DeepLearningOnGraph/decimate_labels.py at master · Conchylicultor/DeepLearningOnGraph · GitHub

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
"""
Match the labels (body parts) of each original mesh to its equivalent in decimated mesh
Use 1-NN to match each face of the mesh
By inverting dirLabels and dirLabelsDecimated, it should be possible to do the inverse
mapping

Just indicate the parameters and target directories.

Use python 3 (but should be working with python 2)
"""

import os, sys
import numpy as np
import utils

# Set directories
root = os.getcwd()
dirMesh            = root + '/../MeshsegBenchmark-1.0/data/off/'
dirMeshDecimated   = root + '/../MeshsegBenchmark-1.0/data/off_decimated/' # Should contain the sames number and filename as above
dirLabels          = root + '/../MeshsegBenchmark-1.0/data/seg/Bench/'
dirLabelsDecimated = root + '/../MeshsegBenchmark-1.0/data/seg/Bench_decimated/'


def main():

    # Global check
    assert(os.path.exists(dirMesh))
    assert(os.path.exists(dirMeshDecimated))
    assert(os.path.exists(dirLabels))
    assert(os.path.exists(dirLabelsDecimated))

    # For each mesh
    filesList = utils.sortFiles(os.listdir(dirMesh))
    for filename in filesList:
        if filename.endswith('.off'): # Candidate
            print('Try matching ', filename)
            idMesh = filename.split('.')[0]

            # Loading the original and reduced meshes and compute the center of each face
            vertices,          faces          = utils.extractMesh(dirMesh          + filename)
            verticesDecimated, facesDecimated = utils.extractMesh(dirMeshDecimated + filename)

            pointCloud          = utils.meshToPointCloud(vertices,          faces)
            pointCloudDecimated = utils.meshToPointCloud(verticesDecimated, facesDecimated)

            # Extract the label list
            labelList = utils.loadLabelList(dirLabels + idMesh + '.seg')
            labelListDecimated = []

            # Use K-NN on each face of the decimated mesh
            for pointDecimated in pointCloudDecimated:
                # Search the closest point index in the original point cloud
                minIndex = 0
                minDist = -1 # Initialize
                for i, point in enumerate(pointCloud):
                    distance = np.linalg.norm(pointDecimated - point)
                    if distance < minDist or minDist == -1:
                        minDist = distance
                        minIndex = i

                labelListDecimated.append(labelList[minIndex])
                if len(labelListDecimated) % 100 == 0:
                    print(len(labelListDecimated) / len(pointCloudDecimated) * 100, '%')

            # Save the values of the labels of the closest match indexes
            saveName = dirLabelsDecimated + idMesh + '.seg'
            print('Saving ', saveName)
            utils.saveLabelList(labelListDecimated, saveName)


if __name__ == "__main__":
    main()