luni/cvpr_2023
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

registration progress save

Browse Source
This commit is contained in:
luni
2023-06-25 23:16:48 +02:00
parent fec3d38a0a
commit 9a7aaefb8f
3 changed files with 1187 additions and 45 deletions
Download Patch File Download Diff File Expand all files Collapse all files
data/registration.py
+282 -28
View File
@@ -7,6 +7,9 @@ import torch
from torch.nn.functional import mse_loss
from pytorch3d.structures import Meshes
from pytorch3d.loss import mesh_laplacian_smoothing, mesh_edge_loss, mesh_normal_consistency, chamfer_distance
from pymeshfix import MeshFix
# os.environ["CUDA_LAUNCH_BLOCKING"] = '1'
sys.path.append(os.path.dirname(os.path.abspath(__file__)) + '/..')
from data.cloth3d_io import (
@@ -26,7 +29,40 @@ from data.cloth3d_io import (
from smpl.smpl_torch import SMPL
# from custom_utils.custom_utils import _maskFaces
smpl = SMPL('cuda', 'smpl')
device = 'cuda'
smpl = SMPL(device, 'smpl')
import torch.nn.functional as F
from pytorch3d.structures import Meshes
def find_closest_vertices(mesh1: Meshes, mesh2: Meshes, vertex_indices: torch.Tensor) -> torch.Tensor:
"""
Given two PyTorch3D meshes and a set of vertex indices for mesh1,
find the closest vertices from mesh2 to these vertices.
Args:
mesh1: A PyTorch3D Meshes object representing the first mesh.
mesh2: A PyTorch3D Meshes object representing the second mesh.
vertex_indices: A tensor of shape (N,) representing the indices of the vertices in mesh1.
Returns:
A tensor of shape (N,) representing the indices of the closest vertices in mesh2.
"""
# Extract the vertices of mesh1 and mesh2.
vertices1 = mesh1.verts_packed()
vertices2 = mesh2.verts_packed()
# Extract the vertices of mesh1 that correspond to the given indices.
vertices1_selected = vertices1[vertex_indices]
# Compute the pairwise distances between the selected vertices of mesh1 and all vertices of mesh2.
distances = F.pairwise_distance(vertices1_selected.unsqueeze(1), vertices2.unsqueeze(0), p=2)
# Find the indices of the closest vertices in mesh2.
closest_indices = torch.argmin(distances, dim=1)
return closest_indices
def _maskFaces(V, F, mask, return_unfixed=False):
@@ -90,6 +126,8 @@ def build_lookup_table(repeated):
from data.cloth3d_io import readPC2Frame
import scipy.io as sio
import scipy
import scipy.spatial as spatial
def local_read_posed_garment_frame(folder, id, garment, frame, pose=None, deorient=False):
pc16_path = os.path.join(folder, id, garment + '.pc16')
verts = readPC2Frame(pc16_path, frame, True)
@@ -97,8 +135,8 @@ def local_read_posed_garment_frame(folder, id, garment, frame, pose=None, deorie
map = sio.loadmat(os.path.join(folder, id, garment + '_map.mat'))['map'].reshape((-1))
if deorient:
R = scipy.spatial.transform.Rotation.from_euler('xyz', pose[:3]).inv()
# R = scipy.spatial.transform.Rotation.from_euler([0,0,-zrot])
R = spatial.transform.Rotation.from_euler('xyz', pose[:3]).inv()
# R = spatial.transform.Rotation.from_euler([0,0,-zrot])
verts = R.apply(verts)
return verts, map
@@ -108,77 +146,293 @@ def local_read_posed_body_frame(folder, id, frame, pose=None, deorient=False):
verts = readPC2Frame(pc16_path, frame, True)
if deorient:
R = scipy.spatial.transform.Rotation.from_euler('xyz', pose[:3]).inv()
# R = scipy.spatial.transform.Rotation.from_euler([0,0,-zrot])
R = spatial.transform.Rotation.from_euler('xyz', pose[:3]).inv()
# R = spatial.transform.Rotation.from_euler([0,0,-zrot])
verts = R.apply(verts)
return verts
def edge_loss_custom(x, y, edges):
x_e = x[:,edges[:,0],:] - x[:,edges[:,1],:]
x_e = torch.sum(x_e ** 2, -1)
x_e = torch.sqrt(x_e + 1e-9)
y_e = y[:,edges[:,0],:] - y[:,edges[:,1],:]
y_e = torch.sum(y_e ** 2, -1)
y_e = torch.sqrt(y_e + 1e-9)
d_e = y_e - x_e
loss = torch.mean(d_e ** 2, dim=-1)[0]
return loss
def crotch(mesh, body, crotch_idx, nr=75):
crotch_middle_on_body = crotch_idx
# create kdtree for body and mesh
body_tree = spatial.cKDTree(body)
mesh_tree = spatial.cKDTree(mesh)
# find closest point on mesh from crotch middle on body along the z axis
crotch_middle_on_mesh = find_closest_point_along_z(body[crotch_middle_on_body], mesh)
crotch_body = body_tree.query(body[crotch_middle_on_body], k=nr)[1]
crotch_mesh = mesh_tree.query(mesh[crotch_middle_on_mesh], k=nr)[1]
body_verts = body[crotch_body]
mesh_verts = mesh[crotch_mesh] - (mesh[crotch_middle_on_mesh] - body[crotch_middle_on_body])
mesh_verts = spatial.cKDTree(mesh_verts)
closest_mesh_verts = mesh_verts.query(body_verts, k=1)[1]
crotch_mesh = crotch_mesh[closest_mesh_verts]
return crotch_body, crotch_mesh
def find_closest_point_along_z(point, verts, eps=0.01):
min_dist = np.inf
closest_point = None
for i, v in enumerate(verts):
dist = np.linalg.norm(point - v)
if dist < min_dist and np.abs(point[0] - v[0]) < eps and np.abs(point[1] - v[1]) < eps:
min_dist = dist
closest_point = i
return closest_point
def mesh_edge_lengths(mesh: Meshes) -> torch.Tensor:
"""
Given a PyTorch3D Meshes object, compute the lengths of all edges in the mesh.
Args:
mesh: A PyTorch3D Meshes object representing the mesh.
Returns:
A tensor of shape (E,) representing the lengths of all edges in the mesh,
where E is the total number of edges in the mesh.
"""
# Extract the vertices and faces of the mesh.
vertices = mesh.verts_packed()
faces = mesh.faces_packed()
