franciszzj
commited on
Commit
•
c81c28f
1
Parent(s):
9ed5c4d
load model before predict
Browse files- app.py +27 -20
- leffa/inference.py +1 -2
app.py
CHANGED
@@ -13,6 +13,28 @@ import gradio as gr
|
|
13 |
# Download checkpoints
|
14 |
snapshot_download(repo_id="franciszzj/Leffa", local_dir="./ckpts")
|
15 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
16 |
|
17 |
def leffa_predict(src_image_path, ref_image_path, control_type):
|
18 |
assert control_type in [
|
@@ -27,20 +49,12 @@ def leffa_predict(src_image_path, ref_image_path, control_type):
|
|
27 |
|
28 |
# Mask
|
29 |
if control_type == "virtual_tryon":
|
30 |
-
automasker = AutoMasker(
|
31 |
-
densepose_path="./ckpts/densepose",
|
32 |
-
schp_path="./ckpts/schp",
|
33 |
-
)
|
34 |
src_image = src_image.convert("RGB")
|
35 |
-
mask =
|
36 |
elif control_type == "pose_transfer":
|
37 |
mask = Image.fromarray(np.ones_like(src_image_array) * 255)
|
38 |
|
39 |
# DensePose
|
40 |
-
densepose_predictor = DensePosePredictor(
|
41 |
-
config_path="./ckpts/densepose/densepose_rcnn_R_50_FPN_s1x.yaml",
|
42 |
-
weights_path="./ckpts/densepose/model_final_162be9.pkl",
|
43 |
-
)
|
44 |
src_image_iuv_array = densepose_predictor.predict_iuv(src_image_array)
|
45 |
src_image_seg_array = densepose_predictor.predict_seg(src_image_array)
|
46 |
src_image_iuv = Image.fromarray(src_image_iuv_array)
|
@@ -52,17 +66,6 @@ def leffa_predict(src_image_path, ref_image_path, control_type):
|
|
52 |
|
53 |
# Leffa
|
54 |
transform = LeffaTransform()
|
55 |
-
if control_type == "virtual_tryon":
|
56 |
-
pretrained_model_name_or_path = "./ckpts/stable-diffusion-inpainting"
|
57 |
-
pretrained_model = "./ckpts/virtual_tryon.pth"
|
58 |
-
elif control_type == "pose_transfer":
|
59 |
-
pretrained_model_name_or_path = "./ckpts/stable-diffusion-xl-1.0-inpainting-0.1"
|
60 |
-
pretrained_model = "./ckpts/pose_transfer.pth"
|
61 |
-
model = LeffaModel(
|
62 |
-
pretrained_model_name_or_path=pretrained_model_name_or_path,
|
63 |
-
pretrained_model=pretrained_model,
|
64 |
-
)
|
65 |
-
inference = LeffaInference(model=model)
|
66 |
|
67 |
data = {
|
68 |
"src_image": [src_image],
|
@@ -71,6 +74,10 @@ def leffa_predict(src_image_path, ref_image_path, control_type):
|
|
71 |
"densepose": [densepose],
|
72 |
}
|
73 |
data = transform(data)
|
|
|
|
|
|
|
|
|
74 |
output = inference(data)
|
75 |
gen_image = output["generated_image"][0]
|
76 |
# gen_image.save("gen_image.png")
|
|
|
13 |
# Download checkpoints
|
14 |
snapshot_download(repo_id="franciszzj/Leffa", local_dir="./ckpts")
|
15 |
|
16 |
+
mask_predictor = AutoMasker(
|
17 |
+
densepose_path="./ckpts/densepose",
|
18 |
+
schp_path="./ckpts/schp",
|
19 |
+
)
|
20 |
+
|
21 |
+
densepose_predictor = DensePosePredictor(
|
22 |
+
config_path="./ckpts/densepose/densepose_rcnn_R_50_FPN_s1x.yaml",
|
23 |
