projet-compression-streamin.../main.py

701 lines
23 KiB
Python
Raw Normal View History

2022-10-03 11:31:38 +00:00
import io
2022-10-17 13:51:37 +00:00
from math import floor
2022-10-03 11:31:38 +00:00
import obja.obja as obja
import numpy as np
import argparse
2022-10-17 13:51:37 +00:00
from time import time
from rich.progress import track, Progress
2022-10-03 11:31:38 +00:00
2022-10-10 10:26:30 +00:00
def cot(x: float):
sin_x = np.sin(x)
if sin_x == 0:
return 1e16
return np.cos(x) / sin_x
def sliding_window(l: list, n: int = 2):
k = n - 1
l2 = l + [l[i] for i in range(k)]
res = [(x for x in l2[i:i+n]) for i in range(len(l2)-k)]
return res
2022-10-03 11:31:38 +00:00
2022-10-17 13:51:37 +00:00
class Edge:
def __init__(self, a, b):
self.a = min(a, b)
self.b = max(a, b)
self.face1 = None
self.face2 = None
self.fold = 0.0
self.curvature = 0.0
def __eq__(self, __o: object) -> bool:
return self.a == __o.a and self.b == __o.b
class Face:
def __init__(self, a, b, c):
self.a = a
self.b = b
self.c = c
self.normal = np.zeros(3)
def to_obja(self):
return obja.Face(self.a, self.b, self.c)
def __eq__(self, __o: object) -> bool:
if __o is None:
return False
return self.a == __o.a and self.b == __o.b and self.c == __o.c
class Vertex:
def __init__(self, pos):
self.pos = pos
self.vertex_ring = []
self.face_ring = []
self.normal = np.zeros(3)
self.area = 0.0
self.curvature = 0.0
def to_obja(self):
return self.pos
2022-10-03 11:31:38 +00:00
class MAPS(obja.Model):
"""_summary_
Args:
obja (_type_): _description_
"""
def __init__(self):
super().__init__()
2022-10-17 13:51:37 +00:00
def parse_file(self, path):
super().parse_file(path)
for i, vertex in enumerate(self.vertices):
self.vertices[i] = Vertex(vertex)
for i, face in enumerate(self.faces):
self.faces[i] = Face(face.a, face.b, face.c)
def update(self):
self.update_edges()
self.update_rings()
self.update_normals()
self.update_area_curvature()
def update_edges(self):
self.edges = {}
remaining_faces = self.faces.copy()
while None in remaining_faces:
remaining_faces.remove(None)
for face in track(self.faces, description='Update edges'):
if face is None:
continue
for a, b in sliding_window([face.a, face.b, face.c], n=2):
new_edge = Edge(a, b)
if self.edges.get(f"{new_edge.a}:{new_edge.b}") is None:
new_edge.face1 = face
if face in remaining_faces:
remaining_faces.remove(face)
for face2 in remaining_faces:
face2_vertices = (face2.a, face2.b, face2.c)
if not (a in face2_vertices and b in face2_vertices):
continue
new_edge.face2 = face2
break
if new_edge.face2 is None:
print('ooooooooooooooooooooooo')
self.edges[f"{new_edge.a}:{new_edge.b}"] = new_edge
def update_rings(self):
for i, vertex in enumerate(self.vertices):
if vertex is None:
continue
vertex_ring, face_ring = self.one_ring(i)
vertex.vertex_ring = vertex_ring
vertex.face_ring = face_ring
def update_area_curvature(self):
for i, vertex in enumerate(self.vertices):
if vertex is None:
continue
area, curvature = self.compute_area_curvature(i)
vertex.area = area
vertex.curvature = curvature
self.feature_edges = []
for edge in self.edges.values():
edge.fold = np.dot(edge.face1.normal, edge.face2.normal)
if edge.fold < 0.5:
self.feature_edges.append(edge)
def update_normals(self):
for face in self.faces:
if face is None:
continue
p1 = self.vertices[face.a].pos
p2 = self.vertices[face.b].pos
p3 = self.vertices[face.c].pos
u = p2 - p1
v = p3 - p1
n = np.cross(u, v)
n /= np.linalg.norm(n)
face.normal = n
self.vertices[face.a].normal += n
self.vertices[face.b].normal += n
self.vertices[face.c].normal += n
