File size: 8,654 Bytes
b213d84
 
16c2627
b213d84
 
 
 
 
24e151d
b213d84
afadbd4
 
16c2627
24e151d
16c2627
afadbd4
b213d84
 
 
 
 
24e151d
 
b213d84
 
 
 
 
 
16c2627
 
 
 
b213d84
 
 
 
 
 
16c2627
 
 
 
b213d84
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
24e151d
 
 
 
 
 
 
 
b213d84
 
 
 
 
 
 
 
24e151d
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
16c2627
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
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
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
import numpy as np
from PIL import Image
from huggingface_hub import snapshot_download
from leffa.transform import LeffaTransform
from leffa.model import LeffaModel
from leffa.inference import LeffaInference
from utils.garment_agnostic_mask_predictor import AutoMasker
from utils.densepose_predictor import DensePosePredictor
from utils.utils import resize_and_center

import gradio as gr

# Download checkpoints
snapshot_download(repo_id="franciszzj/Leffa", local_dir="./ckpts")


def leffa_predict(src_image_path, ref_image_path, control_type):
    assert control_type in [
        "virtual_tryon", "pose_transfer"], "Invalid control type: {}".format(control_type)
    src_image = Image.open(src_image_path)
    ref_image = Image.open(ref_image_path)
    src_image = resize_and_center(src_image, 768, 1024)
    ref_image = resize_and_center(ref_image, 768, 1024)

    src_image_array = np.array(src_image)
    ref_image_array = np.array(ref_image)

    # Mask
    if control_type == "virtual_tryon":
        automasker = AutoMasker(
            densepose_path="./ckpts/densepose",
            schp_path="./ckpts/schp",
        )
        src_image = src_image.convert("RGB")
        mask = automasker(src_image, "upper")["mask"]
    elif control_type == "pose_transfer":
        mask = Image.fromarray(np.ones_like(src_image_array) * 255)

    # DensePose
    densepose_predictor = DensePosePredictor(
        config_path="./ckpts/densepose/densepose_rcnn_R_50_FPN_s1x.yaml",
        weights_path="./ckpts/densepose/model_final_162be9.pkl",
    )
    src_image_iuv_array = densepose_predictor.predict_iuv(src_image_array)
    src_image_seg_array = densepose_predictor.predict_seg(src_image_array)
    src_image_iuv = Image.fromarray(src_image_iuv_array)
    src_image_seg = Image.fromarray(src_image_seg_array)
    if control_type == "virtual_tryon":
        densepose = src_image_seg
    elif control_type == "pose_transfer":
        densepose = src_image_iuv

    # Leffa
    transform = LeffaTransform()
    if control_type == "virtual_tryon":
        pretrained_model_name_or_path = "./ckpts/stable-diffusion-inpainting"
        pretrained_model = "./ckpts/virtual_tryon.pth"
    elif control_type == "pose_transfer":
        pretrained_model_name_or_path = "./ckpts/stable-diffusion-xl-1.0-inpainting-0.1"
        pretrained_model = "./ckpts/pose_transfer.pth"
    model = LeffaModel(
        pretrained_model_name_or_path=pretrained_model_name_or_path,
        pretrained_model=pretrained_model,
    )
    inference = LeffaInference(model=model)

    data = {
        "src_image": [src_image],
        "ref_image": [ref_image],
        "mask": [mask],
        "densepose": [densepose],
    }
    data = transform(data)
    output = inference(data)
    gen_image = output["generated_image"][0]
    # gen_image.save("gen_image.png")
    return np.array(gen_image)


def leffa_predict_vt(src_image_path, ref_image_path):
    return leffa_predict(src_image_path, ref_image_path, "virtual_tryon")


def leffa_predict_pt(src_image_path, ref_image_path):
    return leffa_predict(src_image_path, ref_image_path, "pose_transfer")


if __name__ == "__main__":
    # import sys

    # src_image_path = sys.argv[1]
    # ref_image_path = sys.argv[2]
    # control_type = sys.argv[3]
    # leffa_predict(src_image_path, ref_image_path, control_type)

    title = "## Leffa: Learning Flow Fields in Attention for Controllable Person Image Generation"
    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)."