# Compute the pairwise distances between all pairs of vertices that form an edge.
edge_indices = faces.view(-1, 3)
edge_vertices = vertices[edge_indices]
edge_lengths = torch.norm(edge_vertices[:, 1:] - edge_vertices[:, :-1], dim=2)
return edge_lengths
def register(folder, frame):
id, frame_nr, garment = frame.split()
frame_nr = int(frame_nr)
info = loadInfo(os.path.join(folder, id, 'info.mat'))
raw_mask = read_garment_mask(folder, id, garment, return_raw=True)
# convert it to indices
mask = raw_mask > 0
mesh_posed, map = local_read_posed_garment_frame(folder, id, garment, frame_nr, pose=info['poses'][:,frame_nr], deorient=True)
body_posed = local_read_posed_body_frame(folder, id, frame_nr, pose=info['poses'][:,frame_nr], deorient=True)
template, faces, _, _ = readOBJ(os.path.join(folder, id, garment + '.obj'))
faces = quads2tris(faces)
pose = torch.zeros(72)
shape = torch.zeros(11)
shape[-1] = 1
# boundaries = raw_mask == 2
# # invert map and use it to invert boudaries
# map_inv = np.zeros(map.max()+1, dtype=np.int32)
# map_inv[map] = np.arange(len(map))
# boundaries = boundaries[map_inv]
# # get the boundary indices
# mesh1_boundaries = torch.tensor(np.where(boundaries)[0]).to(device)
# pose = torch.zeros(1, 72).to('cuda')
shape = torch.zeros(1, 11).to(device)
shape[:, :10] = torch.tensor(info['shape']).to(device)
shape[:, -1] = torch.tensor(info['gender']).to(device)
# for star pose
pose[:,5] = 3.1415/4
pose[:,8] = -3.1415/4
star_body = smpl(pose.unsqueeze(0), shape.unsqueeze(0))[0][0].cpu().numpy()
# pose[:, 5] = 3.1415/4
# pose[:,8] = -3.1415/4
STAR_POSE = np.zeros((24, 3), np.float32)
STAR_POSE[1,2] = .15
STAR_POSE[2,2] = -.15
pose = torch.tensor(STAR_POSE.flatten()[None,:]).to(device)
star_body = smpl(pose, shape, remove=True)[0][0].cpu().numpy() - smpl(pose, shape, remove=True)[4][0].cpu().numpy()
# rotate star body with 90 degrees around x axis
R = spatial.transform.Rotation.from_euler('x', np.pi/2)
star_body = R.apply(star_body).astype(np.float32)
is_skirt = (garment == 'Dress' or garment == 'Skirt')
skirt_mod = '_skirt' if is_skirt else ''
F = np.load(f'smpl/smpl_faces{skirt_mod}.npy').astype(np.int32)
plot(template, faces)
##################### mask body #####################
new_crotch = 4729
# star_body = star_body[mask]
# new_crotch = np.zeros((14475), bool)
# new_crotch[4729] = True
# new_crotch = new_crotch[mask]
# new_crotch = np.where(new_crotch)[0][0]
# _, F = _maskFaces(np.zeros((14475, 3)), F, mask)
# plot(template, faces)
# v = mesh_posed[mask,:]
##################### boundaries #####################
# meshfix = MeshFix(template, faces)
# holes = meshfix.extract_holes()
# new_border = []
# for border in np.array(holes.points):
# # find the index of the vertex in the original mesh
# ind = np.argwhere(np.all(template == border, axis=1))
# new_border.extend(ind.ravel())
# mesh1_boundaries = torch.tensor(new_border).to(device)
# np.save('temp/body_boundaries.npy', new_border)
mesh1_boundaries = torch.tensor(np.load('temp/mesh1_boundaries.npy')).to(device)
# meshfix = MeshFix(template, faces)
# holes = meshfix.extract_holes()
# new_border = []
# for border in np.array(holes.points):
# # find the index of the vertex in the original mesh
# ind = np.argwhere(np.all(template == border, axis=1))
# new_border.extend(ind.ravel())
# mesh1_boundaries = torch.tensor(new_border).to(device)
# np.save('temp/body_boundaries.npy', new_border)
# mesh1_boundaries = torch.tensor(np.load('temp/mesh1_boundaries.npy')).to(device)
##################### crotch #####################
crotch_body, crotch_mesh = crotch(template, star_body, crotch_idx=new_crotch, nr=50)
# export star_body as obj
# writeOBJ('temp/star_body.obj', star_body, F)
template = torch.tensor(template).to(device)
template_orig = template.clone()
faces = torch.tensor(faces).to(device)
star_body = torch.tensor(star_body).to(device)
F = torch.tensor(F).to(device)
# verts = template.cpu().numpy()
# star_body = star_body.cpu().numpy()
# faces = faces.detach().cpu().numpy()
# F = F.detach().cpu().numpy()
# fig = go.Figure(data=[
# go.Mesh3d(
# x=verts[:,0], y=verts[:,1], z=verts[:,2],
# i=faces[:,0], j=faces[:,1], k=faces[:,2],
# color='lightpink', opacity=0.50),
# go.Mesh3d(
# x=star_body[:,0], y=star_body[:,1], z=star_body[:,2],
# i=F[:,0], j=F[:,1], k=F[:,2],
# color='lightblue', opacity=0.50),
# go.Scatter3d(
# x=verts[crotch_mesh,0], y=verts[crotch_mesh,1], z=verts[crotch_mesh,2],
# mode='markers', marker=dict(size=2, color='red')),
# go.Scatter3d(
# x=star_body[crotch_body,0], y=star_body[crotch_body,1], z=star_body[crotch_body,2],
# mode='markers', marker=dict(size=2, color='blue')),
# ])
# # make it cuve and limit to -1 and -1
# fig.update_layout(scene = dict(
# xaxis = dict(range=[-1,1],),
# yaxis = dict(range=[-1,1],),
# zaxis = dict(range=[-1,1],),),)
# fig.show()
# Define the meshes to optimize
mesh1 = Meshes(verts=template[None,:,:], faces=faces[None,:,:])
mesh2 = Meshes(verts=star_body, faces=F[None,:,:])
mesh2 = Meshes(verts=star_body[None,:,:], faces=F[None,:,:])
mesh2_boundaries = find_closest_vertices(mesh1, mesh2, mesh1_boundaries)
avg_edge_length = mesh_edge_lengths(mesh1).mean()
# Define the optimizer and learning rate
optimizer = torch.optim.Adam([mesh1.verts_packed()], lr=0.01)
mesh1.verts_packed().requires_grad = True
optimizer = torch.optim.Adam([mesh1.verts_packed()], lr=0.0001)
# optimizer = torch.optim.SGD([mesh1.verts_packed()], lr=0.001)
# Define the loss functions
def laplacian_smoothing_loss(mesh):
return mesh_laplacian_smoothing(mesh, method="uniform")
return mesh_laplacian_smoothing(mesh, method="uniform") # cotcurv uniform cot
def edge_loss(mesh1, mesh2):
return mesh_edge_loss(mesh1, mesh2)
def edge_loss(mesh):
return mesh_edge_loss(mesh, target_length=avg_edge_length)
def normal_consistency_loss(mesh1, mesh2):
return mesh_normal_consistency(mesh1, mesh2)
def normal_consistency_loss(mesh):
return mesh_normal_consistency(mesh)
# Optimization loop
for i in range(100):
for i in range(10000):
optimizer.zero_grad()
