Spaces:
Running
on
L40S
Running
on
L40S
franciszzj
commited on
Commit
β’
68f6086
1
Parent(s):
9b1ec91
update app
Browse files- app.py +151 -108
- utils/utils.py +12 -0
app.py
CHANGED
@@ -6,90 +6,111 @@ from leffa.model import LeffaModel
|
|
6 |
from leffa.inference import LeffaInference
|
7 |
from utils.garment_agnostic_mask_predictor import AutoMasker
|
8 |
from utils.densepose_predictor import DensePosePredictor
|
9 |
-
from utils.utils import resize_and_center
|
10 |
|
11 |
import gradio as gr
|
12 |
|
13 |
# Download checkpoints
|
14 |
snapshot_download(repo_id="franciszzj/Leffa", local_dir="./ckpts")
|
15 |
|
16 |
-
|
17 |
-
|
18 |
-
|
19 |
-
|
20 |
-
|
21 |
-
|
22 |
-
|
23 |
-
|
24 |
-
|
25 |
-
|
26 |
-
|
27 |
-
|
28 |
-
|
29 |
-
|
30 |
-
|
31 |
-
|
32 |
-
|
33 |
-
|
34 |
-
|
35 |
-
|
36 |
-
|
37 |
-
|
38 |
-
|
39 |
-
|
40 |
-
|
41 |
-
|
42 |
-
|
43 |
-
|
44 |
-
|
45 |
-
|
46 |
-
|
47 |
-
|
48 |
-
|
49 |
-
|
50 |
-
|
51 |
-
|
52 |
-
|
53 |
-
|
54 |
-
|
55 |
-
|
56 |
-
|
57 |
-
|
58 |
-
|
59 |
-
|
60 |
-
|
61 |
-
|
62 |
-
|
63 |
-
|
64 |
-
|
65 |
-
|
66 |
-
|
67 |
-
|
68 |
-
|
69 |
-
|
70 |
-
|
71 |
-
|
72 |
-
|
73 |
-
|
74 |
-
"
|
75 |
-
|
76 |
-
|
77 |
-
|
78 |
-
|
79 |
-
|
80 |
-
|
81 |
-
|
82 |
-
|
83 |
-
|
84 |
-
|
85 |
-
|
86 |
-
|
87 |
-
|
88 |
-
|
89 |
-
|
90 |
-
|
91 |
-
|
92 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
93 |
|
94 |
|
95 |
if __name__ == "__main__":
|
@@ -100,14 +121,26 @@ if __name__ == "__main__":
|
|
100 |
# control_type = sys.argv[3]
|
101 |
# leffa_predict(src_image_path, ref_image_path, control_type)
|
102 |
|
|
|
|
|
|
|
|
|
|
|
|
|
103 |
title = "## Leffa: Learning Flow Fields in Attention for Controllable Person Image Generation"
|
104 |
-
link = "[π Paper](https://arxiv.org/abs/2412.08486) - [π₯ Demo](https://huggingface.co/spaces/franciszzj/Leffa) - [π€ Model](https://huggingface.co/franciszzj/Leffa)"
|
|
|
|
|
|
|
|
|
|
|
105 |
description = "Leffa is a unified framework for controllable person image generation that enables precise manipulation of both appearance (i.e., virtual try-on) and pose (i.e., pose transfer)."
|
106 |
-
note = "Note: The models used in the demo are trained solely on academic datasets. Virtual try-on uses VITON-HD, and pose transfer uses DeepFashion."
