Experimental Details

[zhangyahu1]

Thanks for sharing your great work! I have a question about your experimental details on CMU Panoptic dataset: which sequences do you train and test on?

[miaowu99]

train: 160224_haggling1 160226_mafia2 160224_mafia1 160224_mafia2 160224_ultimatum1 160224_ultimatum2

valid: 160422_mafia2

test sampled from： 160226_haggling1 160422_haggling1 160226_mafia1 160226_ultimatum1 160422_ultimatum1 160906_pizza1

[zhangyahu1]

Thanks for quick your response! I download the dataset. However, some 3D annotations from activities are not provided, such as the activity of mafia, 160224_ultimatum1 and 160224_ultimatum2. How do you solve this problem?

[miaowu99]

This is because some annotations of Panoptic only support mpi15, you can try to remove the suffix "coco19" of the download path.

[zhangyahu1]

Thanks!

[nicolasugrinovic]

Hi, Kudos for the great work!

For the test data, do you follow Zanfir et. al.? (ref [41] in your paper). As I understand they use the following activities for their test data and some selected frames which make up to 9600 images. The activities are:

1. '160422_ultimatum1',
2. '160422_haggling1',
3. '160422_mafia2',
4. 160906_pizza1',

Which frames did you use for each activity? As I see that you have additional test data: _160226haggling1 and _160226ultimatum1

Thank you in advance.

[xiaochehe]

train: 160224_haggling1 160226_mafia2 160224_mafia1 160224_mafia2 160224_ultimatum1 160224_ultimatum2

valid: 160422_mafia2

test sampled from： 160226_haggling1 160422_haggling1 160226_mafia1 160226_ultimatum1 160422_ultimatum1 160906_pizza1

Dear authors, do the training samples also come from the cameras 16 and 30?