aaltovision / advio Goto Github PK

Hi, thanks for your work and published the dataset! Since I cannot read the frames from videos which exactly matching the frames.csv, it is always less frames than the csv file, could you please provide image dataset for each frame? Or some scripts that can read correct frames from videos?
Thanks!

I would like first to thank you for your effort in collecting the data and making it public for others to use.I want to ask about accelerometer.csv and gyro.csv.What is the data format? Is it timestamp,x,y,z? And I saw that you gave the iPhone IMU (approximate) additive biases,could you please provide the specific values of white noise and random walk of IMU?

I'm confused about the world coordinate system of the ground truth data. Also, the ARKit support 3 types of coordinate settings (https://developer.apple.com/documentation/arkit/arconfiguration/worldalignment), which one are ADVIO using? Could you please add some detailed explanation?

Hello
In the github readme it says every sensor synchronized to the same system but I check the tango grayscale images and iphone images are quite different around same time. Could you also provide the time offset?
Best

I would like first to thank you for your effort in collecting the data and making it public for others to use.
I read in a different issue that the external calibration will be also released once verified, is there any time estimate on that ? or any updates in general ?
Thank you again !

I have read your creative work ADVIO: An Authentic Dataset for Visual-Inertial Odometry , which is really a great solution for collecting the dataset of VIO problem in the big crowd scene.
I also read the paper Inertial Odometry on Handheld Smartphones which is an implementation for providing the groundtruth.
The problem is how I can evaluate your groundtruth? What accuracy is your groundtruth?
For example, the accuracy of PennCOSYVIO is about 15cm, the accuracy of KITTI is about 10cm, and the accuracy of EuRoC is about 1mm.
Since the measurement update of position fixes is sparse, the trajectory(poses) between two position fixes is propagated by the integration of IMU.
Accounting for uncertainty and inaccuracy in the fixation point locations is taken into account by not enforcing the phone track to match the points, but including a Gaussian measurement noise term with a standard deviation of 25 cm in the position fixes (in all directions), does it mean the accuracy of your groundtruth is about 25cm?

That is a good job, but I cannot find the camerai nternal parameters and camera-inerial external parameters. Could you please provide the sensor calibrated parameter

Is the code for INS paper available online to experiment?

Is it possible to opensource the ARKIT iphone app so that others can collect their own data?
This would be highly beneficial to the community. Hopefully this is feasible. Please do let me know.
Thanks!

Hi, Thanks for your excellent work!
In your paper, "Our sensor data is collected using a standard iPhone 6s device and contains the ground-truth pose trajectory and the raw synchronized data streams from the following sensors: RGB video camera, accelerometer, gyroscope, magnetometer, platform-provided geographic coordinates, and barometer."
But acturally, we find that the data stream from accelerometer and gyroscope are not synchronized. How to deal with it?
How can I run ROVIO on your dataset directly ?

Hi, I want to make a tightly coupled system with GPS
But it has no timestamp message,
how can I assign a GPS timestamp message for Coordinate.latitude and Coordinate.longitude

I saw the description that the floor plan layouts are in the Fix points of Ground-truth folder. However, I don't see any images in the Ground-truth folder.

So each row in pose.csv, (e.g., advio-01/ground-truth/pose.csv) is

time[sec], tx, ty, tz, qw, qx, qy, qz

and the resulting transform can be computed by

T_XY = [R(qw, qx, qy, qz),  [tx, ty, tz]^T; 
              0, 0, 0,           1].

By T_XY, I mean, for a point expressed in coordinate frame Y, p_Y, its coordinate in coordinate frame X, p_X can be computed by

p_X = T_XY * p_Y.

I wonder how are the coordinate frames X and Y defined?
For ease of description, can you please explain the definitions of coordinate frames X and Y by relating to the below figure?

The figure shows several coordinate frames, the camera frame(C), and the IMU frame(B), and a global frame(G) with z axis pointing to negative gravity.
I would deeply appreciate if you could describe say Y like

x axis of Y = x axis of C, y axis of Y = - z axis of C, z axis of Y = y axis of C, origin of Y = optical center of the camera on the iphone 6s.

like 15ms or 30ms 's rolling shutter time ?

Hello,

Thank you for creating this challenging dataset! It has been very useful.

I couldn't find the sensor coordinate frame conventions explicitly mentioned anywhere, but this is what I could infer from the visualization scripts:

The ground-truth, ARCore, ARKit, and Tango poses are all given in a global IMU frame whose y-axis points upwards, z-axis towards the viewer, and x-axis towards the viewer's right. Is the y-axis always aligned with the gravity?

The camera to IMU transform (T_imu_cam) given in the extrinsics file is the matrix that transforms points in the camera's frame to the IMU's frame. This is the inverse of what Kalibr provides i.e. T_cam_imu.

Can someone please confirm this?

I want to ask about arkit.csv.
What is the dataformat?
is it xyz, qx,qy,qz,qw or is it xyz, qw,qx,qy,qz?

First of all, thanks to your contribution.I found that the intrinsics data in the README.md and iphone.yaml files are different. Which data should I choose ? Second, I have unified the time stamps of the accelerometer and the gyroscope , but can't run out of the trajectory which always have problems with inertial and visual align, I want to know which column the x, y, and z axes in the acc file and the gyr file belong to. Finally, I saw that you gave the iPhone IMU (approximate) additive biases, but can you give the specific values of white noise and random walk of IMU? Thank you very much.