|
107 |
|
108 |
with gr.Blocks(theme=gr.themes.Default(primary_hue=gr.themes.colors.pink, secondary_hue=gr.themes.colors.red)).queue() as demo:
|
109 |
gr.Markdown(title)
|
110 |
gr.Markdown(link)
|
|
|
111 |
gr.Markdown(description)
|
112 |
|
113 |
with gr.Tab("Control Appearance (Virtual Try-on)"):
|
@@ -124,12 +157,8 @@ if __name__ == "__main__":
|
|
124 |
|
125 |
gr.Examples(
|
126 |
inputs=vt_src_image,
|
127 |
-
examples_per_page=
|
128 |
-
examples=
|
129 |
-
"./ckpts/examples/person1/01376_00.jpg",
|
130 |
-
"./ckpts/examples/person1/01416_00.jpg",
|
131 |
-
"./ckpts/examples/person1/05976_00.jpg",
|
132 |
-
"./ckpts/examples/person1/06094_00.jpg",],
|
133 |
)
|
134 |
|
135 |
with gr.Column():
|
@@ -144,12 +173,8 @@ if __name__ == "__main__":
|
|
144 |
|
145 |
gr.Examples(
|
146 |
inputs=vt_ref_image,
|
147 |
-
examples_per_page=
|
148 |
-
examples=
|
149 |
-
"./ckpts/examples/garment/01486_00.jpg",
|
150 |
-
"./ckpts/examples/garment/01853_00.jpg",
|
151 |
-
"./ckpts/examples/garment/02070_00.jpg",
|
152 |
-
"./ckpts/examples/garment/03553_00.jpg",],
|
153 |
)
|
154 |
|
155 |
with gr.Column():
|
@@ -163,8 +188,24 @@ if __name__ == "__main__":
|
|
163 |
with gr.Row():
|
164 |
vt_gen_button = gr.Button("Generate")
|
165 |
|
166 |
-
|
167 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
168 |
|
169 |
with gr.Tab("Control Pose (Pose Transfer)"):
|
170 |
with gr.Row():
|
@@ -180,12 +221,8 @@ if __name__ == "__main__":
|
|
180 |
|
181 |
gr.Examples(
|
182 |
inputs=pt_ref_image,
|
183 |
-
examples_per_page=
|
184 |
-
examples=
|
185 |
-
"./ckpts/examples/person1/01376_00.jpg",
|
186 |
-
"./ckpts/examples/person1/01416_00.jpg",
|
187 |
-
"./ckpts/examples/person1/05976_00.jpg",
|
188 |
-
"./ckpts/examples/person1/06094_00.jpg",],
|
189 |
)
|
190 |
|
191 |
with gr.Column():
|
@@ -200,12 +237,8 @@ if __name__ == "__main__":
|
|
200 |
|
201 |
gr.Examples(
|
202 |
inputs=pt_src_image,
|
203 |
-
examples_per_page=
|
204 |
-
examples=
|
205 |
-
"./ckpts/examples/person2/01875_00.jpg",
|
206 |
-
"./ckpts/examples/person2/02532_00.jpg",
|
207 |
-
"./ckpts/examples/person2/02902_00.jpg",
|
208 |
-
"./ckpts/examples/person2/05346_00.jpg",],
|
209 |
)
|
210 |
|
211 |
with gr.Column():
|
@@ -219,8 +252,18 @@ if __name__ == "__main__":
|
|
219 |
with gr.Row():
|
220 |
pose_transfer_gen_button = gr.Button("Generate")
|
221 |
|
222 |
-
|
223 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
224 |
|
225 |
gr.Markdown(note)
|
226 |
|
|
|
6 |
from leffa.inference import LeffaInference
|
7 |
from utils.garment_agnostic_mask_predictor import AutoMasker
|
8 |
from utils.densepose_predictor import DensePosePredictor
|
9 |
+
from utils.utils import resize_and_center, list_dir
|
10 |
|
11 |
import gradio as gr
|
12 |
|
13 |
# Download checkpoints
|
14 |
snapshot_download(repo_id="franciszzj/Leffa", local_dir="./ckpts")
|
15 |
|
16 |
+
|
17 |