# find boundaries with the find holes function?
# Forward pass
mesh1_out = Meshes(verts=[mesh1.verts_packed()], faces=[mesh1.faces_packed()])
loss = chamfer_distance(mesh1_out, mesh2)
loss += laplacian_smoothing_loss(mesh1_out)
loss += edge_loss(mesh1_out)
loss += normal_consistency_loss(mesh1_out)
ch_loss, _ = chamfer_distance(torch.stack(mesh1_out.verts_list()), torch.stack(mesh2.verts_list()))
bound_ch_loss, _ = chamfer_distance(torch.stack(mesh1_out.verts_list())[:,mesh1_boundaries,:], torch.stack(mesh2.verts_list())[:,mesh2_boundaries,:])
# crotch_ch_loss, _ = chamfer_distance(torch.stack(mesh1_out.verts_list())[:,crotch_mesh,:], torch.stack(mesh2.verts_list())[:,crotch_body,:],)
crotch_ch_loss = torch.mean((torch.stack(mesh1_out.verts_list())[:,crotch_mesh,:] - torch.stack(mesh2.verts_list())[:,crotch_body,:])**2, dim=2).mean()
loss: torch.Tensor = ch_loss * 10 + (bound_ch_loss + crotch_ch_loss) * 5
# loss: torch.Tensor = ch_loss + ch_norm_loss + (bound_ch_loss + bound_ch_norm_loss) * 1
loss += laplacian_smoothing_loss(mesh1_out) * 0.05
loss += edge_loss(mesh1_out) * 100
# loss += edge_loss_custom(torch.stack(mesh1_out.verts_list()), template_orig[None,:,:], mesh1.edges_packed()) * 100
loss += normal_consistency_loss(mesh1_out) * 0.1
# Backward pass
loss.backward()
optimizer.step()
# Print the loss
print(f"Loss: {loss.item()}")
if i % 10 == 0:
print(f"Loss: {loss.item()} chamfer: {ch_loss.item()} ")
# create a plotly plot with the body and the mesh1_out
verts = mesh1_out.verts_packed().detach().cpu().numpy()
star_body = star_body.detach().cpu().numpy()
faces = faces.detach().cpu().numpy()
F = F.detach().cpu().numpy()
mesh1_boundaries = mesh1_boundaries.detach().cpu().numpy()
mesh2_boundaries = mesh2_boundaries.detach().cpu().numpy()
fig = go.Figure(data=[
go.Mesh3d(
x=verts[:,0], y=verts[:,1], z=verts[:,2],
i=faces[:,0], j=faces[:,1], k=faces[:,2],
color='lightpink', opacity=0.80),
go.Mesh3d(
x=star_body[:,0], y=star_body[:,1], z=star_body[:,2],
i=F[:,0], j=F[:,1], k=F[:,2],
color='lightblue', opacity=0.50),
go.Scatter3d(
x=verts[mesh1_boundaries,0], y=verts[mesh1_boundaries,1], z=verts[mesh1_boundaries,2],
mode='markers', marker=dict(size=2, color='red')),
go.Scatter3d(
x=star_body[mesh2_boundaries,0], y=star_body[mesh2_boundaries,1], z=star_body[mesh2_boundaries,2],
mode='markers', marker=dict(size=2, color='blue')),
go.Scatter3d(
x=verts[crotch_mesh,0], y=verts[crotch_mesh,1], z=verts[crotch_mesh,2],
mode='markers', marker=dict(size=2, color='red')),
go.Scatter3d(
x=star_body[crotch_body,0], y=star_body[crotch_body,1], z=star_body[crotch_body,2],
mode='markers', marker=dict(size=2, color='blue')),
])
# make it cuve and limit to -1 and -1
fig.update_layout(scene = dict(
xaxis = dict(range=[-1,1],),
yaxis = dict(range=[-1,1],),
zaxis = dict(range=[-1,1],),),)
fig.update_layout(scene_aspectmode='cube')
fig.show()
pass
if __name__ == "__main__":
folder = '/data-net/dataset/cloth3d/train'
# folder = r'D:\Work\Dataset\cloth3d\train'
folder = '/data-net/code/cloth3d/train'
# folder = r'D:\Work\code\cloth3d\train'
# frames = [f'00761_{i}' for i in range(300)]
# folder = r'Z:\Work\Huni\Barcelona\Dataset\cloth3d\train'
frames = get_frames(folder, 'frames_all.txt')
@@ -203,5 +457,5 @@ if __name__ == "__main__":
register(folder, frame)
break
pass
data/registration_post.py
+482
View File
@@ -0,0 +1,482 @@
import os
import sys
import random
import numpy as np
import plotly.graph_objects as go
import torch
from torch.nn.functional import mse_loss
from pytorch3d.structures import Meshes
from pytorch3d.loss import mesh_laplacian_smoothing, mesh_edge_loss, mesh_normal_consistency, chamfer_distance
from pymeshfix import MeshFix
# os.environ["CUDA_LAUNCH_BLOCKING"] = '1'
sys.path.append(os.path.dirname(os.path.abspath(__file__)) + '/..')
from data.cloth3d_io import (
read_garment_mask,
read_garment_template,
read_posed_garment_frame,
loadInfo,
read_posed_body_frame,
read_posed_body,
read_posed_garment,
read_curvatures,
read_curvature_frame,
writeOBJ,
readOBJ,
quads2tris
)
from smpl.smpl_torch import SMPL
# from custom_utils.custom_utils import _maskFaces
device = 'cuda'
smpl = SMPL(device, 'smpl')
import torch.nn.functional as F
from pytorch3d.structures import Meshes
def find_closest_vertices(mesh1: Meshes, mesh2: Meshes, vertex_indices: torch.Tensor) -> torch.Tensor:
"""
Given two PyTorch3D meshes and a set of vertex indices for mesh1,
find the closest vertices from mesh2 to these vertices.
Args:
mesh1: A PyTorch3D Meshes object representing the first mesh.
mesh2: A PyTorch3D Meshes object representing the second mesh.
vertex_indices: A tensor of shape (N,) representing the indices of the vertices in mesh1.
Returns:
A tensor of shape (N,) representing the indices of the closest vertices in mesh2.
"""
# Extract the vertices of mesh1 and mesh2.
vertices1 = mesh1.verts_packed()
vertices2 = mesh2.verts_packed()
# Extract the vertices of mesh1 that correspond to the given indices.
vertices1_selected = vertices1[vertex_indices]
# Compute the pairwise distances between the selected vertices of mesh1 and all vertices of mesh2.
distances = F.pairwise_distance(vertices1_selected.unsqueeze(1), vertices2.unsqueeze(0), p=2)
# Find the indices of the closest vertices in mesh2.
closest_indices = torch.argmin(distances, dim=1)
return closest_indices
def _maskFaces(V, F, mask, return_unfixed=False):
V_mask = mask
F_mask = np.take(V_mask, F.reshape(-1), axis=0)
F_mask = F_mask.reshape((-1, 3)).sum(axis=1) == 3
V_ind = np.zeros(V_mask.shape)
V_ind[V_mask] = np.arange(V_mask.sum())
F_new = np.take(V_ind, F[F_mask, :].reshape(-1), axis=0).reshape((-1, 3))
V_new = V[V_mask, :]
if return_unfixed:
return V_new, F_new, F[F_mask, :]
else:
return V_new, F_new
def get_frames(folder, frames_file):
frames = []
with open(os.path.join(folder, frames_file), 'r') as f:
for line in f:
frames.append(line.strip())
return frames
def plot(mesh, faces):
fig = go.Figure()
fig.add_trace(go.Mesh3d(
x=mesh[:,0],
y=mesh[:,1],
z=mesh[:,2],
i = faces[:,0],
j = faces[:,1],
k = faces[:,2]
),)
fig.update_layout(scene_aspectmode='manual', scene_aspectratio=dict(x=1, y=1, z=1),
)
helper = dict(xaxis = dict(nticks=4, range=[-1,1],),
yaxis = dict(nticks=4, range=[-1,1],),
zaxis = dict(nticks=4, range=[-1,1],),)
fig.update_layout(scene = helper,
)
fig.show()
def build_lookup_table(repeated):
# iterate through the list and create a dict from each sub array in the list where each dict has one element of the sub array as key and the other elements as value as a list and repeat for each element
lookup = {}
for i in range(len(repeated)):
for j in range(len(repeated[i])):
if len(repeated[i]) == 1:
continue
lookup[repeated[i][j]] = [repeated[i][k] for k in range(len(repeated[i]))] # if k != j
return lookup
from data.cloth3d_io import readPC2Frame
import scipy.io as sio
import scipy
import scipy.spatial as spatial
def local_read_posed_garment_frame(folder, id, garment, frame, pose=None, deorient=False):
pc16_path = os.path.join(folder, id, garment + '.pc16')
verts = readPC2Frame(pc16_path, frame, True)
map = sio.loadmat(os.path.join(folder, id, garment + '_map.mat'))['map'].reshape((-1))
if deorient:
R = spatial.transform.Rotation.from_euler('xyz', pose[:3]).inv()
# R = spatial.transform.Rotation.from_euler([0,0,-zrot])
verts = R.apply(verts)
return verts, map
def local_read_posed_body_frame(folder, id, frame, pose=None, deorient=False):
pc16_path = os.path.join(folder, id, 'Body_posed.pc16')
verts = readPC2Frame(pc16_path, frame, True)
if deorient:
R = spatial.transform.Rotation.from_euler('xyz', pose[:3]).inv()
