Spaces:
Sleeping
Sleeping
working proto
Browse files- app.py +46 -13
- model_scripted.pt +1 -1
- requirements.txt +2 -3
- sphere.obj +0 -0
app.py
CHANGED
@@ -1,8 +1,7 @@
|
|
1 |
-
from pytorch3d.io import load_obj
|
2 |
-
from pytorch3d.structures import Meshes
|
3 |
import torch
|
4 |
import gradio as gr
|
5 |
import plotly.graph_objects as go
|
|
|
6 |
|
7 |
device = torch.device("cpu")
|
8 |
model = torch.jit.load('model_scripted.pt').to(device)
|
@@ -14,11 +13,10 @@ def normalize_vertices(verts):
|
|
14 |
scale = max(verts.abs().max(0)[0])
|
15 |
return verts / scale
|
16 |
|
17 |
-
|
18 |
def plot_3d_results(verts, faces, uv_seam_edge_indices):
|
19 |
# Convert vertices to NumPy for easier manipulation
|
20 |
verts_np = verts.cpu().numpy()
|
21 |
-
faces_np = faces.
|
22 |
|
23 |
# Prepare the vertex coordinates for the Mesh3d plot
|
24 |
x, y, z = verts_np[:, 0], verts_np[:, 1], verts_np[:, 2]
|
@@ -56,21 +54,56 @@ def plot_3d_results(verts, faces, uv_seam_edge_indices):
|
|
56 |
|
57 |
|
58 |
def generate_prediction(file_input, treshold_value=0.5):
|
59 |
-
|
60 |
-
|
61 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
62 |
|
63 |
model.eval()
|
64 |
|
65 |
-
test_verts = mesh.verts_packed().to(device)
|
66 |
-
test_edges = mesh.edges_packed().to(device)
|
67 |
-
|
68 |
with torch.no_grad():
|
69 |
-
test_outputs_logits = model(
|
70 |
test_outputs = torch.sigmoid(test_outputs_logits).to(device)
|
71 |
test_predictions = (test_outputs > treshold_value).int().cpu()
|
72 |
-
|
73 |
-
|
|
|
74 |
|
75 |
# Return the HTML content generated by plot_3d_results
|
76 |
return plot_3d_results(verts, faces, uv_seam_edges)
|
|
|
|
|
|
|
1 |
import torch
|
2 |
import gradio as gr
|
3 |
import plotly.graph_objects as go
|
4 |
+
import trimesh
|
5 |
|
6 |
device = torch.device("cpu")
|
7 |
model = torch.jit.load('model_scripted.pt').to(device)
|
|
|
13 |
scale = max(verts.abs().max(0)[0])
|
14 |
return verts / scale
|
15 |
|
|
|
16 |
def plot_3d_results(verts, faces, uv_seam_edge_indices):
|
17 |
# Convert vertices to NumPy for easier manipulation
|
18 |
verts_np = verts.cpu().numpy()
|
19 |
+
faces_np = faces.cpu().numpy()
|
20 |
|
21 |
# Prepare the vertex coordinates for the Mesh3d plot
|
22 |
x, y, z = verts_np[:, 0], verts_np[:, 1], verts_np[:, 2]
|
|
|
54 |
|
55 |
|
56 |
def generate_prediction(file_input, treshold_value=0.5):
|
57 |
+
# Load the triangle mesh
|
58 |
+
mesh = trimesh.load_mesh(file_input)
|
59 |
+
|
60 |
+
# For production, we should use a faster method to preprocess the mesh!
|
61 |
+
|
62 |
+
# Convert vertices to a PyTorch tensor
|
63 |
+
vertices = torch.tensor(mesh.vertices, dtype=torch.float32)
|
64 |
+
|
65 |
+
# Initialize containers for unique vertices and mapping
|
66 |
+
unique_vertices = []
|
67 |
+
vertex_mapping = {}
|
68 |
+
new_faces = []
|
69 |
+
|
70 |
+
# Populate unique vertices and create new faces with updated indices
|
71 |
+
for face in mesh.faces:
|
72 |
+
new_face = []
|
73 |
+
for orig_index in face:
|
74 |
+
vertex = tuple(vertices[orig_index].tolist()) # Convert to tuple (hashable)
|
75 |
+
if vertex not in vertex_mapping:
|
76 |
+
vertex_mapping[vertex] = len(unique_vertices)
|
77 |
+
unique_vertices.append(vertices[orig_index])
|
78 |
+
new_face.append(vertex_mapping[vertex])
|
79 |
+
new_faces.append(new_face)
|
80 |
+
|
81 |
+
# Create edge set to ensure uniqueness
|
82 |
+
edge_set = set()
|
83 |
+
for face in new_faces:
|
84 |
+
# Unpack the vertex indices
|
85 |
+
v1, v2, v3 = face
|
86 |
+
# Create undirected edges (use tuple sorting to ensure uniqueness)
|
87 |
+
edge_set.add(tuple(sorted((v1, v2))))
|
88 |
+
edge_set.add(tuple(sorted((v2, v3))))
|
89 |
+
edge_set.add(tuple(sorted((v1, v3))))
|
90 |
+
|
91 |
+
# Convert edges back to tensor
|
92 |
+
edges = torch.tensor(list(edge_set), dtype=torch.long)
|
93 |
+
|
94 |
+
# Convert unique vertices and new faces back to tensors
|
95 |
+
verts = torch.stack(unique_vertices)
|
96 |
+
faces = torch.tensor(new_faces, dtype=torch.long)
|
97 |
|
98 |
model.eval()
|
99 |
|
|
|
|
|
|
|
100 |
with torch.no_grad():
|
101 |
+
test_outputs_logits = model(verts, edges).to(device)
|
102 |
test_outputs = torch.sigmoid(test_outputs_logits).to(device)
|
103 |
test_predictions = (test_outputs > treshold_value).int().cpu()
|
104 |
+
|
105 |
+
uv_seam_edges_mask = test_predictions.cpu().squeeze() == 1
|
106 |
+
uv_seam_edges = edges[uv_seam_edges_mask].cpu().tolist()
|
107 |
|
108 |
# Return the HTML content generated by plot_3d_results
|
109 |
return plot_3d_results(verts, faces, uv_seam_edges)
|
model_scripted.pt
CHANGED
@@ -1,3 +1,3 @@
|
|
1 |
version https://git-lfs.github.com/spec/v1
|
2 |
-
oid sha256:
|
3 |
size 255324
|
|
|
1 |
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:6f0d5dcd806540ba8061e3d70ed17eac539e560ae0932f04679f706898126588
|
3 |
size 255324
|
requirements.txt
CHANGED
@@ -1,4 +1,3 @@
|
|
1 |
-
git+https://github.com/facebookresearch/pytorch3d.git
|
2 |
torch
|
3 |
-
|
4 |
-
|
|
|
|
|
1 |
torch
|
2 |
+
plotly
|
3 |
+
trimesh
|
sphere.obj
CHANGED
The diff for this file is too large to render.
See raw diff
|
|