+
class LeffaPredictor(object):
|
18 |
+
def __init__(self):
|
19 |
+
self.mask_predictor = AutoMasker(
|
20 |
+
densepose_path="./ckpts/densepose",
|
21 |
+
schp_path="./ckpts/schp",
|
22 |
+
)
|
23 |
+
|
24 |
+
self.densepose_predictor = DensePosePredictor(
|
25 |
+
config_path="./ckpts/densepose/densepose_rcnn_R_50_FPN_s1x.yaml",
|
26 |
+
weights_path="./ckpts/densepose/model_final_162be9.pkl",
|
27 |
+
)
|
28 |
+
|
29 |
+
vt_model = LeffaModel(
|
30 |
+
pretrained_model_name_or_path="./ckpts/stable-diffusion-inpainting",
|
31 |
+
pretrained_model="./ckpts/virtual_tryon.pth",
|
32 |
+
)
|
33 |
+
self.vt_inference = LeffaInference(model=vt_model)
|
34 |
+
self.vt_model_type = "viton_hd"
|
35 |
+
|
36 |
+
pt_model = LeffaModel(
|
37 |
+
pretrained_model_name_or_path="./ckpts/stable-diffusion-xl-1.0-inpainting-0.1",
|
38 |
+
pretrained_model="./ckpts/pose_transfer.pth",
|
39 |
+
)
|
40 |
+
self.pt_inference = LeffaInference(model=pt_model)
|
41 |
+
|
42 |
+
def change_vt_model(self, vt_model_type):
|
43 |
+
if vt_model_type == self.vt_model_type:
|
44 |
+
return
|
45 |
+
if vt_model_type == "viton_hd":
|
46 |
+
pretrained_model = "./ckpts/virtual_tryon.pth"
|
47 |
+
elif vt_model_type == "dress_code":
|
48 |
+
pretrained_model = "./ckpts/virtual_tryon_dc.pth"
|
49 |
+
vt_model = LeffaModel(
|
50 |
+
pretrained_model_name_or_path="./ckpts/stable-diffusion-inpainting",
|
51 |
+
pretrained_model=pretrained_model,
|
52 |
+
)
|
53 |
+
self.vt_inference = LeffaInference(model=vt_model)
|
54 |
+
self.vt_model_type = vt_model_type
|
55 |
+
|
56 |
+
def leffa_predict(self, src_image_path, ref_image_path, control_type, step=50, scale=2.5, seed=42):
|
57 |
+
assert control_type in [
|
58 |
+
"virtual_tryon", "pose_transfer"], "Invalid control type: {}".format(control_type)
|
59 |
+
src_image = Image.open(src_image_path)
|
60 |
+
ref_image = Image.open(ref_image_path)
|
61 |
+
src_image = resize_and_center(src_image, 768, 1024)
|
62 |
+
ref_image = resize_and_center(ref_image, 768, 1024)
|
63 |
+
|
64 |
+
src_image_array = np.array(src_image)
|
65 |
+
|
66 |
+
# Mask
|
67 |
+
if control_type == "virtual_tryon":
|
68 |
+
src_image = src_image.convert("RGB")
|
69 |
+
mask = self.mask_predictor(src_image, "upper")["mask"]
|
70 |
+
elif control_type == "pose_transfer":
|
71 |
+
mask = Image.fromarray(np.ones_like(src_image_array) * 255)
|
72 |
+
|
73 |
+
# DensePose
|
74 |
+
if control_type == "virtual_tryon":
|
75 |
+
src_image_seg_array = self.densepose_predictor.predict_seg(
|
76 |
+
src_image_array)
|
77 |
+
src_image_seg = Image.fromarray(src_image_seg_array)
|
78 |
+
densepose = src_image_seg
|
79 |
+
elif control_type == "pose_transfer":
|
80 |
+
src_image_iuv_array = self.densepose_predictor.predict_iuv(
|
81 |
+
src_image_array)
|
82 |
+
src_image_iuv = Image.fromarray(src_image_iuv_array)
|
83 |
+
densepose = src_image_iuv
|
84 |
+
|
