Open lx1037 opened 3 months ago
I have the same problem. I found that the network code can only provide instance segmentation of the complete point cloud (.txt), and does not include the generation of mesh and the acquisition of edge mentioned in the paper (not conclude .json script).
I have the problem, too. Could you provide an example of the files ‘param_4147.txt’ and 'param_inter_lines_4147.json'? Thanks so much!
Hello, I have uploaded the fitting codes at 'Fitting_patches_and_edges'. Youcan use the primitive_forward_v2.py to calculate the patches and edges from output files from SED-Net, and then use arg2mesh.py to generate mesh.
When I use primitive_forward_v2.py, I noticed a splinenet model is used.
` def init(self, sample_ratio, closed_splinenet_path="splinenet/ckpt/train_closed_spline_1_700_0.9_bt_36_lr_0.001_trsz_28000_tsz_3000_wght_0.9.pth",
closed_splinenet_path="splinenet/ckpt/closed_spline.pth",
open_splinenet_path="splinenet/ckpt/open_spline.pth")`
Could you also provide these two pretrained model? Thanks!
You can find the pretrain spline-net here https://github.com/SimingYan/HPNet, in the 'log/pretrained_model' directory.
How can I obtain the input for arg2mesh.py, which are txt file and json file containing param information? Is there a specific program I should run to generate these files?
if __name__ == "__main__": arg2mesh("E:\\arg2mesh\\test\\output\\","E:\\arg2mesh\\test\\param_4147.txt","E:\\arg2mesh\\test\\param_inter_lines_4147.json")