Open JianWu313 opened 3 years ago
3D WBF only supports rectangular parallelepipeds as input, given as 2 opposite corners (x1, y1, z1) and (x2, y2, z2). If you can convert your 3D boxes in this format you are good.
@ZFTurbo thanks for your quick reply ,I still not so sure why we need to normalize all the coordinate
Actually algorithm will work without normalization (in this case you need to remove checks and sligtly remade IoU calculation). But anyway it will require rectangular parallelepipeds as input.
@ZFTurbo thanks a lot, I still have one more question about the parameter"weights". Is it used to change the orginial score in each models,for example,w1s1, w2s2, if we have many models ,can we just set all the weights to be 1 (I have found the weights in your paper but don't understand the usage)
Weights needed to say which model predictions are more important. By default all weights equal to 1.
Is there any implementation of yaw angle fusion in 3d boxes...?
Is there any implementation of yaw angle fusion in 3d boxes...?
Hello, did you find a solution? thanks @bringBackm
@ZFTurbo thanks a lot, I still have one more question about the parameter"weights". Is it used to change the orginial score in each models,for example,w1_s1, w2_s2, if we have many models ,can we just set all the weights to be 1 (I have found the weights in your paper but don't understand the usage)
Hello, did you find a solution? thanks @JianWu313
most of the 3d object-detector need to predict a heading angle(yaw angle),but is seems that the 3d wbf don't use it and another problem is the normalization,.by 2D case it is quite easy because we just need to divided by width and height, but 3D detection are the coordinate in real world with respect to camera coordinate(not just image plane)