for vertex in self.vertices:
if vertex is None:
continue
norm = np.linalg.norm(vertex.normal)
if norm != 0:
vertex.normal /= norm
2022-10-03 11:31:38 +00:00
def one_ring(self, index: int) -> tuple[list[int], list[int]]:
""" Return the corresponding 1-ring
Args:
index (int): index of the 1-ring's main vertex
Returns:
list[int]: ordered list of the 1-ring vertices
"""
2022-10-17 13:51:37 +00:00
if self.vertices[index] is None:
return None, None
2022-10-03 11:31:38 +00:00
# Find the 1-ring faces
ring_faces, ring_face_indices = [], []
for face_index, face in enumerate(self.faces):
2022-10-17 13:51:37 +00:00
if face is None:
2022-10-03 11:31:38 +00:00
continue
if index in (face.a, face.b, face.c):
ring_faces.append(face)
ring_face_indices.append(face_index)
# Initialize the ring
2022-10-17 13:51:37 +00:00
start_index = (ring_faces[0].a if ring_faces[0].a != index and ring_faces[0].c != index else
ring_faces[0].b if ring_faces[0].a != index and ring_faces[0].b != index else
ring_faces[0].c)
2022-10-03 11:31:38 +00:00
ring = [start_index]
ring_faces.pop(0)
# Select the indexes of the ring in the right order
while len(ring_faces) > 0:
broke = False
prev_index = ring[-1]
for i, face in enumerate(ring_faces):
if prev_index in (face.a, face.b, face.c):
# Found the face that correspond to the next vertex
current_index = ( # select the next vertex from the face
face.a if face.a != index and face.a != prev_index else
face.b if face.b != index and face.b != prev_index else
face.c
)
ring.append(current_index)
ring_faces.pop(i)
broke = True
break
if not broke:
raise ValueError(
f"Vertex {prev_index} is not in the remaining faces {ring_faces}. Origin {ring} on {index}")
return ring, ring_face_indices
2022-10-17 13:51:37 +00:00
def compute_area_curvature(self, index: int) -> tuple[float, float]:
2022-10-03 11:31:38 +00:00
""" Compute area and curvature the corresponding 1-ring
Args:
index (int): index of the 1-ring's main vertex
Returns:
tuple[float, float]: area and curvature
"""
2022-10-17 13:51:37 +00:00
if self.vertices[index] is None:
return None, None
ring = self.vertices[index].vertex_ring
p1 = self.vertices[index].pos
n1 = self.vertices[index].normal
2022-10-03 11:31:38 +00:00
area_sum = 0
2022-10-17 13:51:37 +00:00
curvature = 0
for index1, index2 in sliding_window(ring, n=2):
# the second vertice of the triangle
p2 = self.vertices[index1].pos
p3 = self.vertices[index2].pos # the third vertice of the triangle
n2 = self.vertices[index1].normal
2022-10-03 11:31:38 +00:00
M = np.array([ # build the matrix, used to compute the area
[p1[0], p2[0], p3[0]],
[p1[1], p2[1], p3[1]],
[p1[2], p2[2], p3[2]],
])
area = abs(np.linalg.det(M) / 2) # compute the area
area_sum += area
2022-10-17 13:51:37 +00:00
edge_curvature = np.dot(n2 - n1, p2 - p1) / \
np.linalg.norm(p2 - p1)**2
edge_curvature = abs(edge_curvature)
edge_key = f"{min(index, index1)}:{max(index, index1)}"
self.edges[edge_key].curvature = edge_curvature
2022-10-10 10:26:30 +00:00
2022-10-17 13:51:37 +00:00
curvature += edge_curvature
2022-10-03 11:31:38 +00:00
2022-10-17 13:51:37 +00:00
curvature /= len(ring)
2022-10-03 11:31:38 +00:00
2022-10-17 13:51:37 +00:00
return area_sum, curvature
2022-10-10 10:26:30 +00:00
def compute_priority(self, lamb: float = 0.0, max_length: int = 12) -> list[float]:
""" Compute selection priority of vertices (0.0 -> hight priority ; 1.0 -> low priority)
2022-10-03 11:31:38 +00:00
Args:
lamb (float, optional): convex combination factor. Defaults to 0.5.
max_length (int, optional): 1-ring maximum length to be prioritary. Defaults to 12.