# R = spatial.transform.Rotation.from_euler([0,0,-zrot])
verts = R.apply(verts)
return verts
def edge_loss_custom(x, y, edges):
x_e = x[:,edges[:,0],:] - x[:,edges[:,1],:]
x_e = torch.sum(x_e ** 2, -1)
x_e = torch.sqrt(x_e + 1e-9)
y_e = y[:,edges[:,0],:] - y[:,edges[:,1],:]
y_e = torch.sum(y_e ** 2, -1)
y_e = torch.sqrt(y_e + 1e-9)
d_e = y_e - x_e
loss = torch.mean(d_e ** 2, dim=-1)[0]
return loss
def crotch(mesh, body, crotch_idx, nr=75):
crotch_middle_on_body = crotch_idx
# create kdtree for body and mesh
body_tree = spatial.cKDTree(body)
mesh_tree = spatial.cKDTree(mesh)
# find closest point on mesh from crotch middle on body along the z axis
crotch_middle_on_mesh = find_closest_point_along_z(body[crotch_middle_on_body], mesh)
crotch_body = body_tree.query(body[crotch_middle_on_body], k=nr)[1]
crotch_mesh = mesh_tree.query(mesh[crotch_middle_on_mesh], k=nr)[1]
body_verts = body[crotch_body]
mesh_verts = mesh[crotch_mesh] - (mesh[crotch_middle_on_mesh] - body[crotch_middle_on_body])
mesh_verts = spatial.cKDTree(mesh_verts)
closest_mesh_verts = mesh_verts.query(body_verts, k=1)[1]
crotch_mesh = crotch_mesh[closest_mesh_verts]
return crotch_body, crotch_mesh
def find_closest_point_along_z(point, verts, eps=0.01):
min_dist = np.inf
closest_point = None
for i, v in enumerate(verts):
dist = np.linalg.norm(point - v)
if dist < min_dist and np.abs(point[0] - v[0]) < eps and np.abs(point[1] - v[1]) < eps:
min_dist = dist
closest_point = i
return closest_point
def mesh_edge_lengths(mesh: Meshes) -> torch.Tensor:
"""
Given a PyTorch3D Meshes object, compute the lengths of all edges in the mesh.
Args:
mesh: A PyTorch3D Meshes object representing the mesh.
Returns:
A tensor of shape (E,) representing the lengths of all edges in the mesh,
where E is the total number of edges in the mesh.
"""
# Extract the vertices and faces of the mesh.
vertices = mesh.verts_packed()
faces = mesh.faces_packed()
# Compute the pairwise distances between all pairs of vertices that form an edge.
edge_indices = faces.view(-1, 3)
edge_vertices = vertices[edge_indices]
edge_lengths = torch.norm(edge_vertices[:, 1:] - edge_vertices[:, :-1], dim=2)
return edge_lengths
# try to optimize the already registered garment ??????????
def register(folder, frame):
id, frame_nr, garment = frame.split()
frame_nr = int(frame_nr)
info = loadInfo(os.path.join(folder, id, 'info.mat'))
raw_mask = read_garment_mask(folder, id, garment, return_raw=True)
# convert it to indices
mask = raw_mask > 0
mesh_posed, map = local_read_posed_garment_frame(folder, id, garment, frame_nr, pose=info['poses'][:,frame_nr], deorient=True)
body_posed = local_read_posed_body_frame(folder, id, frame_nr, pose=info['poses'][:,frame_nr], deorient=True)
template, faces, _, _ = readOBJ(os.path.join(folder, id, garment + '.obj'))
faces = quads2tris(faces)
boundaries = raw_mask == 2
registered_boundaries = torch.tensor(np.where(boundaries[mask])[0]).to(device)
# invert map and use it to invert boudaries
# map_inv = np.zeros(map.max()+1, dtype=np.int32)
# map_inv[map] = np.arange(len(map))
# boundaries = boundaries[map_inv]
# # get the boundary indices
# mesh1_boundaries = torch.tensor(np.where(boundaries)[0]).to(device)
# # pose = torch.zeros(1, 72).to('cuda')
# shape = torch.zeros(1, 11).to(device)
# shape[:, :10] = torch.tensor(info['shape']).to(device)
# shape[:, -1] = torch.tensor(info['gender']).to(device)
# # for star pose
# # pose[:, 5] = 3.1415/4
# # pose[:,8] = -3.1415/4
# STAR_POSE = np.zeros((24, 3), np.float32)
# STAR_POSE[1,2] = .15
# STAR_POSE[2,2] = -.15
# pose = torch.tensor(STAR_POSE.flatten()[None,:]).to(device)
# star_body = smpl(pose, shape, remove=True)[0][0].cpu().numpy() - smpl(pose, shape, remove=True)[4][0].cpu().numpy()
# # rotate star body with 90 degrees around x axis
# R = spatial.transform.Rotation.from_euler('x', np.pi/2)
# star_body = R.apply(star_body).astype(np.float32)
is_skirt = (garment == 'Dress' or garment == 'Skirt')
skirt_mod = '_skirt' if is_skirt else ''
F = np.load(f'smpl/smpl_faces{skirt_mod}.npy').astype(np.int32)
##################### mask body #####################
new_crotch = 4729
# star_body = star_body[mask]
new_crotch = np.zeros((14475), bool)
new_crotch[4729] = True
new_crotch = new_crotch[mask]
new_crotch = np.where(new_crotch)[0][0]
_, F = _maskFaces(np.zeros((14475, 3)), F, mask)
registered = template[map][mask]
# plot(template, faces)
# v = mesh_posed[mask,:]
##################### boundaries #####################
# meshfix = MeshFix(template, faces)
# holes = meshfix.extract_holes()
# new_border = []
# for border in np.array(holes.points):
# # find the index of the vertex in the original mesh
# ind = np.argwhere(np.all(template == border, axis=1))
# new_border.extend(ind.ravel())
# mesh1_boundaries = torch.tensor(new_border).to(device)
# np.save('temp/body_boundaries.npy', new_border)
template_boundaries = torch.tensor(np.load('temp/mesh1_boundaries.npy')).to(device)
# meshfix = MeshFix(template, faces)
# holes = meshfix.extract_holes()
# new_border = []
# for border in np.array(holes.points):
# # find the index of the vertex in the original mesh
# ind = np.argwhere(np.all(template == border, axis=1))
# new_border.extend(ind.ravel())
# mesh1_boundaries = torch.tensor(new_border).to(device)
# np.save('temp/body_boundaries.npy', new_border)
# mesh1_boundaries = torch.tensor(np.load('temp/mesh1_boundaries.npy')).to(device)
##################### crotch #####################
# crotch_body, crotch_mesh = crotch(template, star_body, crotch_idx=new_crotch, nr=50)
# export star_body as obj
# writeOBJ('temp/star_body.obj', star_body, F)
template = torch.tensor(template).to(device)
template_orig = template.clone()
faces = torch.tensor(faces).to(device)
registered = torch.tensor(registered).to(device)
F = torch.tensor(F).to(device)
# verts = template.cpu().numpy()
# star_body = star_body.cpu().numpy()
# faces = faces.detach().cpu().numpy()
# F = F.detach().cpu().numpy()
# fig = go.Figure(data=[
# go.Mesh3d(
# x=verts[:,0], y=verts[:,1], z=verts[:,2],
# i=faces[:,0], j=faces[:,1], k=faces[:,2],
# color='lightpink', opacity=0.50),
# go.Mesh3d(
# x=star_body[:,0], y=star_body[:,1], z=star_body[:,2],
# i=F[:,0], j=F[:,1], k=F[:,2],
# color='lightblue', opacity=0.50),
# go.Scatter3d(
# x=verts[crotch_mesh,0], y=verts[crotch_mesh,1], z=verts[crotch_mesh,2],
# mode='markers', marker=dict(size=2, color='red')),
# go.Scatter3d(
# x=star_body[crotch_body,0], y=star_body[crotch_body,1], z=star_body[crotch_body,2],
# mode='markers', marker=dict(size=2, color='blue')),
# ])
# # make it cuve and limit to -1 and -1
# fig.update_layout(scene = dict(
# xaxis = dict(range=[-1,1],),
# yaxis = dict(range=[-1,1],),
# zaxis = dict(range=[-1,1],),),)
# fig.show()
# Define the meshes to optimize
mesh1 = Meshes(verts=registered[None,:,:], faces=F[None,:,:])
mesh2 = Meshes(verts=template[None,:,:], faces=faces[None,:,:])
mesh1_boundaries = registered_boundaries
mesh2_boundaries = template_boundaries
# mesh2_boundaries = find_closest_vertices(mesh1, mesh2, mesh1_boundaries)
avg_edge_length = mesh_edge_lengths(mesh1).mean()
# Define the optimizer and learning rate
mesh1.verts_packed().requires_grad = True
optimizer = torch.optim.Adam([mesh1.verts_packed()], lr=0.0001)
# optimizer = torch.optim.SGD([mesh1.verts_packed()], lr=0.001)
# Define the loss functions
def laplacian_smoothing_loss(mesh):
return mesh_laplacian_smoothing(mesh, method="uniform") # cotcurv uniform cot
def edge_loss(mesh):
return mesh_edge_loss(mesh, target_length=avg_edge_length)
def normal_consistency_loss(mesh):
return mesh_normal_consistency(mesh)
# Optimization loop
for i in range(10000):
optimizer.zero_grad()
