Does this result normal?

[lucasjinreal]

You said yellow is the 3d box location, but I just can not find it anyway:

There is not car. I think the model predict something wrong.

beside there are some error in codes. One of them is that input placeholder are (1, 1024, 4), but input of a simple image are (9, 1024, 4) why?

[KleinYuan]

@jinfagang Could you elaborate more so that I can understand the issue more clearly? Like how you reproduce this? Which variable are which not expected state?

[lucasjinreal]

@KleinYuan Thanks for your reply. This is the inference result.

Just download the model and feed the image with point clouds. After fix some errors and logic error, got this prediction.

I am not sure the yellow and red indicates, but the input is white, I suppose it's to predict the car but prediction are offset. Besizes the real data input lidar shape are (9, 1024, 4) from kitti, but input required (1, 1024, 4), I supports it's just an angle that 1/9 of 360 and only a section ?

[aditbhrgv]

Hi,

I also face the similar issue. It seems 3D bounding box (in yellow) is offset by some distance. I used testing/00001.png and corresponding calib & 0001.bin velodyne data. Below is my output:

[tianchengdw]

Hi,

I also face the similar issue. It seems 3D bounding box (in yellow) is offset by some distance. I used testing/00001.png and corresponding calib & 0001.bin velodyne data. Below is my output:

Hey, I also have the problem just as you had. Have you already solved the problem? I used the 00000.png.

[lucasjinreal]

@tianchengdw I didn't digger deep in this question, even the result is good the speed is still very slow.... we'd better using some more faster methods for 3d object perception. We have a community to talk about this: strangeai talk