85 |
+
# Leffa
|
86 |
+
transform = LeffaTransform()
|
87 |
+
|
88 |
+
data = {
|
89 |
+
"src_image": [src_image],
|
90 |
+
"ref_image": [ref_image],
|
91 |
+
"mask": [mask],
|
92 |
+
"densepose": [densepose],
|
93 |
+
}
|
94 |
+
data = transform(data)
|
95 |
+
if control_type == "virtual_tryon":
|
96 |
+
inference = self.vt_inference
|
97 |
+
elif control_type == "pose_transfer":
|
98 |
+
inference = self.pt_inference
|
99 |
+
output = inference(
|
100 |
+
data,
|
101 |
+
num_inference_steps=step,
|
102 |
+
guidance_scale=scale,
|
103 |
+
seed=seed,)
|
104 |
+
gen_image = output["generated_image"][0]
|
105 |
+
# gen_image.save("gen_image.png")
|
106 |
+
return np.array(gen_image)
|
107 |
+
|
108 |
+
def leffa_predict_vt(self, src_image_path, ref_image_path, step, scale, seed, vt_model_type="viton_hd"):
|
109 |
+
self.change_vt_model(vt_model_type)
|
110 |
+
return self.leffa_predict(src_image_path, ref_image_path, "virtual_tryon", step, scale, seed)
|
111 |
+
|
112 |
+
def leffa_predict_pt(self, src_image_path, ref_image_path, step, scale, seed):
|
113 |
+
return self.leffa_predict(src_image_path, ref_image_path, "pose_transfer", step, scale, seed)
|
114 |
|
115 |
|
116 |
if __name__ == "__main__":
|
|
|
121 |
# control_type = sys.argv[3]
|
122 |
# leffa_predict(src_image_path, ref_image_path, control_type)
|
123 |
|
124 |
+
leffa_predictor = LeffaPredictor()
|
125 |
+
example_dir = "./ckpts/examples"
|
126 |
+
person1_images = list_dir(f"{example_dir}/person1")
|
127 |
+
person2_images = list_dir(f"{example_dir}/person2")
|
128 |
+
garment_images = list_dir(f"{example_dir}/garment")
|
129 |
+
|
130 |
title = "## Leffa: Learning Flow Fields in Attention for Controllable Person Image Generation"
|
131 |
+
link = "[π Paper](https://arxiv.org/abs/2412.08486) - [π€ Code](https://github.com/franciszzj/Leffa) - [π₯ Demo](https://huggingface.co/spaces/franciszzj/Leffa) - [π€ Model](https://huggingface.co/franciszzj/Leffa)"
|
132 |
+
news = """## News
|
133 |
+
- 16/Dec/2024, the virtual try-on [model](https://huggingface.co/franciszzj/Leffa/blob/main/virtual_tryon_dc.pth) trained on DressCode is released.
|
134 |
+
- 12/Dec/2024, the HuggingFace [demo](https://huggingface.co/spaces/franciszzj/Leffa) and [models](https://huggingface.co/franciszzj/Leffa) (virtual try-on model trained on VITON-HD and pose transfer model trained on DeepFashion) are released.
|
135 |
+
- 11/Dec/2024, the [arXiv](https://arxiv.org/abs/2412.08486) version of the paper is released.
|
136 |
+
"""
|
137 |
description = "Leffa is a unified framework for controllable person image generation that enables precise manipulation of both appearance (i.e., virtual try-on) and pose (i.e., pose transfer)."
|
138 |
+
note = "Note: The models used in the demo are trained solely on academic datasets. Virtual try-on uses VITON-HD/DressCode, and pose transfer uses DeepFashion."
