Spaces:
Running
on
Zero
Running
on
Zero
File size: 5,158 Bytes
74a242e |
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 |
from typing import TypedDict
import diffusers.image_processor
import gradio as gr
import pillow_heif # pyright: ignore[reportMissingTypeStubs]
import spaces # pyright: ignore[reportMissingTypeStubs]
import torch
from PIL import Image
from pipeline import TryOffAnyone
pillow_heif.register_heif_opener() # pyright: ignore[reportUnknownMemberType]
pillow_heif.register_avif_opener() # pyright: ignore[reportUnknownMemberType]
torch.set_float32_matmul_precision("high")
torch.backends.cuda.matmul.allow_tf32 = True
TITLE = """
# Try Off Anyone
## ⚠️ Important
1. Choose an example image or upload your own
2. Use the Pen tool to draw a mask over the clothing area you want to extract
[](https://arxiv.org/abs/2412.08573)
"""
DEVICE = torch.device("cuda" if torch.cuda.is_available() else "cpu")
DTYPE = torch.bfloat16 if torch.cuda.is_bf16_supported() else torch.float32
pipeline_tryoff = TryOffAnyone(
device=DEVICE,
dtype=DTYPE,
)
mask_processor = diffusers.image_processor.VaeImageProcessor(
vae_scale_factor=8,
do_normalize=False,
do_binarize=True,
do_convert_grayscale=True,
)
vae_processor = diffusers.image_processor.VaeImageProcessor(
vae_scale_factor=8,
)
class ImageData(TypedDict):
background: Image.Image
composite: Image.Image
layers: list[Image.Image]
@spaces.GPU
def process(
image_data: ImageData,
image_width: int,
image_height: int,
num_inference_steps: int,
condition_scale: float,
seed: int,
) -> Image.Image:
assert image_width > 0
assert image_height > 0
assert num_inference_steps > 0
assert condition_scale > 0
assert seed >= 0
# extract image and mask from image_data
image = image_data["background"]
mask = image_data["layers"][0]
# preprocess image
image = image.convert("RGB").resize((image_width, image_height))
image_preprocessed = vae_processor.preprocess( # pyright: ignore[reportUnknownMemberType,reportAssignmentType]
image=image,
width=image_width,
height=image_height,
)[0]
# preprocess mask
mask = mask.getchannel("A").resize((image_width, image_height))
mask_preprocessed = mask_processor.preprocess( # pyright: ignore[reportUnknownMemberType]
image=mask,
width=image_width,
height=image_height,
)[0]
# generate the TryOff image
gen = torch.Generator(device=DEVICE).manual_seed(seed)
tryoff_image = pipeline_tryoff(
image_preprocessed,
mask_preprocessed,
inference_steps=num_inference_steps,
scale=condition_scale,
generator=gen,
)[0]
return tryoff_image
with gr.Blocks() as demo:
gr.Markdown(TITLE)
with gr.Row():
with gr.Column():
input_image = gr.ImageMask(
label="Input Image",
height=1024, # https://github.com/gradio-app/gradio/issues/10236
type="pil",
interactive=True,
)
run_button = gr.Button(
value="Extract Clothing",
)
gr.Examples(
examples=[
["examples/model_1.jpg"],
["examples/model_2.jpg"],
["examples/model_3.jpg"],
["examples/model_4.jpg"],
["examples/model_5.jpg"],
["examples/model_6.jpg"],
["examples/model_7.jpg"],
["examples/model_8.jpg"],
["examples/model_9.jpg"],
],
inputs=[input_image],
)
with gr.Column():
output_image = gr.Image(
label="TryOff result",
height=1024,
image_mode="RGB",
type="pil",
)
with gr.Accordion("Advanced Settings", open=True):
seed = gr.Slider(
label="Seed",
minimum=0,
maximum=100_000,
value=69_420,
step=1,
)
scale = gr.Slider(
label="Scale",
minimum=0.5,
maximum=5,
value=2.5,
step=0.05,
)
num_inference_steps = gr.Slider(
label="Number of inference steps",
minimum=1,
maximum=50,
value=25,
step=1,
)
with gr.Row():
image_width = gr.Slider(
label="Image Width",
minimum=64,
maximum=1024,
value=384,
step=8,
)
image_height = gr.Slider(
label="Image Height",
minimum=64,
maximum=1024,
value=512,
step=8,
)
run_button.click(
fn=process,
inputs=[
input_image,
image_width,
image_height,
num_inference_steps,
scale,
seed,
],
outputs=output_image,
)
demo.launch()
|