# find boundaries with the find holes function?
# Forward pass
mesh1_out = Meshes(verts=[mesh1.verts_packed()], faces=[mesh1.faces_packed()])
ch_loss, _ = chamfer_distance(torch.stack(mesh1_out.verts_list()), torch.stack(mesh2.verts_list()))
bound_ch_loss, _ = chamfer_distance(torch.stack(mesh1_out.verts_list())[:,mesh1_boundaries,:], torch.stack(mesh2.verts_list())[:,mesh2_boundaries,:])
# crotch_ch_loss, _ = chamfer_distance(torch.stack(mesh1_out.verts_list())[:,crotch_mesh,:], torch.stack(mesh2.verts_list())[:,crotch_body,:],)
# crotch_ch_loss = torch.mean((torch.stack(mesh1_out.verts_list())[:,crotch_mesh,:] - torch.stack(mesh2.verts_list())[:,crotch_body,:])**2, dim=2).mean()
loss: torch.Tensor = ch_loss * 3 + (bound_ch_loss) * 20
# loss: torch.Tensor = ch_loss + ch_norm_loss + (bound_ch_loss + bound_ch_norm_loss) * 1
lap_loss = laplacian_smoothing_loss(mesh1_out)
e_loss = edge_loss(mesh1_out)
# loss += edge_loss_custom(torch.stack(mesh1_out.verts_list()), template_orig[None,:,:], mesh1.edges_packed()) * 100
norm_loss = normal_consistency_loss(mesh1_out)
loss += lap_loss * 0.05 + e_loss * 100 + norm_loss * 0.1
# Backward pass
loss.backward()
optimizer.step()
# Print the loss
if i % 10 == 0:
print(f"it: {i} : loss: {loss:.5f} chamfer: {ch_loss:.6f} bound_chamfer: {bound_ch_loss:.6f} laplacian: {lap_loss:.6f} edge: {e_loss:.6f} normal: {norm_loss:.5f}")
# create a plotly plot with the body and the mesh1_out
registered = mesh1_out.verts_packed().detach().cpu().numpy()
template = mesh2.verts_packed().detach().cpu().numpy()
faces = faces.detach().cpu().numpy()
F = F.detach().cpu().numpy()
# faces, F = F, faces
# verts, star_body = star_body, verts
mesh1_boundaries, mesh2_boundaries = mesh1_boundaries.detach().cpu().numpy(), mesh2_boundaries.detach().cpu().numpy()
# mesh2_boundaries = mesh1_boundaries.detach().cpu().numpy()
fig = go.Figure(data=[
go.Mesh3d(
x=template[:,0], y=template[:,1], z=template[:,2],
i=faces[:,0], j=faces[:,1], k=faces[:,2],
color='lightblue', opacity=0.80),
go.Mesh3d(
x=registered[:,0], y=registered[:,1], z=registered[:,2],
i=F[:,0], j=F[:,1], k=F[:,2],
color='lightpink', opacity=0.50),
go.Scatter3d(
x=registered[mesh1_boundaries,0], y=registered[mesh1_boundaries,1], z=registered[mesh1_boundaries,2],
mode='markers', marker=dict(size=2, color='red')),
go.Scatter3d(
x=template[mesh2_boundaries,0], y=template[mesh2_boundaries,1], z=template[mesh2_boundaries,2],
mode='markers', marker=dict(size=2, color='blue')),
# go.Scatter3d(
# x=verts[crotch_mesh,0], y=verts[crotch_mesh,1], z=verts[crotch_mesh,2],
# mode='markers', marker=dict(size=2, color='red')),
# go.Scatter3d(
# x=star_body[crotch_body,0], y=star_body[crotch_body,1], z=star_body[crotch_body,2],
# mode='markers', marker=dict(size=2, color='blue')),
])
# make it cuve and limit to -1 and -1
fig.update_layout(scene = dict(
xaxis = dict(range=[-1,1],),
yaxis = dict(range=[-1,1],),
zaxis = dict(range=[-1,1],),),)
fig.update_layout(scene_aspectmode='cube')
fig.show()
# create a plotly wireframe plot of registered
fig = go.Figure()
for edge in mesh1_out.edges_packed().detach().cpu().numpy():
fig.add_trace(go.Scatter3d(
x=registered[edge,0], y=registered[edge,1], z=registered[edge,2],
mode='lines', line=dict(color='red', width=1)))
fig.update_layout(scene = dict(
xaxis = dict(range=[-1,1],),
yaxis = dict(range=[-1,1],),
zaxis = dict(range=[-1,1],),),)
fig.update_layout(scene_aspectmode='cube')
fig.show()
pass
if __name__ == "__main__":
folder = '/data-net/code/cloth3d/train'
# folder = r'D:\Work\code\cloth3d\train'
# frames = [f'00761_{i}' for i in range(300)]
# folder = r'Z:\Work\Huni\Barcelona\Dataset\cloth3d\train'
frames = get_frames(folder, 'frames_all.txt')
root = '/'.join(folder.split('/')[:-1])
type = folder.split('/')[-1]
use_no_orient = False
no_shape = False
is_skirt = False
skirt_modifier = '_skirt' if is_skirt else ''
orient_mod = '_no_orient' if use_no_orient else ''
shape_mod = '_no_shape' if no_shape else ''
# mask = np.load('/data-net/dataset/00761/mesh_mask/00761.npy')
# faces = np.load('/data-net/dataset/00761/faces/00761.npy')
for frame in frames:
frame = random.choice(frames)
# frame = '00761 0 Tshirt'
frame = '01893 256 Jumpsuit'
register(folder, frame)
break
pass
data/registration_weights.py
+423 -17
View File
@@ -30,7 +30,7 @@ from data.cloth3d_io import (
from smpl.smpl_torch import SMPL
# from custom_utils.custom_utils import _maskFaces
device = 'cuda'
device = 'cpu'
smpl = SMPL(device, 'smpl')
@@ -315,7 +315,7 @@ def register(folder, frame):
return mesh_normal_consistency(mesh)
# Optimization loop
for i in range(4000):
for i in range(1):
optimizer.zero_grad()
