search

CathIAS / TLIO

Tight Learned Inertial Odometry
Other
289 stars 70 forks source link

Issue with window size #25

Closed leom97 closed 2 years ago

leom97 commented 3 years ago

Hello, whenever the number of IMU measurements that the network sees is a multiple of 32, I get a layer mismatch error during the training process of the neural network.

For instance, if the IMU frequency is 200, the past size is .4s, the window size is 2s, the network sees 480 measurements, and a mismatch error is thrown.

Thank you in advance.

JinghangLi commented 3 years ago

Well, I don't know if this will help, I do see the number 32 in the network parameters section of the code here but I don't know exactly what it does either.

leom97 commented 3 years ago

Well, I don't know if this will help, I do see the number 32 in the network parameters section of the code here but I don't know exactly what it does either.

Hi! Exactly, that's how I figured out that 32 itself was a problem. I guess one should dig into the various dimensions of the ResNet...