+
weights_path="./ckpts/densepose/model_final_162be9.pkl",
|
24 |
+
)
|
25 |
+
|
26 |
+
vt_model = LeffaModel(
|
27 |
+
pretrained_model_name_or_path="./ckpts/stable-diffusion-inpainting",
|
28 |
+
pretrained_model="./ckpts/virtual_tryon.pth",
|
29 |
+
)
|
30 |
+
vt_inference = LeffaInference(model=vt_model)
|
31 |
+
|
32 |
+
pt_model = LeffaModel(
|
33 |
+
pretrained_model_name_or_path="./ckpts/stable-diffusion-xl-1.0-inpainting-0.1",
|
34 |
+
pretrained_model="./ckpts/pose_transfer.pth",
|
35 |
+
)
|
36 |
+
pt_inference = LeffaInference(model=pt_model)
|
37 |
+
|
38 |
|
39 |
def leffa_predict(src_image_path, ref_image_path, control_type):
|
40 |
assert control_type in [
|
|
|
49 |
|
50 |
# Mask
|
51 |
if control_type == "virtual_tryon":
|
|
|
|
|
|
|
|
|
52 |
src_image = src_image.convert("RGB")
|
53 |
+
mask = mask_predictor(src_image, "upper")["mask"]
|
54 |
elif control_type == "pose_transfer":
|
55 |
mask = Image.fromarray(np.ones_like(src_image_array) * 255)
|
56 |
|
57 |
# DensePose
|
|
|
|
|
|
|
|
|
58 |
src_image_iuv_array = densepose_predictor.predict_iuv(src_image_array)
|
59 |
src_image_seg_array = densepose_predictor.predict_seg(src_image_array)
|
60 |
src_image_iuv = Image.fromarray(src_image_iuv_array)
|
|
|
66 |
|
67 |
# Leffa
|
68 |
transform = LeffaTransform()
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
69 |
|
70 |
data = {
|
71 |
"src_image": [src_image],
|
|
|
74 |
"densepose": [densepose],
|
75 |
}
|
76 |
data = transform(data)
|
77 |
+
if control_type == "virtual_tryon":
|
78 |
+
inference = vt_inference
|
79 |
+
elif control_type == "pose_transfer":
|
80 |
+
inference = pt_inference
|
81 |
output = inference(data)
|
82 |
gen_image = output["generated_image"][0]
|
83 |
# gen_image.save("gen_image.png")
|
leffa/inference.py
CHANGED
@@ -17,7 +17,6 @@ class LeffaInference(object):
|
|
17 |
self,
|
18 |
model: nn.Module,
|
19 |
ckpt_path: Optional[str] = None,
|
20 |
-
repaint: bool = False,
|
21 |
) -> None:
|
22 |
self.model: torch.nn.Module = model
|
23 |
self.device = "cuda" if torch.cuda.is_available() else "cpu"
|
@@ -28,7 +27,7 @@ class LeffaInference(object):
|
|
28 |
self.model = self.model.to(self.device)
|
29 |
self.model.eval()
|
30 |
|
31 |
-
self.pipe = LeffaPipeline(model=self.model
|
32 |
|
33 |
def to_gpu(self, data: Dict[str, Any]) -> Dict[str, Any]:
|
34 |
for k, v in data.items():
|
|
|
17 |
self,
|
18 |
model: nn.Module,
|
19 |
ckpt_path: Optional[str] = None,
|
|
|
20 |
) -> None:
|
21 |
self.model: torch.nn.Module = model
|
22 |
self.device = "cuda" if torch.cuda.is_available() else "cpu"
|
|
|
27 |
self.model = self.model.to(self.device)
|
28 |
self.model.eval()
|
29 |
|
30 |
+
self.pipe = LeffaPipeline(model=self.model)
|
31 |
|
32 |
def to_gpu(self, data: Dict[str, Any]) -> Dict[str, Any]:
|
33 |
for k, v in data.items():
|