zhanggang001 / hednet Goto Github PK

The difference between hednet_8x_1f_onestage_D1_1x.yaml and hednet_8x_1f_onestage_D1_2x.yaml

Great job with excellent performance. May I ask when the code can be made public? Thank you.

Hi, when predicting the final heatmap and object size, it looks like you are using normal convolution. Why not use sparse convolution here?

您好，SAFDNet代码公开了吗？？您这上面上传的和HDNet有啥区别吗

In the prediction head, the box encode in center(x,y), enter_z(z), dim(l,w,h), rot(sina,cosa). What iou representation?
@zhanggang001

Hi, does the 3D-EDB mentioned in your paper correspond to self.sed_layers in the figure below, and does the 2D-EDB correspond to self.afd_layers? Thanks!!

Great work and appreciate your willing to open source it!
Any plan on when to release your code?

@zhanggang001

I'm very interested in your adapt feature diffusion in SAFDNet and want to study how is it implemented to incorporate it in my work.Please release it soon.
Thanks!

I see that HEDNet gets 76.9/75.0 for mAP/mAPH in your paper. But the result is 75.66/73.77 on the leaderboard.

Thank you for releasing the code. Regarding safdnet_20e_nuscenes.yaml, I have a question about whether the file nuscenes_10sweeps_withvelo_lidar.npy on line 17 is correct. Because during the generation of NuScenes files, there is no nuscenes_10sweeps_withvelo_lidar.npy file generated.