Returns:
list[float]: priority values
"""
2022-10-17 13:51:37 +00:00
max_area = max(
[vertex.area for vertex in self.vertices if vertex is not None])
max_curvature = max(
[vertex.curvature for vertex in self.vertices if vertex is not None])
2022-10-03 11:31:38 +00:00
# Compute priorities
priorities = []
2022-10-17 13:51:37 +00:00
for vertex in self.vertices:
if vertex is not None and len(vertex.vertex_ring) < max_length:
2022-10-03 11:31:38 +00:00
# Compute priority
2022-10-17 13:51:37 +00:00
priority = (
lamb * vertex.area / max_area +
(1.0 - lamb) * vertex.curvature / max_curvature
)
2022-10-03 11:31:38 +00:00
else:
# Vertex with low priority
priority = 2.0
priorities.append(priority)
return priorities
def select_vertices(self) -> list[int]:
""" Select vertices for the current level reduction
Returns:
list[int]: selected vertices
"""
# Order vertices by priority
priorities = self.compute_priority()
vertices = [i[0]
for i in sorted(enumerate(priorities), key=lambda p: p[1])]
selected_vertices = []
2022-10-17 13:51:37 +00:00
with Progress() as progress:
task = progress.add_task('Select vertices', total=len(vertices))
while not progress.finished:
# Select prefered vertex
vertex = vertices.pop(0) # remove it from remaining vertices
progress.advance(task)
if priorities[vertex] == 2.0:
continue
incident_count = 0
for feature_edge in self.feature_edges:
if vertex in (feature_edge.a, feature_edge.b):
incident_count += 1
if incident_count > 2:
2022-10-03 11:31:38 +00:00
continue
2022-10-17 13:51:37 +00:00
selected_vertices.append(vertex)
# Remove neighbors
# for face in remaining_faces:
for face in self.faces:
if face is None:
continue
face_vertices = (face.a, face.b, face.c)
if vertex in face_vertices:
2022-10-03 11:31:38 +00:00
2022-10-17 13:51:37 +00:00
# Remove face and face's vertices form remainings
# remaining_faces.remove(face)
for face_vertex in face_vertices:
if face_vertex in vertices:
vertices.remove(face_vertex)
progress.advance(task)
2022-10-03 11:31:38 +00:00
2022-10-17 13:51:37 +00:00
return selected_vertices[:floor(1.0 * len(selected_vertices))]
2022-10-03 11:31:38 +00:00
def project_polar(self, index: int) -> list[np.ndarray]:
""" Flatten the 1-ring to retriangulate
Args:
index (int): main vertex of the 1-ring
Returns:
list[np.ndarray]: list the cartesian coordinates of the flattened 1-ring projected in the plane
"""
2022-10-17 13:51:37 +00:00
ring = self.vertices[index].vertex_ring
2022-10-03 11:31:38 +00:00
radius, angles = [], []
teta = 0.0 # cumulated angles
2022-10-10 10:26:30 +00:00
for index1, index2 in sliding_window(ring):
2022-10-17 13:51:37 +00:00
r = np.linalg.norm(
self.vertices[index].pos - self.vertices[index1].pos)
2022-10-03 11:31:38 +00:00
teta += self.compute_angle(index1, index, index2) # add new angle
radius.append(r)
angles.append(teta)
angles = [2 * np.pi * a / teta for a in angles] # normalize angles
coordinates = [np.array([r * np.cos(a), r * np.sin(a)])
for r, a in zip(radius, angles)] # parse polar to cartesian
return coordinates, ring
def compute_angle(self, i: int, j: int, k: int) -> float:
""" Calculate the angle defined by three points
Args:
i (int): previous index
j (int): central index
k (int): next index
Returns:
float: angle defined by the three points
"""
2022-10-17 13:51:37 +00:00
a = self.vertices[i].pos
b = self.vertices[j].pos
c = self.vertices[k].pos
2022-10-03 11:31:38 +00:00
u = a - b
v = c - b
u /= np.linalg.norm(u)
v /= np.linalg.norm(v)
res = np.dot(u, v)
return np.arccos(np.clip(res, -1, 1))
def clip_ear(self, index: int) -> tuple[list[obja.Face], int]:
""" Retriangulate a polygon using the ear clipping algorithm
Args:
index (int): index of 1-ring
Returns:
tuple[list[obja.Face], int]: list the triangles
"""
2022-10-17 13:51:37 +00:00
polygon_, ring_ = self.project_polar(index)
main_v = []
for i, r in enumerate(ring_):
for feature_edge in self.feature_edges:
feat_edge_vertices = (feature_edge.a, feature_edge.b)
if r in feat_edge_vertices and index in feat_edge_vertices:
