Spaces:
Running
on
L40S
Running
on
L40S
update
Browse files- app.py +2 -2
- infer_api.py +34 -33
- refine/mesh_refine.py +23 -12
app.py
CHANGED
@@ -47,8 +47,8 @@ This is official demo for our CVPR 2025 paper <a href="">StdGEN: Semantic-Decomp
|
|
47 |
|
48 |
Code: <a href='https://github.com/hyz317/StdGEN' target='_blank'>GitHub</a>. Paper: <a href='https://arxiv.org/abs/2411.05738' target='_blank'>ArXiv</a>.
|
49 |
|
50 |
-
❗️❗️❗️**Important Notes:**
|
51 |
-
1. Refinement stage takes about ~
|
52 |
|
53 |
2. You can upload any reference image (with or without background), A-pose images are also supported (white bkg required). If the image has an alpha channel (transparency), background segmentation will be automatically performed. Alternatively, you can pre-segment the background using other tools and upload the result directly.
|
54 |
|
|
|
47 |
|
48 |
Code: <a href='https://github.com/hyz317/StdGEN' target='_blank'>GitHub</a>. Paper: <a href='https://arxiv.org/abs/2411.05738' target='_blank'>ArXiv</a>.
|
49 |
|
50 |
+
❗️❗️❗️**Important Notes:** This is only a **PREVIEW** version with lower quality. We only perform color back-projection to clothes and hair. Please refer to GitHub repo for complete version.
|
51 |
+
1. Refinement stage takes about ~2.5min, and the mesh result may possibly delayed due to the server load, please wait patiently.
|
52 |
|
53 |
2. You can upload any reference image (with or without background), A-pose images are also supported (white bkg required). If the image has an alpha channel (transparency), background segmentation will be automatically performed. Alternatively, you can pre-segment the background using other tools and upload the result directly.
|
54 |
|
infer_api.py
CHANGED
@@ -542,19 +542,19 @@ def get_distract_mask(generator, color_0, color_1, normal_0=None, normal_1=None,
|
|
542 |
return distract_mask, distract_bbox, random_sampled_points, final_mask
|
543 |
|
544 |
|
545 |
-
infer_refine_sam = sam_model_registry["vit_h"](checkpoint="./ckpt/sam_vit_h_4b8939.pth").cuda()
|
546 |
-
infer_refine_generator = SamAutomaticMaskGenerator(
|
547 |
-
|
548 |
-
|
549 |
-
|
550 |
-
|
551 |
-
|
552 |
-
|
553 |
-
|
554 |
-
)
|
555 |
infer_refine_outside_ratio = 0.20
|
556 |
|
557 |
-
|
558 |
def infer_refine(meshes, imgs):
|
559 |
fixed_v, fixed_f, fixed_t = None, None, None
|
560 |
flow_vert, flow_vector = None, None
|
@@ -564,7 +564,6 @@ def infer_refine(meshes, imgs):
|
|
564 |
|
565 |
mv, proj = make_star_cameras_orthographic(8, 1, r=1.2)
|
566 |
mv = mv[[4, 3, 2, 0, 6, 5]]
|
567 |
-
renderer = NormalsRenderer(mv,proj,(1024,1024))
|
568 |
|
569 |
results = []
|
570 |
|
@@ -576,12 +575,17 @@ def infer_refine(meshes, imgs):
|
|
576 |
mesh_v, mesh_f = mesh.vertices, mesh.faces
|
577 |
