Open ashwinvaswani opened 4 years ago
Can also look at this: https://github.com/immuno121/Automatic-highlight-generation-from-Sports-Videos
I would like to take this up
The model in the link you provided does not incorporate audio information so might not be the best suitable approach. I found a paper which suggests the use of semantic concept mining and incrporates both audio and visual data:
sci-hub.tw/10.1007/s11042-009-0337-1
Can you take a look at this and see if we can use this
Yeah we can definitely use that imo, looks like really good work. However implementing the network might be difficult if we start from that. I think we should first build something simple, like in this article https://www.analyticsvidhya.com/blog/2019/09/guide-automatic-highlight-generation-python-without-machine-learning/, and then slowly improve it to the network in the paper. Once we have something of a base which works, imo it will be easier for us to improve on it later on.
Ya makes sense. I'll start implementing what's given in the link
I have implemented this. Should I open a PR if this approach is going to be incorporated in the library?
Yes put up a PR
Is there any need for enhancement right now or can the issue be closed?
I think let's keep it open for now. We are aiming to add the enhancements ultimately after all.
@Rijul1999 Basic implementation is done. Check on additional work in this
A good starting point would be to use the intensity (pitch / something else) of audio from commentary to filter out "exciting events" and compile the clips from those events to generate highlight.