# find boundaries with the find holes function?
@@ -430,20 +430,139 @@ def register(folder, frame):
pass
def order(verts, edges):
# Choose a starting vertex
start_vertex = 0
# Initialize the visited and ordered arrays
visited = np.zeros(verts.shape[0], dtype=bool)
ordered_vertices = []
ordered_edges = []
# Define the depth-first search function
def dfs(vertex):
visited[vertex] = True
ordered_vertices.append(vertex)
for i in range(edges.shape[0]):
if edges[i, 0] == vertex and not visited[edges[i, 1]]:
ordered_edges.append(i)
dfs(edges[i, 1])
elif edges[i, 1] == vertex and not visited[edges[i, 0]]:
ordered_edges.append(i)
dfs(edges[i, 0])
# Perform depth-first search starting from the chosen vertex
dfs(start_vertex)
return ordered_vertices, ordered_edges
def reindex_edges(edges, mask):
lines_new = np.zeros_like(edges)
for i in range(edges.shape[0]):
line_1 = edges[i,0]
line_2 = edges[i,1]
line_1_new = line_1 - np.count_nonzero(~mask[:line_1])
line_2_new = line_2 - np.count_nonzero(~mask[:line_2])
lines_new[i] = [line_1_new, line_2_new]
lines_new = np.asarray(lines_new)
return lines_new
def reindex_vertices(verts, mask):
verts_new = np.zeros_like(verts)
for i in range(verts.shape[0]):
vert = verts[i]
vert_new = vert - np.count_nonzero(~mask[:vert])
verts_new[i] = vert_new
verts_new = np.asarray(verts_new)
return verts_new
def generate_edges(registered_cloth, registered_holes, registered_border):
lines = registered_holes.lines
lines_filtered = []
for i in range(0, lines.shape[0]-2, 3):
ind = np.argwhere(np.all(registered_cloth == registered_holes.points[lines[i]], axis=1))
ind2 = np.argwhere(np.all(registered_cloth == registered_holes.points[lines[i+1]], axis=1))
if ind.ravel() in registered_border and ind2.ravel() in registered_border:
lines_filtered.append(lines[i+1:i+3])
lines = np.array(lines_filtered)
return lines
def reindex_lines(lines, registered_border_mask):
lines_new = np.zeros_like(lines)
for i in range(lines.shape[0]):
line_1 = lines[i,0]
line_2 = lines[i,1]
line_1_new = line_1 - np.count_nonzero(~registered_border_mask[:line_1])
line_2_new = line_2 - np.count_nonzero(~registered_border_mask[:line_2])
lines_new[i] = [line_1_new, line_2_new]
lines_new = np.asarray(lines_new)
return lines_new
def get_boundary_components(registered_border, lines_new):
from scipy.sparse import csr_matrix
from scipy.sparse.csgraph import connected_components
n = len(registered_border)
adj_matrix = csr_matrix((np.ones(lines_new.shape[0]), (lines_new[:, 0], lines_new[:, 1])), shape=(n, n))
# Find the connected components of the graph
n_components, labels = connected_components(adj_matrix)
registered_border = np.array(registered_border)
bounds = []
bounds_edges_global = []
bounds_edges_local = []
for i in range(n_components):
ind = np.argwhere(labels == i)[:,0]
if len(ind) < 10:
continue
bounds.append(registered_border[ind])
temp_global = []
temp_local = []
for j in range(lines_new.shape[0]):
if lines_new[j, 0] in ind and lines_new[j, 1] in ind:
# if lines_new[j, 0] < lines_new[j, 1]:
temp_global.append(registered_border[lines_new[j]])
temp_local.append(lines_new[j])
# else:
# temp.append(registered_border[np.asarray([lines_new[j, 1], lines_new[j, 0]])])
bounds_edges_global.append(np.asarray(temp_global))
bounds_edges_local.append(np.asarray(temp_local))
return bounds, bounds_edges_global, bounds_edges_local, labels
def get_boundaries(folder, frame):
from pymeshfix import MeshFix
from scipy.spatial import KDTree
id, frame_nr, garment = frame.split()
frame_nr = int(frame_nr)
registered_cloth = np.load('temp/registered_cloth.npy')
registered_faces = np.load('temp/registered_faces.npy')
registred_weights = np.load('temp/registered_weights.npy')
registered_cloth = np.load('temp/matlab_match_registered.npy')
registered_faces = np.load('temp/matlab_match_faces.npy')
registered_weight_faces = np.load('temp/matlab_match_weights_faces.npy')
registred_weights = np.load('temp/matlab_match_weights_weights.npy')
_, faces, _, _ = readOBJ(os.path.join(folder, id, garment + '.obj'))
faces = quads2tris(faces)
template = np.load('temp/mapping_cloth.npy')
registered_meshfix = MeshFix(registered_cloth, registered_faces)
registered_cloth = (template[registered_weight_faces] * registred_weights[:,:,None]).sum(1)
registered_meshfix = MeshFix(registered_cloth, registered_faces.astype(np.int32))
registered_holes = registered_meshfix.extract_holes()
template_meshfix = MeshFix(template, faces)
@@ -457,17 +576,57 @@ def get_boundaries(folder, frame):
template_border_tree = KDTree(template[template_border])
registered_border = []
for border in np.array(registered_holes.points):
registered_border_mask = np.zeros(registered_holes.points.shape[0], dtype=bool)
for i, border in enumerate(np.array(registered_holes.points)):
q = template_border_tree.query(border)
if q[0] > 0.02:
continue
# find the index of the vertex in the original mesh
ind = np.argwhere(np.all(registered_cloth == border, axis=1))
registered_border.extend(ind.ravel())
registered_border_mask[i] = True
np.save('temp/registered_border.npy', registered_border)
np.save('temp/template_border.npy', template_border)
lines = generate_edges(registered_cloth, registered_holes, registered_border)
lines_new = reindex_lines(lines, registered_border_mask)
bounds, bounds_edges_global, bounds_edges_local, labels = get_boundary_components(registered_border, lines_new)
for k in range(labels.max()+1):
bounds_edges_reindexed = reindex_edges(bounds_edges_local[k].copy(), labels == k)
verts_order, edges_order = order(bounds[k], bounds_edges_reindexed)
traces = []
for i in range(bounds_edges_global[k].shape[0]):
traces.append(go.Scatter3d(
x=registered_cloth[bounds_edges_global[k][i],0], y=registered_cloth[bounds_edges_global[k][i],1], z=registered_cloth[bounds_edges_global[k][i],2],
mode='lines', line=dict(color='red', width=1)))
for i in range(15):
traces.append(go.Scatter3d(
x=registered_cloth[bounds_edges_global[k][edges_order][i],0], y=registered_cloth[bounds_edges_global[k][edges_order][i],1], z=registered_cloth[bounds_edges_global[k][edges_order][i],2],
mode='lines', line=dict(color='blue', width=1)))
fig = go.Figure(data=traces)
fig.update_layout(scene = dict(
xaxis = dict(range=[-1,1],),
yaxis = dict(range=[-1,1],),
zaxis = dict(range=[-1,1],),),)
fig.update_layout(scene_aspectmode='cube')
fig.show()
k = 0
traces = [go.Scatter3d(
x=registered_cloth[bounds[k],0], y=registered_cloth[bounds[k],1], z=registered_cloth[bounds[k],2],
mode='markers', line=dict(color='red', width=1))]
fig = go.Figure(data=traces)
fig.update_layout(scene = dict(
xaxis = dict(range=[-1,1],),
yaxis = dict(range=[-1,1],),
zaxis = dict(range=[-1,1],),),)
fig.update_layout(scene_aspectmode='cube')
fig.show()
# plot registered cloth and tempalte as scatterplots and add another two scatter plots for the two boundary sets
fig = go.Figure(data=[
go.Scatter3d(
@@ -501,6 +660,7 @@ def get_boundaries(folder, frame):
pass
def optimize_boundaries(folder, frame):
id, frame_nr, garment = frame.split()
frame_nr = int(frame_nr)
@@ -519,7 +679,6 @@ def optimize_boundaries(folder, frame):
template_border = np.load('temp/template_border.npy')
registered_border = np.load('temp/registered_border.npy')
template = torch.tensor(template).to(device).to(torch.float32)
faces = torch.tensor(faces).to(device)
registered_faces = torch.tensor(registered_faces).to(device)
@@ -527,7 +686,6 @@ def optimize_boundaries(folder, frame):
registered_weights = torch.tensor(registered_weights).to(device).to(torch.float32)
registered_cloth = torch.tensor(registered_cloth).to(device).to(torch.float32)
mesh2 = Meshes(verts=template[None,:,:], faces=faces[None,:,:])
avg_edge_length = mesh_edge_lengths(mesh2).mean()
@@ -553,7 +711,7 @@ def optimize_boundaries(folder, frame):
return mesh_normal_consistency(mesh)
# Optimization loop
for i in range(4000):
for i in range(1):
optimizer.zero_grad()