|
578 |
if last_colors is None:
|
579 |
-
|
580 |
-
|
581 |
-
|
582 |
-
|
583 |
-
|
584 |
-
|
|
|
|
|
|
|
|
|
|
|
585 |
|
586 |
colors, normals = [], []
|
587 |
for i in range(6):
|
@@ -604,18 +608,15 @@ def infer_refine(meshes, imgs):
|
|
604 |
colors.append(color)
|
605 |
normals.append(normal)
|
606 |
|
607 |
-
if last_front_color is not None and level == 0:
|
608 |
-
|
609 |
-
else:
|
610 |
-
|
611 |
-
|
612 |
-
|
613 |
-
last_front_color = np.array(colors[0]).astype(np.float32) / 255.0
|
614 |
-
last_front_normal = np.array(normals[0]).astype(np.float32) / 255.0
|
615 |
|
616 |
if last_colors is None:
|
617 |
from copy import deepcopy
|
618 |
-
last_colors
|
619 |
|
620 |
# my mesh flow weight by nearest vertexs
|
621 |
if fixed_v is not None and fixed_f is not None and level == 1:
|
@@ -643,8 +644,8 @@ def infer_refine(meshes, imgs):
|
|
643 |
|
644 |
t = trimesh.Trimesh(vertices=mesh_v, faces=mesh_f)
|
645 |
|
646 |
-
mesh_v = torch.tensor(mesh_v,
|
647 |
-
mesh_f = torch.tensor(mesh_f
|
648 |
|
649 |
new_mesh, simp_v, simp_f = geo_refine(mesh_v, mesh_f, colors, normals, fixed_v=fixed_v, fixed_f=fixed_f, distract_mask=distract_mask, distract_bbox=distract_bbox)
|
650 |
|
@@ -659,22 +660,22 @@ def infer_refine(meshes, imgs):
|
|
659 |
_, idx_anchor = kdtree_anchor.query(new_mesh_v, k=1)
|
660 |
_, idx_mesh_v = kdtree_mesh_v.query(new_mesh_v, k=25)
|
661 |
idx_anchor = idx_anchor.squeeze()
|
662 |
-
neighbors = torch.tensor(new_mesh_v)
|
663 |
# calculate the distances neighbors [V, 25, 3]; new_mesh_v [V, 3] -> [V, 25]
|
664 |
-
neighbor_dists = torch.norm(neighbors - torch.tensor(new_mesh_v)
|
665 |
neighbor_dists[neighbor_dists > 0.06] = 114514.
|
666 |
neighbor_weights = torch.exp(-neighbor_dists * 1.)
|
667 |
neighbor_weights = neighbor_weights / neighbor_weights.sum(dim=1, keepdim=True)
|
668 |
anchors = fixed_v[idx_anchor] # V, 3
|
669 |
anchor_normals = calc_vertex_normals(fixed_v, fixed_f)[idx_anchor] # V, 3
|
670 |
-
dis_anchor = torch.clamp(((anchors - torch.tensor(new_mesh_v)
|
671 |
vec_anchor = dis_anchor[:, None] * anchor_normals # V, 3
|
672 |
vec_anchor = vec_anchor[idx_mesh_v] # V, 25, 3
|
673 |
weighted_vec_anchor = (vec_anchor * neighbor_weights[:, :, None]).sum(1) # V, 3
|
674 |
new_mesh_v += weighted_vec_anchor.cpu().numpy()
|
675 |
|
676 |
# replace new_mesh verts with new_mesh_v
|
677 |
-
new_mesh = Meshes(verts=[torch.tensor(new_mesh_v
|
678 |
|
679 |
except Exception as e:
|
680 |
pass
|
|
|
542 |
return distract_mask, distract_bbox, random_sampled_points, final_mask
|
543 |
|
544 |
|
545 |
+
# infer_refine_sam = sam_model_registry["vit_h"](checkpoint="./ckpt/sam_vit_h_4b8939.pth").cuda()
|
546 |
+
# infer_refine_generator = SamAutomaticMaskGenerator(
|
547 |
+
# model=infer_refine_sam,
|
