search

maudzung / TTNet-Real-time-Analysis-System-for-Table-Tennis-Pytorch

Unofficial implementation of "TTNet: Real-time temporal and spatial video analysis of table tennis" (CVPR 2020)
https://arxiv.org/pdf/2004.09927.pdf
586 stars 157 forks source link

Score and Game Analysis #13

Open koren88i opened 3 years ago

koren88i commented 3 years ago

Hey, After training the model I ran the test and the demo.. All I see in the output video is prediction for ball boune and net (when the ball is over the net).

Where are the stuff in the github video? score tracking etc?

IdeaKing commented 3 years ago

I had the same question too.

I read their research paper and in section 7 it states: "As it was mentioned above, the predictions of the proposed neural network should be supplemented with the serve event spotting approach. It was done with a separate convolutional neural network, which adopts architecture and the training method from the event spotting branch of the multi-task neural network."

And this isn't the official repo, and since their game rules module isn't public I'm guessing that you will have to develop your own version.

pinjiang commented 3 years ago

@koren88i Did you see the correct prediction of the ball? Here I see constant predicted place of the ball.

egehancosgun commented 3 years ago

Did anyone of you managed to get correct predictions? I am getting the same prediction for all the frames.

hz35572 commented 5 months ago

Hey, After training the model I ran the test and the demo.. All I see in the output video is prediction for ball boune and net (when the ball is over the net).

Where are the stuff in the github video? score tracking etc?

Have you developed your own version? I'd like to learn from it.