VDIGPKU / RCBEVDet

[CVPR 2024] RCBEVDet: Radar-camera Fusion in Bird’s Eye View for 3D Object Detection
101 stars 5 forks source link

some confusions about code #5

Open xiaobh1519 opened 3 months ago

xiaobh1519 commented 3 months ago

image In the provided code, the heatmap tensor, which is suspected to correspond to fRCS from the original text, is initialized to zeros but does not seem to undergo further processing before being concatenated. Could you please explain for this?

Distrber commented 3 months ago

According to the functions provided by mmdet3d, the heatmap tensor has already been assigned values. Therefore, before concatenation, the heatmap tensor is not an empty tensor. Snipaste_2024-07-07_13-22-46

xiaobh1519 commented 3 months ago

According to the functions provided by mmdet3d, the heatmap tensor has already been assigned values. Therefore, before concatenation, the heatmap tensor is not an empty tensor. Snipaste_2024-07-07_13-22-46

But it seems that you didn't assign the return value of draw_gaussian_map to heatmap.

Distrber commented 3 months ago

According to the functions provided by mmdet3d, the heatmap tensor has already been assigned values. Therefore, before concatenation, the heatmap tensor is not an empty tensor. Snipaste_2024-07-07_13-22-46

But it seems that you didn't assign the return value of draw_gaussian_map to heatmap.

During your debugging process, is the heatmap always set to zero? In my opinion, the masked_heatmap in this function is a portion of the heatmap, so it directly modifies the values of the heatmap.

xiaobh1519 commented 3 months ago

According to the functions provided by mmdet3d, the heatmap tensor has already been assigned values. Therefore, before concatenation, the heatmap tensor is not an empty tensor. Snipaste_2024-07-07_13-22-46

But it seems that you didn't assign the return value of draw_gaussian_map to heatmap.

During your debugging process, is the heatmap always set to zero? In my opinion, the masked_heatmap in this function is a portion of the heatmap, so it directly modifies the values of the heatmap.

Thanks for your reply, I notice that when you project radar features, the equation is r = (c_x^2+ cy^2) \times V{RCS}, I would like to know why this "r" is relative to cx and cy?

VDIGPKU commented 2 months ago

(c_x^2+ c_y^2) means the distance of the radar point from the sensor. According to the definition of the RCS, it is inversely proportional to distance. So, we should multiply (c_x^2+ c_y^2) to calculate the true Radar reflection power.