548 |
+
# points_per_side=64,
|
549 |
+
# pred_iou_thresh=0.80,
|
550 |
+
# stability_score_thresh=0.92,
|
551 |
+
# crop_n_layers=1,
|
552 |
+
# crop_n_points_downscale_factor=2,
|
553 |
+
# min_mask_region_area=100,
|
554 |
+
# )
|
555 |
infer_refine_outside_ratio = 0.20
|
556 |
|
557 |
+
|
558 |
def infer_refine(meshes, imgs):
|
559 |
fixed_v, fixed_f, fixed_t = None, None, None
|
560 |
flow_vert, flow_vector = None, None
|
|
|
564 |
|
565 |
mv, proj = make_star_cameras_orthographic(8, 1, r=1.2)
|
566 |
mv = mv[[4, 3, 2, 0, 6, 5]]
|
|
|
567 |
|
568 |
results = []
|
569 |
|
|
|
575 |
mesh_v, mesh_f = mesh.vertices, mesh.faces
|
576 |
|
577 |
if last_colors is None:
|
578 |
+
@spaces.GPU()
|
579 |
+
def get_mask():
|
580 |
+
renderer = NormalsRenderer(mv,proj,(1024,1024))
|
581 |
+
images = renderer.render(
|
582 |
+
torch.tensor(mesh_v, device='cuda').float(),
|
583 |
+
torch.ones_like(torch.from_numpy(mesh_v), device='cuda').float(),
|
584 |
+
torch.tensor(mesh_f, device='cuda'),
|
585 |
+
)
|
586 |
+
mask = (images[..., 3] < 0.9).cpu().numpy()
|
587 |
+
return mask
|
588 |
+
mask = get_mask()
|
589 |
|
590 |
colors, normals = [], []
|
591 |
for i in range(6):
|
|
|
608 |
colors.append(color)
|
609 |
normals.append(normal)
|
610 |
|
611 |
+
# if last_front_color is not None and level == 0:
|
612 |
+
# original_mask, distract_bbox, _, distract_mask = get_distract_mask(infer_refine_generator, last_front_color, np.array(colors[0]).astype(np.float32) / 255.0, outside_ratio=infer_refine_outside_ratio)
|
613 |
+
# else:
|
614 |
+
distract_mask = None
|
615 |
+
distract_bbox = None
|
|
|
|
|
|
|
616 |
|
617 |
if last_colors is None:
|
618 |
from copy import deepcopy
|
619 |
+
last_colors = deepcopy(colors)
|
620 |
|
621 |
# my mesh flow weight by nearest vertexs
|
622 |
if fixed_v is not None and fixed_f is not None and level == 1:
|
|
|
644 |
|
645 |
t = trimesh.Trimesh(vertices=mesh_v, faces=mesh_f)
|
646 |
|
647 |
+
mesh_v = torch.tensor(mesh_v, dtype=torch.float32)
|
648 |
+
mesh_f = torch.tensor(mesh_f)
|
649 |
|
650 |
new_mesh, simp_v, simp_f = geo_refine(mesh_v, mesh_f, colors, normals, fixed_v=fixed_v, fixed_f=fixed_f, distract_mask=distract_mask, distract_bbox=distract_bbox)
|
651 |
|
|
|
660 |
_, idx_anchor = kdtree_anchor.query(new_mesh_v, k=1)
|
661 |
_, idx_mesh_v = kdtree_mesh_v.query(new_mesh_v, k=25)
|
662 |
idx_anchor = idx_anchor.squeeze()
|
663 |
+
neighbors = torch.tensor(new_mesh_v)[idx_mesh_v] # V, 25, 3
|
664 |
# calculate the distances neighbors [V, 25, 3]; new_mesh_v [V, 3] -> [V, 25]
|
665 |
+
neighbor_dists = torch.norm(neighbors - torch.tensor(new_mesh_v)[:, None], dim=-1)
|
666 |
neighbor_dists[neighbor_dists > 0.06] = 114514.
|
667 |
neighbor_weights = torch.exp(-neighbor_dists * 1.)
