swathikirans/ego-rnn

the test accuracy

Opened this issue · 5 comments

CVsaber commented

Hi @swathikirans
I trained the network in your way and then tested it in two stream approach, but in any case my accuracy could not reach 77.59% on the GTEA 61 dataset. My best result is 75%.So I want to ask how you trained? How to use RGB Model and Flow Model to train in two stream?
Thank you!

swathikirans commented

Hi @DEepLiker,

Run main-run-twoStream.py after training the RGB and Flow models separately. We reported the best performing model, after several runs, in the paper. Since GTEA61 is imbalanced and because of the stochastic nature of training, it is not guaranteed to obtain the exact same accuracy. I would suggest using EGTEA dataset since it is large enough to train on deep models.

CVsaber commented

Thanks a lot
when i run main-run-twoStream.py, I see i need to choose the RGB model and Flow model, I want to ask which model should I choose? the last epoch model or the val best model? or others?

swathikirans commented

Both the model from last epoch and best model should ideally give the best result since we are training down till the last layers of the individual streams.

  • 👍1
frostinassiky commented

@DEepLiker When you run main-run-twoStream.py, do you change any parameters?
I got only 71.55% for joint training.
Now I am testing different hyper-parameters with the hope that the score can change - -

  • 👀1
chenwenhe11 commented

@frostinassiky I have the same problem, have you solved it？