|
139 |
|
140 |
with gr.Blocks(theme=gr.themes.Default(primary_hue=gr.themes.colors.pink, secondary_hue=gr.themes.colors.red)).queue() as demo:
|
141 |
gr.Markdown(title)
|
142 |
gr.Markdown(link)
|
143 |
+
gr.Markdown(news)
|
144 |
gr.Markdown(description)
|
145 |
|
146 |
with gr.Tab("Control Appearance (Virtual Try-on)"):
|
|
|
157 |
|
158 |
gr.Examples(
|
159 |
inputs=vt_src_image,
|
160 |
+
examples_per_page=10,
|
161 |
+
examples=person1_images,
|
|
|
|
|
|
|
|
|
162 |
)
|
163 |
|
164 |
with gr.Column():
|
|
|
173 |
|
174 |
gr.Examples(
|
175 |
inputs=vt_ref_image,
|
176 |
+
examples_per_page=10,
|
177 |
+
examples=garment_images,
|
|
|
|
|
|
|
|
|
178 |
)
|
179 |
|
180 |
with gr.Column():
|
|
|
188 |
with gr.Row():
|
189 |
vt_gen_button = gr.Button("Generate")
|
190 |
|
191 |
+
with gr.Accordion("Advanced Options", open=False):
|
192 |
+
vt_step = gr.Number(
|
193 |
+
label="Inference Steps", minimum=30, maximum=100, step=1, value=50)
|
194 |
+
|
195 |
+
vt_scale = gr.Number(
|
196 |
+
label="Guidance Scale", minimum=0.1, maximum=5.0, step=0.1, value=2.5)
|
197 |
+
|
198 |
+
vt_seed = gr.Number(
|
199 |
+
label="Random Seed", minimum=-1, maximum=2147483647, step=1, value=42)
|
200 |
+
|
201 |
+
vt_model_type = gr.Radio(
|
202 |
+
choices=["viton_hd", "dress_code"],
|
203 |
+
value="viton_hd",
|
204 |
+
label="Model Type",
|
205 |
+
)
|
206 |
+
|
207 |
+
vt_gen_button.click(fn=leffa_predictor.leffa_predict_vt, inputs=[
|
208 |
+
vt_src_image, vt_ref_image, vt_step, vt_scale, vt_seed, vt_model_type], outputs=[vt_gen_image])
|
209 |
|
210 |
with gr.Tab("Control Pose (Pose Transfer)"):
|
211 |
with gr.Row():
|
|
|
221 |
|
222 |
gr.Examples(
|
223 |
inputs=pt_ref_image,
|
224 |
+
examples_per_page=10,
|
225 |
+
examples=person1_images,
|
|
|
|
|
|
|
|
|
226 |
)
|
227 |
|
228 |
with gr.Column():
|
|
|
237 |
|
238 |
gr.Examples(
|
239 |
inputs=pt_src_image,
|
240 |
+
examples_per_page=10,
|
241 |
+
examples=person2_images,
|
|
|
|
|
|
|
|
|
242 |
)
|
243 |
|
244 |
with gr.Column():
|
|
|
252 |
with gr.Row():
|
253 |
pose_transfer_gen_button = gr.Button("Generate")
|
254 |
|
255 |
+
with gr.Accordion("Advanced Options", open=False):
|
256 |
+
pt_step = gr.Number(
|
257 |
+
label="Inference Steps", minimum=30, maximum=100, step=1, value=50)
|
258 |
+
|
259 |
+
pt_scale = gr.Number(
|
260 |
+
label="Guidance Scale", minimum=0.1, maximum=5.0, step=0.1, value=2.5)
|
261 |
+
|
262 |
+
pt_seed = gr.Number(
|
263 |
+
label="Random Seed", minimum=-1, maximum=2147483647, step=1, value=42)
|
264 |
+
|
265 |
+
pose_transfer_gen_button.click(fn=leffa_predictor.leffa_predict_pt, inputs=[
|
266 |
+
pt_src_image, pt_ref_image, pt_step, pt_scale, pt_seed], outputs=[pt_gen_image])
|
267 |
|
268 |
gr.Markdown(note)
|
269 |
|
utils/utils.py
CHANGED
@@ -1,3 +1,4 @@
|
|
|
|
1 |
import cv2
|
2 |
import numpy as np
|
3 |
from PIL import Image
|
@@ -29,3 +30,14 @@ def resize_and_center(image, target_width, target_height):
|
|
29 |
padded_img[top:top + new_height, left:left + new_width] = resized_img
|
30 |
|
31 |
return Image.fromarray(padded_img)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import os
|
2 |
import cv2
|
3 |
import numpy as np
|
4 |
from PIL import Image
|
|
|
30 |
padded_img[top:top + new_height, left:left + new_width] = resized_img
|
31 |
|
32 |
return Image.fromarray(padded_img)
|
33 |
+
|
34 |
+
|
35 |
+
def list_dir(folder_path):
|
36 |
+
# Collect all file paths within the directory
|
37 |
+
file_paths = []
|
38 |
+
for root, _, files in os.walk(folder_path):
|
39 |
+
for file in files:
|
40 |
+
file_paths.append(os.path.join(root, file))
|
41 |
+
|
42 |
+
file_paths = sorted(file_paths)
|
43 |
+
return file_paths
|