform segmentation version (that's been used for Twitch-LOL) uploaded to a dedicated folder.

This could be useful for an ablation study for example.

msaintja avatar Jun 13 '20 20:06 msaintja