Results on testing split

[sshaoshuai]

Hi,
Thanks for your great codes.
I run your codes with pyramid_cars_with_aug_example configuration, and final get the AP about (84.0, 74.0, 67.7) on val split, which is comparable with the results on your paper.
But when I use the best model to run on the test split, and I only get AP 56.4 on moderate split from the official testing service.
So do you know where is the problem?
Thanks very much.

[melfm]

We used the 50-50 split on the validation for a fair comparison with other methods, however for the submission to the test server, we used our own custom 85-15 split. We highly recommend you create your own custom split as well for submitting.

It might also depend on the ground-plane estimation technique you are using.

[sshaoshuai]

I see, thanks for your response.

[griffintin]

@melfm @sshaoshuai

Could you give some hint on generating ground-plane, which algorthm is suitable?

[hulaifeng]

@sshaoshuai @melfm @kujason Hi.How to set the ground-plane belong to the test set? Or Could you give some hint on generating ground-plane, which algorthm is suitable?

[hulaifeng]

@ @melfm create_train_test_txt.py
Could you please send the val.txt and training.txt of car ,pedestrian,cyclist to me? Here's my email: 201221100501@std.uestc.edu.cn
thanks

[tilltheworldend]

@sshaoshuai Hi.I run the codes with pyramid_cars_with_aug_example configuration and didn't change any parameters in it.But I only got （76.8，72.4，66.9）ap using the 120000 checkpoint.Do you know why cause this?