Spaces:
Running
on
Zero
Running
on
Zero
bug-fix
Browse files- inference_utils.py +18 -20
inference_utils.py
CHANGED
@@ -1,5 +1,6 @@
|
|
1 |
import os
|
2 |
-
import torch
|
|
|
3 |
|
4 |
seed = 1024
|
5 |
random.seed(seed)
|
@@ -10,12 +11,12 @@ torch.backends.cudnn.deterministic = True
|
|
10 |
torch.backends.cudnn.benchmark = False
|
11 |
|
12 |
# SPIGA ckpt downloading always fails, so we download it manually and put it in the right place.
|
13 |
-
import
|
14 |
from gdown import download
|
15 |
|
16 |
-
|
17 |
spiga_file_id = "1YrbScfMzrAAWMJQYgxdLZ9l57nmTdpQC"
|
18 |
-
ckpt_path = os.path.join(
|
19 |
if not os.path.exists(ckpt_path):
|
20 |
os.makedirs(os.path.dirname(ckpt_path), exist_ok=True)
|
21 |
download(id=spiga_file_id, output=ckpt_path)
|
@@ -30,7 +31,6 @@ from diffusers import DDIMScheduler, ControlNetModel
|
|
30 |
from diffusers import UNet2DConditionModel as OriginalUNet2DConditionModel
|
31 |
from detail_encoder.encoder_plus import detail_encoder
|
32 |
|
33 |
-
|
34 |
detector = FaceDetector(weight_path="./models/mobilenet0.25_Final.pth")
|
35 |
|
36 |
|
@@ -64,21 +64,21 @@ def concatenate_images(image_files, output_file):
|
|
64 |
|
65 |
def init_pipeline():
|
66 |
# Initialize the model
|
67 |
-
model_id
|
68 |
base_path = "./checkpoints/stablemakeup"
|
69 |
folder_id = "1397t27GrUyLPnj17qVpKWGwg93EcaFfg"
|
70 |
if not os.path.exists(base_path):
|
71 |
download_folder(id=folder_id, output=base_path, quiet=False, use_cookies=False)
|
72 |
makeup_encoder_path = base_path + "/pytorch_model.bin"
|
73 |
-
id_encoder_path
|
74 |
-
pose_encoder_path
|
75 |
-
|
76 |
-
Unet
|
77 |
-
id_encoder
|
78 |
-
pose_encoder
|
79 |
-
makeup_encoder
|
80 |
-
id_state_dict
|
81 |
-
pose_state_dict
|
82 |
makeup_state_dict = torch.load(makeup_encoder_path)
|
83 |
id_encoder.load_state_dict(id_state_dict, strict=False)
|
84 |
pose_encoder.load_state_dict(pose_state_dict, strict=False)
|
@@ -99,10 +99,8 @@ pipeline, makeup_encoder = init_pipeline()
|
|
99 |
|
100 |
|
101 |
def inference(id_image_pil, makeup_image_pil, guidance_scale=1.6, size=512):
|
102 |
-
id_image
|
103 |
makeup_image = makeup_image_pil.resize((size, size))
|
104 |
-
pose_image
|
105 |
-
result_img
|
106 |
-
id_image=[id_image, pose_image], makeup_image=makeup_image, pipe=pipeline, guidance_scale=guidance_scale
|
107 |
-
)
|
108 |
return result_img
|
|
|
1 |
import os
|
2 |
+
import torch
|
3 |
+
import random
|
4 |
|
5 |
seed = 1024
|
6 |
random.seed(seed)
|
|
|
11 |
torch.backends.cudnn.benchmark = False
|
12 |
|
13 |
# SPIGA ckpt downloading always fails, so we download it manually and put it in the right place.
|
14 |
+
import spiga
|
15 |
from gdown import download
|
16 |
|
17 |
+
pkg_path = spiga.__file__.replace("/__init__.py", "")
|
18 |
spiga_file_id = "1YrbScfMzrAAWMJQYgxdLZ9l57nmTdpQC"
|
19 |
+
ckpt_path = os.path.join(pkg_path, "spiga/models/weights/spiga_300wpublic.pt")
|
20 |
if not os.path.exists(ckpt_path):
|
21 |
os.makedirs(os.path.dirname(ckpt_path), exist_ok=True)
|
22 |
download(id=spiga_file_id, output=ckpt_path)
|
|
|
31 |
from diffusers import UNet2DConditionModel as OriginalUNet2DConditionModel
|
32 |
from detail_encoder.encoder_plus import detail_encoder
|
33 |
|
|
|
34 |
detector = FaceDetector(weight_path="./models/mobilenet0.25_Final.pth")
|
35 |
|
36 |
|
|
|
64 |
|
65 |
def init_pipeline():
|
66 |
# Initialize the model
|
67 |
+
model_id = "runwayml/stable-diffusion-v1-5" # or your local sdv1-5 path
|
68 |
base_path = "./checkpoints/stablemakeup"
|
69 |
folder_id = "1397t27GrUyLPnj17qVpKWGwg93EcaFfg"
|
70 |
if not os.path.exists(base_path):
|
71 |
download_folder(id=folder_id, output=base_path, quiet=False, use_cookies=False)
|
72 |
makeup_encoder_path = base_path + "/pytorch_model.bin"
|
73 |
+
id_encoder_path = base_path + "/pytorch_model_1.bin"
|
74 |
+
pose_encoder_path = base_path + "/pytorch_model_2.bin"
|
75 |
+
|
76 |
+
Unet = OriginalUNet2DConditionModel.from_pretrained(model_id, subfolder="unet").to("cuda")
|
77 |
+
id_encoder = ControlNetModel.from_unet(Unet)
|
78 |
+
pose_encoder = ControlNetModel.from_unet(Unet)
|
79 |
+
makeup_encoder = detail_encoder(Unet, "openai/clip-vit-large-patch14", "cuda", dtype=torch.float32)
|
80 |
+
id_state_dict = torch.load(id_encoder_path)
|
81 |
+
pose_state_dict = torch.load(pose_encoder_path)
|
82 |
makeup_state_dict = torch.load(makeup_encoder_path)
|
83 |
id_encoder.load_state_dict(id_state_dict, strict=False)
|
84 |
pose_encoder.load_state_dict(pose_state_dict, strict=False)
|
|
|
99 |
|
100 |
|
101 |
def inference(id_image_pil, makeup_image_pil, guidance_scale=1.6, size=512):
|
102 |
+
id_image = id_image_pil.resize((size, size))
|
103 |
makeup_image = makeup_image_pil.resize((size, size))
|
104 |
+
pose_image = get_draw(id_image, size=size)
|
105 |
+
result_img = makeup_encoder.generate(id_image=[id_image, pose_image], makeup_image=makeup_image, pipe=pipeline, guidance_scale=guidance_scale)
|
|
|
|
|
106 |
return result_img
|