Datasets:
Failed installation for "omnifall"
Hi all! Thanks a lot for the amazing work you did are are still doing!
I have just tried to install the omnifall package via pip install omnifall, but it keeps failing saying it does not exist?
Any idea how I can solve this? Do I need a specific python version?
Also, I could not really find that mentioned anywhere else other than here on HF, is there (or will there be) more information about the package?
Thanks!
Dear simopippa,
I will upload the package in the next few days. I have not yet had the opportunity to test it fully.
If you are interested in the synthetic dataset, the videos are already available here:
https://huggingface.co/datasets/simplexsigil2/omnifall/blob/main/data_files/omnifall-synthetic_av1.tar
The labels should match the videos directly.
For the Omnifall-ITW dataset, you can download the OOPS Dataset and then apply the following path mapping:
https://huggingface.co/datasets/simplexsigil2/omnifall/blob/main/data_files/oops_video_mapping.csv
After applying this mapping, our labels should align with the videos.
The main purpose of the package is to perform these steps automatically and to provide a standardized data loader.
For the remaining datasets, the process is slightly more involved. In some cases the frame rate must be adapted. The relevant scripts for this step are available here. Especially for UP-Fall you need to use our approach or it will not fit (UP-Fall needs variable framerate and sometimes frame gaps can be up to 0.5 seconds):
https://github.com/simplexsigil/omnifall-helper-scripts/tree/master/video_conversion
The Omnifall package is intended to automate this process as well, including downloading and converting the datasets automagically as far as possible (currently CMDFall needs to be requested from the authors).
Hi and thank you for the very detailed answer! That is greatly appreciated!
That sounds great that you are planning to automate all this, in my current work I had already done something similar, mapping your labels to the original videos and load them together, but it would be awesome if that come directly from an installable package!
Already with the labels you made our work so much easier, now it seems like it will be even better!
I will wait until the package can be installed correctly, if you remember maybe tag me in this message, else I will check myself every here and there.