About Testing

[AmingWu]

Where are the testing results?

[HaozheQi]

Hi, @AmingWu, actually 'succ/prec' in the picture is the result, but there may be something wrong about your result. The true result is about 56.2/72.8. you may need to check the input settings of the code.

[AmingWu]

I find the testing data does not include label_02. The label_02 is only for training. Is it correct?

[AmingWu]

[AmingWu]

data_tracking_label_2 only contains the training data. And it does not contain the test data.

[HaozheQi]

Yes, the KITTI tracking dataset only gives the label of training set. So we only used its training set to train and test our P2B, which was split as follows: scenes 0-16 for training, 17-18 for validation, and 19-20 for testing.

[AmingWu]

I retrain the code. However, I still could not reproduce your results.

[AmingWu]

I use the default settings of your code.

[HaozheQi]

The result 6.1/3.9 indicates the tracker can not give a right prediction at all, I think there may still exist some errors in your code. Can you get a right result with the model I trained?

[AmingWu]

This is the training process. Is it correct?

[AmingWu]

The training is right?

[HaozheQi]

The output of your terminal seems to be correct. I'm not sure why your test result is wrong.

[AmingWu]

I use the default settings. And I do not modify the code. The dataset is correct. But, I do not reproduce your result.

[AmingWu]

Could you share your trained model?

[HaozheQi]

The model trained for car is already included in this repository, and I'm working on the release of other categories.

[AmingWu]

OK, I have test your model. I find the result is still wrong.

[HaozheQi]

You may check the code related to the test stage, I think your training stage has no error now.

[AmingWu]

dataset_Test = SiameseTest( input_size=1024, path='/home/room/wam/P2B/trackingdata/testing', split='Test', category_name='Car', offset_BB=0, scale_BB=1.25)

I only modify the path from opt.data_dir to '/home/room/wam/P2B/trackingdata/testing',

[HaozheQi]

The path of training and test should be the same, does this matter?

[AmingWu]

Thank you very much. I have reproduced your results. The path of training and test should be the same.

[AmingWu]

If I want to run other categories, I only need to modify parser.add_argument('--category_name', type=str, default = 'Car', help='Object to Track (Car/Pedetrian/Van/Cyclist)')?

[zhanghm1995]

@AmingWu @HaozheQi, Hello, I wonder how did you final solve this problem, I run the following scripts:

python test_tracking.py --data_dir=/mnt/study/Datasets/KITTI/tracking/training

And also got the result like yours:

Test 1: Time 0.030s (it:0.031s) Data:0.409s (it:0.518s), Succ/Prec:6.6/4.4: 100%

And I used the KITTI training data and the default offline netR_36.pth model file.

[zjtgit]

@AmingWu @HaozheQi, Hello, I wonder how did you final solve this problem, I run the following scripts:

python test_tracking.py --data_dir=/mnt/study/Datasets/KITTI/tracking/training

And also got the result like yours:

Test 1: Time 0.030s (it:0.031s) Data:0.409s (it:0.518s), Succ/Prec:6.6/4.4: 100%

And I used the KITTI training data and the default offline netR_36.pth model file.

I run the test script in pycharm, and I got the same result as yours. I am sure the data which I used to test is KITTI training sequence, but it's still give me a wrong result.

Test 1: Time 0.037s (it:0.041s) Data:0.436s (it:0.493s), Succ/Prec:6.6/4.4: 100%|██████████| 6424/6424 [03:56<00:00, 27.15it/s]