jonasschult / dcm-net Goto Github PK

in graph_level_generation.py--->quadric_error_metric--->csv2npy:
FileNotFoundError: [Errno 2] No such file or directory: 'Data/scannet_qem_train_rooms/scannet/curr_mesh.csv'
I debugged the code and couldn't find the.csv file in the output directory，only a curr_mesh.ply file.
`def quadric_error_metric(curr_file_path, ratio, old_vertices):
os.system(f"tridecimator "
f"{curr_file_path} {curr_file_path} {ratio} -On -C > /dev/null")

mesh = open3d.read_triangle_mesh(curr_file_path)
if not mesh.has_vertices():
    raise QEMError('no vertices left')
coords = np.asarray(mesh.vertices)
edges_list = edges_from_faces(np.asarray(mesh.triangles))
edge_out = []
for key, group in enumerate(edges_list):
    for elem in group:
        edge_out.append([key, elem])

reverse_trace = csv2npy(curr_file_path.replace(
    '.ply', '.csv'), old_vertices=old_vertices, new_vertices=coords)

return coords, edge_out, reverse_trace`

Hi,

Thank you so much for creating this great work. I have one question regarding the vcglib that you shared. It seems to me that there are some undeclared variables in tridecimator.cpp. Would you mind fixing this issue for us?

Best,
Xingtong

hi, I was wondering if you have some expertiments on this dataset?

Hi Jonas,

Thank you for the amazing work and the code!

I followed the installation step by step, everything went good, until when I ran the "make" for QUADRIC ERROR METRICS. I got the errors below. Seems like something wrong in the cpp file. Could you please have a look at that?

And to clearify, by "Add vcglib/apps/tridecimator and vcglib/apps/sample/trimesh_clustering to your environment path variable!", is this the right thing to do or not?

PATH=$PATH:vcglib/apps/tridecimator
PATH=$PATH:vcglib/apps/sample/trimesh_clustering

Appreciate your attention!

tridecimator.cpp: In function ‘int main(int, char**)’:
tridecimator.cpp:165:147: error: ‘dup_face’ was not declared in this scope
face %i and edge %i and deg vertex %i and deg face %i\n",dup,unref, dup_face, dup_edge, deg_vertex, deg_face);
^~~~~~~~
tridecimator.cpp:165:157: error: ‘dup_edge’ was not declared in this scope
d edge %i and deg vertex %i and deg face %i\n",dup,unref, dup_face, dup_edge, deg_vertex, deg_face);
^~~~~~~~
tridecimator.cpp:165:167: error: ‘deg_vertex’ was not declared in this scope
and deg vertex %i and deg face %i\n",dup,unref, dup_face, dup_edge, deg_vertex, deg_face);
^~~~~~~~~~
tridecimator.cpp:165:179: error: ‘deg_face’ was not declared in this scope
tex %i and deg face %i\n",dup,unref, dup_face, dup_edge, deg_vertex, deg_face);
^~~~~~~~
tridecimator.cpp:171:32: error: ‘FinalSize’ was not declared in this scope
printf("reducing it to %i\n",FinalSize);
^~~~~~~~~

Hello.

I want to know the format of input.
Is it (BatchSize, NumOfPoints, 9)? (9 is xyzrgb + normal(xyz))

Thank you.

Hi,

It seems to me that the code only provides a way of generating geodesic neighbors in the base dataset class. Could you please kindly guide me through how I can get the spatial neighbors?

Hi, thank you for making your work available. I tried running the model using the scannet_benchmark.json configuration but I got this error

I checked torch_geometric.data documentation but I couldn't find GraphLevelDataLoader, is it in a different version of torch-geometric or in another library? Thank you

Thanks, @JonasSchult for sharing your work.

I'm trying to prepare a dataset for a regression task. I'm not sure how to handle crops as I think the following piece of code assigns labels to level 0 and is written for segmentation task?.

Any idea how can I assign regression labels to level 0.

Thanks

In utils/preprocess/crop_training_samples.py

if level == 0 and is_train:
                        pooled_levels = np.zeros(
                            (traces[0].max() + 1, labels.max() + 1), dtype=np.int)
                        for i in range(len(labels)):
                            pooled_levels[traces[0][i]][labels[i]] += 1

                        pooled_levels = np.argmax(pooled_levels, axis=1)
                        block_labels = pooled_levels[filters[-1]]

                    # map from old indices to new ones
                    block_edges[-1] = np.unique(block_edges[-1],
                                                return_inverse=True)[1].reshape(-1, 2)

Thanks for sharing your wonderful work.
Could you please help me to preprocess(prepare) other dataset for training, such as ShapeNet.

Thank you for sharing your amazing work!

I'd like to ask if you are planning to release the preprocessing code for S3DIS data set?

Best,
HU Zeyu

On the following line you apply random rotation to help teach your network to be invariant to absolute position. This is important because even though you formulated your input EdgeConv operator to be invariant to absolute position, it still contains absolute orientation. However you neglect to apply this random rotation to your normals as well. This not only means you're learning on absolute orientation but also that your normals and positions no longer match. I suspect fixing this issue could lead to noticeable improvements in your segmentation IoU scores.

https://github.com/VisualComputingInstitute/dcm-net/blob/e021c69d7a27fc985450b6afadef0a3327161a1c/transform/random_rotation.py#L18

I am more interested in your visualization in Figure 1. Could you share the visualization software you use?
@JonasSchult