main_v.append(i)
if len(main_v) < 2:
polygons_rings = [(polygon_, ring_)]
else:
v1 = ring_[main_v[0]]
v2 = ring_[main_v[1]]
ring1, ring2 = [], []
polygon1, polygon2, = [], []
start = ring_.index(v1)
while ring_[start] != v2:
ring1.append(ring_[start])
polygon1.append(polygon_[start])
start += 1
start %= len(ring_)
ring1.append(ring_[start])
polygon1.append(polygon_[start])
start = ring_.index(v2)
while ring_[start] != v1:
ring2.append(ring_[start])
polygon2.append(polygon_[start])
start += 1
start %= len(ring_)
ring2.append(ring_[start])
polygon2.append(polygon_[start])
polygons_rings = [(polygon1, ring1), (polygon2, ring2)]
2022-10-03 11:31:38 +00:00
faces = [] # the final list of faces
2022-10-17 13:51:37 +00:00
for polygon, ring in polygons_rings:
indices = [(local_i, global_i)
for local_i, global_i in enumerate(ring)] # remainging vertices
node_index = 0
cycle_counter = 0
while len(indices) > 2:
# Extract indices
local_i, global_i = indices[node_index - 1]
local_j, global_j = indices[node_index]
local_k, global_k = indices[node_index + 1]
# Extract verticies
prev_vert = polygon[local_i]
curr_vert = polygon[local_j]
next_vert = polygon[local_k]
is_convex = MAPS.is_convex(prev_vert, curr_vert, next_vert)
is_ear = True
if is_convex or cycle_counter > len(indices): # the triangle needs to be convext to be an ear
# Begin with the point next to the triangle
test_node_index = (node_index + 2) % len(indices)
while indices[test_node_index][0] != local_i and is_ear:
test_vert = polygon[indices[test_node_index][0]]
is_ear = not MAPS.is_inside(prev_vert,
curr_vert,
next_vert,
test_vert)
test_node_index = (test_node_index + 1) % len(indices)
else:
is_ear = False
cycle_counter += 1
if is_ear:
faces.append(Face(global_i, global_j, global_k))
indices.pop(node_index) # remove the point from the ring
cycle_counter = 0
node_index = (node_index + 2) % len(indices) - 1
2022-10-03 11:31:38 +00:00
return faces
def is_convex(prev_vert: np.ndarray, curr_vert: np.ndarray, next_vert: np.ndarray) -> bool:
""" Check if the angle less than pi
Args:
prev_vert (np.ndarray): first point
curr_vert (np.ndarray): middle point
next_vert (np.ndarray): last point
Returns:
bool: angle smaller than pi
"""
a = prev_vert - curr_vert
b = next_vert - curr_vert
dot = a[0] * b[0] + a[1] * b[1]
det = a[0] * b[1] - a[1] * b[0]
angle = np.arctan2(det, dot)
if angle < 0.0:
angle = 2.0 * np.pi + angle
internal_angle = angle
return internal_angle >= np.pi
def is_inside(a: np.ndarray, b: np.ndarray, c: np.ndarray, p: np.ndarray) -> bool:
""" Check if p is in the triangle a b c
Args:
a (np.ndarray): point one
b (np.ndarray): point two
c (np.ndarray): point three
p (np.ndarray): point to check
Returns:
bool: if the point to check is in a b c
"""
# Compute vectors
v0 = c - a
v1 = b - a
v2 = p - a
# Compute dot products
dot00 = np.dot(v0, v0)
dot01 = np.dot(v0, v1)
dot02 = np.dot(v0, v2)
dot11 = np.dot(v1, v1)
dot12 = np.dot(v1, v2)
# Compute barycentric coordinates
denom = dot00 * dot11 - dot01 * dot01
if abs(denom) < 1e-20:
return True
invDenom = 1.0 / denom
u = (dot11 * dot02 - dot01 * dot12) * invDenom
v = (dot00 * dot12 - dot01 * dot02) * invDenom
# Check if point is in triangle
return (u >= 0) and (v >= 0) and (u + v < 1)
2022-10-17 13:51:37 +00:00
2022-10-10 10:26:30 +00:00
def truc(self, output):
2022-10-17 13:51:37 +00:00
self.update()
2022-10-10 10:26:30 +00:00
priorities = self.compute_priority()
2022-10-17 13:51:37 +00:00
# min_p = min(priorities)
# priorities = [p - min_p for p in priorities]
# max_p = max(priorities)
colors = [priorities[face.a] + priorities[face.b] +
priorities[face.c] if face is not None else 0.0 for face in self.faces]
min_c = min(colors)
colors = [c - min_c for c in colors]
max_c = max(colors)
2022-10-10 10:26:30 +00:00
operations = []
for i, face in enumerate(self.faces):
if face != None:
2022-10-17 13:51:37 +00:00
r, g, b = colors[i] / max_c, 1.0, 1.0
c = 0
for x in (face.a, face.b, face.c):
for feature_edge in self.feature_edges:
if x in feature_edge:
c += 1
break
if c > 1:
r, g, b = 1.0, 0.0, 0.0
operations.append(('fc', i, (r, g, b), 0, 0, 0))
2022-10-10 10:26:30 +00:00
operations.append(('af', i, face, 0, 0, 0))
for i, vertex in enumerate(self.vertices):
2022-10-17 13:51:37 +00:00
if vertex is None:
# r, g, b = priorities[i] / max_p , 1.0, 1.0
operations.append(('av', i, vertex, 1.0, 1.0, 1.0))
2022-10-10 10:26:30 +00:00
operations.reverse()
# Write the result in output file
output_model = obja.Output(output)
for (op, index, value, r, g, b) in operations:
if op == 'av':
output_model.add_vertex_rgb(index, value, r, g, b)
elif op == 'af':
output_model.add_face(index, value)
elif op == 'ev':
output_model.edit_vertex(index, value)
elif op == 'ef':
output_model.edit_face(index, value)
elif op == 'fc':
print('fc {} {} {} {}'.format(
len(output_model.face_mapping),
value[0],
value[1],
value[2]),
file=output
)
2022-10-17 13:51:37 +00:00
def compress(self, output: io.TextIOWrapper, final_only: bool) -> None:
2022-10-03 11:31:38 +00:00
""" Compress the 3d model
Args:
output (io.TextIOWrapper): Output file descriptor
"""
operations = []
# while len(self.vertices) > 64:
2022-10-17 13:51:37 +00:00
for _ in range(2):
self.update()
2022-10-03 11:31:38 +00:00
selected_vertices = self.select_vertices() # find the set of vertices to remove
2022-10-17 13:51:37 +00:00
for v_index in track(selected_vertices, description="Compression"):
2022-10-03 11:31:38 +00:00
# Extract ring faces
2022-10-17 13:51:37 +00:00
ring_faces = self.vertices[v_index].face_ring
2022-10-03 11:31:38 +00:00
# Apply retriangulation algorithm
2022-10-17 13:51:37 +00:00
faces = self.clip_ear(v_index)
2022-10-03 11:31:38 +00:00
# Edit the first faces
for i in range(len(faces)):
2022-10-17 13:51:37 +00:00
if not final_only:
operations.append(
('ef', ring_faces[i], self.faces[ring_faces[i]].to_obja()))
2022-10-03 11:31:38 +00:00
self.faces[ring_faces[i]] = faces[i]
# Remove the last faces
for i in range(len(faces), len(ring_faces)):
2022-10-17 13:51:37 +00:00
if not final_only:
operations.append(
('af', ring_faces[i], self.faces[ring_faces[i]].to_obja()))
2022-10-03 11:31:38 +00:00
self.faces[ring_faces[i]] = None
# Remove the vertex
2022-10-17 13:51:37 +00:00
if not final_only:
operations.append(
('av', v_index, self.vertices[v_index].to_obja()))
self.vertices[v_index] = None
2022-10-03 11:31:38 +00:00
2022-10-03 11:36:23 +00:00
# Register remaining vertices and faces
2022-10-03 11:31:38 +00:00
for i, face in enumerate(self.faces):
2022-10-17 13:51:37 +00:00
if face is not None:
operations.append(('af', i, face.to_obja()))
for i, v_index in enumerate(self.vertices):
if v_index is not None:
operations.append(('av', i, v_index.to_obja()))
2022-10-03 11:31:38 +00:00
# To rebuild the model, run operations in reverse order
operations.reverse()
# Write the result in output file
output_model = obja.Output(output)
for (op, index, value) in operations:
if op == 'av':
output_model.add_vertex(index, value)
elif op == 'af':
output_model.add_face(index, value)
elif op == 'ev':
output_model.edit_vertex(index, value)
elif op == 'ef':
output_model.edit_face(index, value)
2022-10-10 10:26:30 +00:00
elif op == 'fc':
print('fc {} {} {} {}'.format(
index,
value[0],
value[1],
value[2]),
file=output
)
2022-10-03 11:31:38 +00:00
def main(args):
""" Run MAPS model compression
Args:
args (Namespace): arguments (input and output path)
"""
model = MAPS()
model.parse_file(args.input)
with open(args.output, 'w') as output:
2022-10-17 13:51:37 +00:00
model.compress(output, args.final)
# with open(args.output, 'w') as output:
# model.truc(output)
2022-10-03 11:31:38 +00:00
if __name__ == '__main__':
parser = argparse.ArgumentParser()
parser.add_argument('-i', '--input', type=str, required=True)
parser.add_argument('-o', '--output', type=str, required=True)
2022-10-17 13:51:37 +00:00
parser.add_argument('-f', '--final', type=bool, default=False)
2022-10-03 11:31:38 +00:00
args = parser.parse_args()
main(args)