# find boundaries with the find holes function?
@@ -576,7 +734,7 @@ def optimize_boundaries(folder, frame):
e_loss = edge_loss(mesh1_out)
# loss += edge_loss_custom(torch.stack(mesh1_out.verts_list()), template_orig[None,:,:], mesh1.edges_packed()) * 100
norm_loss = normal_consistency_loss(mesh1_out)
loss += lap_loss * 0.05 + e_loss * 2 + norm_loss * 0.1
loss += lap_loss * 0.5 + e_loss * 2 + norm_loss * 0.1
# Backward pass
loss.backward()
@@ -584,8 +742,8 @@ def optimize_boundaries(folder, frame):
# Print the loss
if i % 10 == 0:
print(f"it: {i} : loss: {loss:.5f} bound_chamfer: {ch_bound_loss:.6f}")
# print(f"it: {i} : loss: {loss:.5f} bound_chamfer: {ch_bound_loss:.6f} chamfer: {ch_loss:.6f} laplacian: {lap_loss:.6f} edge: {e_loss:.6f} normal: {norm_loss:.5f}")
# print(f"it: {i} : loss: {loss:.5f} bound_chamfer: {ch_bound_loss:.6f}")
print(f"it: {i} : loss: {loss:.5f} bound_chamfer: {ch_bound_loss:.6f} laplacian: {lap_loss:.6f} edge: {e_loss:.6f} normal: {norm_loss:.5f}")
registered = mesh1_out.verts_packed().detach().cpu().numpy()
template = mesh2.verts_packed().detach().cpu().numpy()
@@ -613,6 +771,163 @@ def optimize_boundaries(folder, frame):
fig.update_layout(scene_aspectmode='cube')
fig.show()
fig = go.Figure(data=[
go.Scatter3d(
x=registered[:,0], y=registered[:,1], z=registered[:,2],
mode='markers', marker=dict(size=4, color='red')),
go.Scatter3d(
x=template[:,0], y=template[:,1], z=template[:,2],
mode='markers', marker=dict(size=2, color='blue')),
])
fig.update_layout(scene = dict(
xaxis = dict(range=[-1,1],),
yaxis = dict(range=[-1,1],),
zaxis = dict(range=[-1,1],),),)
fig.update_layout(scene_aspectmode='cube')
fig.show()
pass
def load_selected_saves():
registered = np.load('temp/matlab_match_registered.npy')
garment = np.load('temp/matlab_match_garment.npy')
mask = np.load('temp/matlab_match_mask.npy')
mask_bool = np.load('temp/matlab_match_mask.npy')
weights_faces = np.load('temp/matlab_match_weights_faces.npy')
weights_weights = np.load('temp/matlab_match_weights_weights.npy')
faces = np.load('temp/matlab_match_faces.npy')
boundaries = np.load('temp/matlab_match_boundaries.npy')
return registered, garment, mask, mask_bool, weights_faces, weights_weights, faces, boundaries
def optimize_boundaries_2(folder, frame):
registered, garment, mask, mask_bool, weights_faces, weights_weights, faces, boundaries = load_selected_saves()
id, frame_nr, garment_name = frame.split()
registered_cloth = registered
registered_faces = faces
registered_weights = weights_weights
_, template_faces, _, _ = readOBJ(os.path.join(folder, id, garment_name + '.obj'))
template_faces = quads2tris(template_faces)
template = np.load('temp/mapping_cloth.npy')
mapping_mask = mask
mapping_faces = weights_faces
template_border = np.load('temp/template_border.npy')
registered_border = np.load('temp/registered_border.npy')
template = torch.tensor(template).to(device).to(torch.float32)
template_faces = torch.tensor(template_faces).to(device)
registered_faces = torch.tensor(registered_faces).to(device)
mapping_faces = torch.tensor(mapping_faces.astype(np.int64)).to(device)
registered_weights = torch.tensor(registered_weights).to(device).to(torch.float32)
registered_cloth = torch.tensor(registered_cloth).to(device).to(torch.float32)
registered_border = torch.tensor(registered_border).to(device).to(torch.int64)
mesh2 = Meshes(verts=template[None,:,:], faces=template_faces[None,:,:])
avg_edge_length = mesh_edge_lengths(mesh2).mean()
optim_mask = torch.zeros(registered_cloth.shape[0]).to(device).to(torch.bool)
optim_mask[registered_border] = True
mapping_weights_masked = registered_weights[optim_mask]
mapping_weights_nonmasked = registered_weights[~optim_mask]
mapping_weights_masked.requires_grad = True
# Define the optimizer and learning rate
optimizer = torch.optim.Adam([mapping_weights_masked], lr=0.001)
# optimizer = torch.optim.SGD([mesh1.verts_packed()], lr=0.001)
# Define the loss functions
def laplacian_smoothing_loss(mesh):
return mesh_laplacian_smoothing(mesh, method="uniform") # cotcurv uniform cot
def edge_loss(mesh):
return mesh_edge_loss(mesh, target_length=avg_edge_length)
def normal_consistency_loss(mesh):
return mesh_normal_consistency(mesh)
# Optimization loop
for i in range(1000):
optimizer.zero_grad()
