For detecting players in different teams using colors, I understand the problem you explained. But have you considered gray-scaling the image and then classify by the shade threshold? I found your channel and your work today and I am really interested in this project. I want to learn more so as to hopefully contribute to it.
Unfortunately many colors will come up as the same shade of gray after gray-scaling. There are different gray-scaling algos but image that we take simple RGB average. Team red (255, 0, 0) and team blue (0, 0, 255) play each other. The shade of gray will be 85 in both cases. Glad you like the channel! There are few videos on the channel that tie to this one. You should find them interesting ;)
@@Roboflow Oh yes you're absolutely correct. And yes I've been going through your videos today and I have lined up a few that I'm gonna watch more carefully and follow along. You deserve millions of subscribers
I have a MoviNet A3 stream that does action recognition (classification) its input shape is (1,1,256,256,3). What's the most efficient way of using it on videos 2500+ frames in length? Since same actions can span multiple frames frame skipping will lead to misses.
For detecting players in different teams using colors, I understand the problem you explained. But have you considered gray-scaling the image and then classify by the shade threshold?
I found your channel and your work today and I am really interested in this project. I want to learn more so as to hopefully contribute to it.
Unfortunately many colors will come up as the same shade of gray after gray-scaling. There are different gray-scaling algos but image that we take simple RGB average. Team red (255, 0, 0) and team blue (0, 0, 255) play each other. The shade of gray will be 85 in both cases.
Glad you like the channel! There are few videos on the channel that tie to this one. You should find them interesting ;)
@@Roboflow Oh yes you're absolutely correct.
And yes I've been going through your videos today and I have lined up a few that I'm gonna watch more carefully and follow along. You deserve millions of subscribers
@@NphiniT hahaha millions? thanks a lot! I'm hoping for 100k for now
Thank you so much
I have a MoviNet A3 stream that does action recognition (classification) its input shape is (1,1,256,256,3).
What's the most efficient way of using it on videos 2500+ frames in length?
Since same actions can span multiple frames frame skipping will lead to misses.
does the model process a single frame at the time?
@@Roboflow Yes
so I set a window of 8 then the detected action is the max of this window. 8 because the training videos were each 8 frames long