Visualize bbox3d

[no-Seaweed]

Hi,

I am writing the code for visualizing both GT and predict 3d BBox on validation set images by multiplying the output 3d coordinates with the corresponding calibration matrix. Your output matches the GT, but both of those 2 projected bboxes are a bit off to the actual car. What the information I am missing? Here is the example I have.

Best, Shukai

[githubAccR]

Did you ever find out how to fix this?

[JianWu313]

@no-Seaweed can you send me a skript only for the prediction,I just wanna used the pretrained weights for inference,thanks in advanve

[Colin131]

in "rpn_util.py" function "test_kitti_3d" delete line:

y3d += h3d/2

then you will be fine

[Sondosmohamed1]

@no-Seaweed could you illustrate how did you visualize the result ? i want to reimplement the algorithm using camera data i retrained it but i do not know how to implement it using new camera data(for example png /jpg image ) .