felixchenfy / Realtime-Action-Recognition

Apply ML to the skeletons from OpenPose; 9 actions; multiple people. (WARNING: I'm sorry that this is only good for course demo, not for real world applications !!! Those ary very difficult !!!)
MIT License
875 stars 256 forks source link

can we use for real time scenario? #66

Open yogidosalwar opened 3 years ago

yogidosalwar commented 3 years ago

Hi @felixchenfy ,

Can we use for the real time scenario? Means multiple people in a single frame.
You have mentioned: Highlights: 9 actions; multiple people (<=5); If I want use for more than 5 people, then where can I make changes.

Thank you!!!

felixchenfy commented 3 years ago

Hi, really sorry about the confusion! You don't need to change anything. The program can detect >5 people.
(I'm 95% sure. The 5% unsure is due to that I wrote it a long time ago -- 2 years ago)

The number "5" is just an estimation -- If there are >5 people in the image, each person will probably be too small and degrade the performance

yogidosalwar commented 3 years ago

Hi @felixchenfy , Thank you so much for the quick response.
The program can detect more than 5 people, but its not recognized action when multiple persons in the frame. When 2-3 persons in the frame, then model is giving better accuracy.

Thank you.

felixchenfy commented 3 years ago

The reason why action is not recognized may because of this criteria: https://github.com/felixchenfy/Realtime-Action-Recognition/blob/master/src/s5_test.py#L253-L255 When the size of the person is small, or the detected joints are too few, the program will skip that person.

Sorry about that, this project only works fine for 2~3 people, and their size in the image should be big

On Wed, Mar 3, 2021 at 3:44 AM yogidosalwar notifications@github.com wrote:

Hi @felixchenfy https://github.com/felixchenfy , Thank you so much for the quick response. The program can detect more than 5 people, but its not recognized action when multiple persons in the frame. When 2-3 persons in the frame, then model is giving better accuracy.

Thank you.

— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub https://github.com/felixchenfy/Realtime-Action-Recognition/issues/66#issuecomment-789655277, or unsubscribe https://github.com/notifications/unsubscribe-auth/AFJY7WPI2JTBUAOJB264VHDTBYOIZANCNFSM4YOMLEIQ .