# find boundaries with the find holes function?
registered_masked = (template[mapping_faces[optim_mask],:] * mapping_weights_masked[:,:,None]).sum(1)
registered_nonmasked = (template[mapping_faces[~optim_mask],:] * mapping_weights_nonmasked[:,:,None]).sum(1)
# registered = torch.cat((registered_masked, registered_nonmasked), 0)
registered = torch.zeros_like(registered_weights)
registered[optim_mask] = registered_masked
registered[~optim_mask] = registered_nonmasked
# Forward pass
mesh1_out = Meshes(verts=registered[None,:,:], faces=registered_faces[None,:,:])
# ch_loss, _ = chamfer_distance(torch.stack(mesh1_out.verts_list()), torch.stack(mesh2.verts_list()),
# )
ch_bound_loss, _ = chamfer_distance(torch.stack(mesh1_out.verts_list())[:,registered_border,:], torch.stack(mesh2.verts_list())[:, template_border,:],
)
# loss: torch.Tensor = ch_loss * 1 + ch_bound_loss * 20
loss: torch.Tensor = ch_bound_loss * 50
lap_loss = laplacian_smoothing_loss(mesh1_out)
e_loss = edge_loss(mesh1_out)
# loss += edge_loss_custom(torch.stack(mesh1_out.verts_list()), template_orig[None,:,:], mesh1.edges_packed()) * 100
norm_loss = normal_consistency_loss(mesh1_out)
loss += lap_loss * 0.5 + e_loss * 2 + norm_loss * 0.1
# Backward pass
loss.backward()
optimizer.step()
# Print the loss
if i % 10 == 0:
# print(f"it: {i} : loss: {loss:.5f} bound_chamfer: {ch_bound_loss:.6f}")
print(f"it: {i} : loss: {loss:.5f} bound_chamfer: {ch_bound_loss:.6f} laplacian: {lap_loss:.6f} edge: {e_loss:.6f} normal: {norm_loss:.5f}")
registered = mesh1_out.verts_packed().detach().cpu().numpy()
template = mesh2.verts_packed().detach().cpu().numpy()
template_faces = template_faces.detach().cpu().numpy()
optim_mask = optim_mask.detach().cpu().numpy()
registered_faces = registered_faces.detach().cpu().numpy()
registered_border = registered_border.detach().cpu().numpy()
# faces, F = F, faces
# verts, star_body = star_body, verts
from scipy.spatial import KDTree
template_border_tree = KDTree(template[template_border,:])
for i, border in enumerate(registered_border):
dist, idx = template_border_tree.query(registered[border,:], k=2)
weights_faces[border,:] = np.concatenate((template_border[idx], [0]))
weights_weights[border,:] = np.concatenate((1 - dist / np.sum(dist), [0]))
registered = (template[weights_faces] * weights_weights[:,:,None]).sum(1)
fig = go.Figure(data=[
go.Mesh3d(
x=template[:,0], y=template[:,1], z=template[:,2],
i=template_faces[:,0], j=template_faces[:,1], k=template_faces[:,2],
color='lightblue', opacity=0.80),
go.Mesh3d(
x=registered[:,0], y=registered[:,1], z=registered[:,2],
i=registered_faces[:,0], j=registered_faces[:,1], k=registered_faces[:,2],
color='lightpink', opacity=0.50),
go.Scatter3d(
x=registered[registered_border,0], y=registered[registered_border,1], z=registered[registered_border,2],
mode='markers', marker=dict(size=4, color='red')),
go.Scatter3d(
x=template[template_border,0], y=template[template_border,1], z=template[template_border,2],
mode='markers', marker=dict(size=2, color='blue')),
])
fig.update_layout(scene = dict(
xaxis = dict(range=[-1,1],),
yaxis = dict(range=[-1,1],),
zaxis = dict(range=[-1,1],),),)
fig.update_layout(scene_aspectmode='cube')
fig.show()
fig = go.Figure(data=[
go.Scatter3d(
@@ -631,9 +946,98 @@ def optimize_boundaries(folder, frame):
pass
def flatten_boundary(folder, frame):
from scipy.spatial import KDTree
registered, garment, mask, mask_bool, weights_faces, weights_weights, faces, boundaries = load_selected_saves()
id, frame_nr, garment_name = frame.split()
registered_faces = faces
registered_weights = weights_weights
_, template_faces, _, _ = readOBJ(os.path.join(folder, id, garment_name + '.obj'))
template_faces = quads2tris(template_faces)
template = np.load('temp/mapping_cloth.npy')
registered_cloth = (template[weights_faces] * weights_weights[:,:,None]).sum(1)
template_border = np.load('temp/template_border.npy')
registered_border = np.load('temp/registered_border.npy')
template_border_tree = KDTree(template[template_border,:])
for i, border in enumerate(registered_border):
dist, idx = template_border_tree.query(registered_cloth[border,:], k=2)
weights_faces[border,:] = np.concatenate((template_border[idx], [0]))
weights_weights[border,:] = np.concatenate((1 - dist / np.sum(dist), [0]))
registered_cloth = (template[weights_faces] * weights_weights[:,:,None]).sum(1)
fig = go.Figure(data=[
go.Mesh3d(
x=template[:,0], y=template[:,1], z=template[:,2],
i=template_faces[:,0], j=template_faces[:,1], k=template_faces[:,2],
color='lightblue', opacity=0.80),
go.Mesh3d(
x=registered_cloth[:,0], y=registered_cloth[:,1], z=registered_cloth[:,2],
i=registered_faces[:,0], j=registered_faces[:,1], k=registered_faces[:,2],
color='lightpink', opacity=0.50),
go.Scatter3d(
x=registered_cloth[registered_border,0], y=registered_cloth[registered_border,1], z=registered_cloth[registered_border,2],
mode='markers', marker=dict(size=4, color='red')),
go.Scatter3d(
x=template[template_border,0], y=template[template_border,1], z=template[template_border,2],
mode='markers', marker=dict(size=2, color='blue')),
])
fig.update_layout(scene = dict(
xaxis = dict(range=[-1,1],),
yaxis = dict(range=[-1,1],),
zaxis = dict(range=[-1,1],),),)
fig.update_layout(scene_aspectmode='cube')
fig.show()
pass
def fill_holes(folder, frame):
from scipy.spatial import KDTree
from pymeshfix import MeshFix
import pyvista as pv
id, frame_nr, garment = frame.split()
frame_nr = int(frame_nr)
registered_cloth = np.load('temp/registered_cloth.npy')
registered_faces = np.load('temp/registered_faces.npy')
registered_weights = np.load('temp/registered_weights.npy')
_, faces, _, _ = readOBJ(os.path.join(folder, id, garment + '.obj'))
faces = quads2tris(faces)
template = np.load('temp/mapping_cloth.npy')
mapping_mask = np.load('temp/mapping_mask.npy')
mapping_faces = np.load('temp/mapping_faces.npy')
mapping_weights = np.load('temp/mapping_weights.npy')
mapping_faces, mapping_weights = mapping_faces[mapping_mask], mapping_weights[mapping_mask]
template_border = np.load('temp/template_border.npy')
registered_border = np.load('temp/registered_border.npy')
template_border_tree = KDTree(template[template_border])
registered_border_tree = KDTree(registered_cloth[registered_border])
registered_meshfix = MeshFix(registered_cloth, registered_faces)
registered_holes = registered_meshfix.extract_holes()
template_meshfix = MeshFix(template, faces)
template_holes = template_meshfix.extract_holes()
for border in np.array(registered_holes.points):
q = registered_border_tree.query(border)
if __name__ == "__main__":
folder = '/data-net/code/cloth3d/train'
# folder = r'D:\Work\code\cloth3d\train'
# folder = '/data-net/code/cloth3d/train'
folder = r'D:\Work\code\cloth3d\train'
# frames = [f'00761_{i}' for i in range(300)]
# folder = r'Z:\Work\Huni\Barcelona\Dataset\cloth3d\train'
frames = get_frames(folder, 'frames_all.txt')
@@ -659,7 +1063,9 @@ if __name__ == "__main__":
# register(folder, frame)
# get_boundaries(folder, frame)
optimize_boundaries(folder, frame)
get_boundaries(folder, frame)
# optimize_boundaries_2(folder, frame)
# flatten_boundary(folder, frame)
# fill_holes(folder, frame)
break
pass