EgoExo4D Aria - Intrinsic Parameters

[rupalsaxena]

Hello,

Thank you for amazing work on EgoExo4D dataset!

I have a doubt regarding intrinsic parameters!

For Aria "camera-rgb" sensor, I can get intrinsic parameters from two sources:

1. From json file: annotation/ego_pose/train/camera_pose/uuid.json

2. Using projectaria_tools, example code below:

   class VrsHandler:
   def __init__(self, vrs_file, sensor_name = "camera-rgb"):
       self.provider = data_provider.create_vrs_data_provider(vrs_file)
       assert self.provider is not None, "Cannot open file"
       self.sensor_stream_id = self.provider.get_stream_id_from_label(sensor_name)
   
       # get calib and calibration algo
       self.initial_calib = self.provider.get_device_calibration().get_camera_calib(sensor_name)
       self.w, self.h = self.initial_calib.get_image_size()
       self.updated_calib = calibration.get_linear_camera_calibration(
           self.w, 
           self.h, 
           self.initial_calib.get_focal_lengths()[0], 
           "pinhole",
           self.initial_calib.get_transform_device_camera()
       )
       self.sensor_name = sensor_name
   
   def get_intrinsic_mattress(self):
       fx, fy = self.updated_calib.get_focal_lengths()
       cx, cy = self.updated_calib.get_principal_point()
   
       self.K = np.array([
           [fx, 0,   cx],
           [0,   fy, cy],
           [0,   0,   1]
       ])
   
       return self.K
   

   Now using both the methods, I am getting different K mattress. For eg for georgiatech_covid_09_2, from 1st method, I got [[150.0, 0.0, 255.5] [0.0, 150.0, 255.5] [0.0, 0.0, 1.0]]

From 2nd method, i.e. the code, I wrote, I got intrinsic mattress as [[610.52437819, 0, 703.5 ] [ 0, 610.52437819, 703.5 ] [ 0, 0, 1 ]]

I am confused, why are these values different??

Please let me know, thanks :) Best regards :)