|
668 |
neighbor_weights = neighbor_weights / neighbor_weights.sum(dim=1, keepdim=True)
|
669 |
anchors = fixed_v[idx_anchor] # V, 3
|
670 |
anchor_normals = calc_vertex_normals(fixed_v, fixed_f)[idx_anchor] # V, 3
|
671 |
+
dis_anchor = torch.clamp(((anchors - torch.tensor(new_mesh_v)) * anchor_normals).sum(-1), min=0) + 0.01
|
672 |
vec_anchor = dis_anchor[:, None] * anchor_normals # V, 3
|
673 |
vec_anchor = vec_anchor[idx_mesh_v] # V, 25, 3
|
674 |
weighted_vec_anchor = (vec_anchor * neighbor_weights[:, :, None]).sum(1) # V, 3
|
675 |
new_mesh_v += weighted_vec_anchor.cpu().numpy()
|
676 |
|
677 |
# replace new_mesh verts with new_mesh_v
|
678 |
+
new_mesh = Meshes(verts=[torch.tensor(new_mesh_v)], faces=new_mesh.faces_list(), textures=new_mesh.textures)
|
679 |
|
680 |
except Exception as e:
|
681 |
pass
|
refine/mesh_refine.py
CHANGED
@@ -1,4 +1,5 @@
|
|
1 |
import torch
|
|
|
2 |
import numpy as np
|
3 |
import trimesh
|
4 |
from PIL import Image
|
@@ -267,6 +268,15 @@ def run_mesh_refine(vertices, faces, pils: List[Image.Image], fixed_v=None, fixe
|
|
267 |
|
268 |
def geo_refine(mesh_v, mesh_f, rgb_ls, normal_ls, expansion_weight=0.1, fixed_v=None, fixed_f=None,
|
269 |
distract_mask=None, distract_bbox=None, thres=3e-6, no_decompose=False):
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
270 |
rm_normals = simple_remove(normal_ls)
|
271 |
|
272 |
# transfer the alpha channel of rm_normals to img_list
|
@@ -282,6 +292,9 @@ def geo_refine(mesh_v, mesh_f, rgb_ls, normal_ls, expansion_weight=0.1, fixed_v=
|
|
282 |
if no_decompose:
|
283 |
stage1_lr = 0.03
|
284 |
stage1_remesh_interval = 30
|
|
|
|
|
|
|
285 |
|
286 |
vertices, faces = reconstruct_stage1(rm_normals, steps=200, vertices=mesh_v, faces=mesh_f, fixed_v=fixed_v, fixed_f=fixed_f,
|
287 |
lr=stage1_lr, remesh_interval=stage1_remesh_interval, start_edge_len=0.02,
|
@@ -290,20 +303,18 @@ def geo_refine(mesh_v, mesh_f, rgb_ls, normal_ls, expansion_weight=0.1, fixed_v=
|
|
290 |
|
291 |
vertices, faces = run_mesh_refine(vertices, faces, rm_normals, fixed_v=fixed_v, fixed_f=fixed_f, steps=100, start_edge_len=0.005, end_edge_len=0.0002,
|
292 |
decay=0.99, update_normal_interval=20, update_warmup=5, process_inputs=False, process_outputs=False, remesh_interval=1)
|
293 |
-
|
294 |
-
|
|
|
|
|
295 |
simp_vertices, simp_faces = meshes.verts_packed(), meshes.faces_packed()
|
296 |
vertices, faces = simp_vertices.detach().cpu().numpy(), simp_faces.detach().cpu().numpy()
|
|
|
|
|
|
|
297 |
|
298 |
-
|
299 |
-
|
300 |
-
new_mesh = mesh.split(only_watertight=False)
|
301 |
-
|
302 |
-
new_mesh = [ j for j in new_mesh if len(j.vertices) >= 200 ]
|
303 |
-
mesh = trimesh.Scene(new_mesh).dump(concatenate=True)
|
304 |
-
vertices, faces = mesh.vertices.astype('float32'), mesh.faces
|
305 |
-
|
306 |
-
vertices, faces = trimesh.remesh.subdivide(vertices, faces)
|
307 |
origin_len_v, origin_len_f = len(vertices), len(faces)
|
308 |
# concatenate fixed_v and fixed_f
|
309 |
if fixed_v is not None and fixed_f is not None:
|
@@ -316,4 +327,4 @@ def geo_refine(mesh_v, mesh_f, rgb_ls, normal_ls, expansion_weight=0.1, fixed_v=
|
|
316 |
if fixed_v is not None and fixed_f is not None:
|
317 |
new_meshes = Meshes(verts=[new_meshes.verts_packed()[:origin_len_v]], faces=[new_meshes.faces_packed()[:origin_len_f]],
