mdiodato / cu-iccv-2019-learning-to-drive-winners Goto Github PK

Hi, I have reproduce your result by your code, thanks for your efforts.
But we have no lables for the test dataset , could you please provide some results on validation set to hlep me confirm whether the model I have trained is good or not?

The loss for my model is epoch5 | 858.24963 | 6.695664 | 864.945294 on validation with 106219 data.

The result in your paper is about 889 for model3 epoch2 on test dataset with 279863 data, so maybe my trained model is not that good??

Another question is that, I noticed your LSTM model takes two frames' features as input, but the input sequence is in an order of [t, t-4] instead of [t-4, t] where t is the frame index, could you give some explanation for this ?

Thanks very much if you can help!