    with gr.Blocks(theme=gr.themes.Default(primary_hue=gr.themes.colors.pink, secondary_hue=gr.themes.colors.red)).queue() as demo:
        gr.Markdown(title)
        gr.Markdown(description)

        with gr.Tab("Control Appearance (Virtual Try-on)"):
            with gr.Row():
                with gr.Column():
                    gr.Markdown("#### Person Image")
                    vt_src_image = gr.Image(
                        sources=["upload"],
                        type="filepath",
                        label="Person Image",
                        width=512,
                        height=512,
                    )

                    gr.Examples(
                        inputs=vt_src_image,
                        examples_per_page=5,
                        examples=["./ckpts/examples/person1/01320_00.jpg",
                                  "./ckpts/examples/person1/01350_00.jpg",
                                  "./ckpts/examples/person1/01365_00.jpg",
                                  "./ckpts/examples/person1/01376_00.jpg",
                                  "./ckpts/examples/person1/01416_00.jpg",],
                    )

                with gr.Column():
                    gr.Markdown("#### Garment Image")
                    vt_ref_image = gr.Image(
                        sources=["upload"],
                        type="filepath",
                        label="Garment Image",
                        width=512,
                        height=512,
                    )

                    gr.Examples(
                        inputs=vt_ref_image,
                        examples_per_page=5,
                        examples=["./ckpts/examples/garment/01449_00.jpg",
                                  "./ckpts/examples/garment/01486_00.jpg",
                                  "./ckpts/examples/garment/01853_00.jpg",
                                  "./ckpts/examples/garment/02070_00.jpg",
                                  "./ckpts/examples/garment/03553_00.jpg",],
                    )

                with gr.Column():
                    gr.Markdown("#### Generated Image")
                    vt_gen_image = gr.Image(
                        label="Generated Image",
                        width=512,
                        height=512,
                    )

                    with gr.Row():
                        vt_gen_button = gr.Button("Generate")

                vt_gen_button.click(fn=leffa_predict_vt, inputs=[
                    vt_src_image, vt_ref_image], outputs=[vt_gen_image])

        with gr.Tab("Control Pose (Pose Transfer)"):
            with gr.Row():
                with gr.Column():
                    gr.Markdown("#### Person Image")
                    pt_ref_image = gr.Image(
                        sources=["upload"],
                        type="filepath",
                        label="Person Image",
                        width=512,
                        height=512,
                    )

                    gr.Examples(
                        inputs=vt_src_image,
                        examples_per_page=5,
                        examples=["./ckpts/examples/person1/01320_00.jpg",
                                  "./ckpts/examples/person1/01350_00.jpg",
                                  "./ckpts/examples/person1/01365_00.jpg",
                                  "./ckpts/examples/person1/01376_00.jpg",
                                  "./ckpts/examples/person1/01416_00.jpg",],
                    )

                with gr.Column():
                    gr.Markdown("#### Target Pose Person Image")
                    pt_src_image = gr.Image(
                        sources=["upload"],
                        type="filepath",
                        label="Target Pose Person Image",
                        width=512,
                        height=512,
                    )

                    gr.Examples(
                        inputs=pt_src_image,
                        examples_per_page=5,
                        examples=["./ckpts/examples/person2/01850_00.jpg",
                                  "./ckpts/examples/person2/01875_00.jpg",
                                  "./ckpts/examples/person2/02532_00.jpg",
                                  "./ckpts/examples/person2/02902_00.jpg",
                                  "./ckpts/examples/person2/05346_00.jpg",],
                    )

                with gr.Column():
                    gr.Markdown("#### Generated Image")
                    pt_gen_image = gr.Image(
                        label="Generated Image",
                        width=512,
                        height=512,
                    )

                    with gr.Row():
                        pose_transfer_gen_button = gr.Button("Generate")

                pose_transfer_gen_button.click(fn=leffa_predict_pt, inputs=[
                    pt_src_image, pt_ref_image], outputs=[pt_gen_image])

        demo.launch(share=True, server_port=7860)