|
318 |
textures=pytorch3d.renderer.mesh.textures.TexturesVertex([new_meshes.textures.verts_features_packed()[:origin_len_v]]))
|
319 |
-
return new_meshes,
|
|
|
1 |
import torch
|
2 |
+
import spaces
|
3 |
import numpy as np
|
4 |
import trimesh
|
5 |
from PIL import Image
|
|
|
268 |
|
269 |
def geo_refine(mesh_v, mesh_f, rgb_ls, normal_ls, expansion_weight=0.1, fixed_v=None, fixed_f=None,
|
270 |
distract_mask=None, distract_bbox=None, thres=3e-6, no_decompose=False):
|
271 |
+
vertices, faces = geo_refine_1(mesh_v, mesh_f, rgb_ls, normal_ls, expansion_weight=expansion_weight, fixed_v=fixed_v, fixed_f=fixed_f,
|
272 |
+
distract_mask=distract_mask, distract_bbox=distract_bbox, thres=thres, no_decompose=no_decompose)
|
273 |
+
vertices, faces = geo_refine_2(vertices, faces, fixed_v=fixed_v)
|
274 |
+
return geo_refine_3(vertices, faces, rgb_ls, fixed_v=fixed_v, fixed_f=fixed_f, distract_mask=distract_mask)
|
275 |
+
|
276 |
+
@spaces.GPU()
|
277 |
+
def geo_refine_1(mesh_v, mesh_f, rgb_ls, normal_ls, expansion_weight=0.1, fixed_v=None, fixed_f=None,
|
278 |
+
distract_mask=None, distract_bbox=None, thres=3e-6, no_decompose=False):
|
279 |
+
|
280 |
rm_normals = simple_remove(normal_ls)
|
281 |
|
282 |
# transfer the alpha channel of rm_normals to img_list
|
|
|
292 |
if no_decompose:
|
293 |
stage1_lr = 0.03
|
294 |
stage1_remesh_interval = 30
|
295 |
+
|
296 |
+
if fixed_v is not None:
|
297 |
+
return mesh_v, mesh_f
|
298 |
|
299 |
vertices, faces = reconstruct_stage1(rm_normals, steps=200, vertices=mesh_v, faces=mesh_f, fixed_v=fixed_v, fixed_f=fixed_f,
|
300 |
lr=stage1_lr, remesh_interval=stage1_remesh_interval, start_edge_len=0.02,
|
|
|
303 |
|
304 |
vertices, faces = run_mesh_refine(vertices, faces, rm_normals, fixed_v=fixed_v, fixed_f=fixed_f, steps=100, start_edge_len=0.005, end_edge_len=0.0002,
|
305 |
decay=0.99, update_normal_interval=20, update_warmup=5, process_inputs=False, process_outputs=False, remesh_interval=1)
|
306 |
+
return vertices, faces
|
307 |
+
|
308 |
+
def geo_refine_2(vertices, faces, fixed_v=None):
|
309 |
+
meshes = simple_clean_mesh(to_pyml_mesh(vertices, faces), apply_smooth=True, stepsmoothnum=2, apply_sub_divide=False, sub_divide_threshold=0.25)
|
310 |
simp_vertices, simp_faces = meshes.verts_packed(), meshes.faces_packed()
|
311 |
vertices, faces = simp_vertices.detach().cpu().numpy(), simp_faces.detach().cpu().numpy()
|
312 |
+
if fixed_v is not None:
|
313 |
+
vertices, faces = trimesh.remesh.subdivide(vertices, faces)
|
314 |
+
return vertices, faces
|
315 |
|
316 |
+
@spaces.GPU()
|
317 |
+
def geo_refine_3(vertices, faces, rgb_ls, fixed_v=None, fixed_f=None, distract_mask=None):
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
318 |
origin_len_v, origin_len_f = len(vertices), len(faces)
|
319 |
# concatenate fixed_v and fixed_f
|
320 |
if fixed_v is not None and fixed_f is not None:
|
|
|
327 |
if fixed_v is not None and fixed_f is not None:
|
328 |
new_meshes = Meshes(verts=[new_meshes.verts_packed()[:origin_len_v]], faces=[new_meshes.faces_packed()[:origin_len_f]],
|
329 |
textures=pytorch3d.renderer.mesh.textures.TexturesVertex([new_meshes.textures.verts_features_packed()[:origin_len_v]]))
|
330 |
+
return new_meshes.to("cpu"), vertices, faces
|