Spaces:
Running
on
Zero
Running
on
Zero
File size: 8,744 Bytes
a858bb2 7f9fea1 a858bb2 9668e16 a858bb2 7f9fea1 a858bb2 7f9fea1 a858bb2 9668e16 a858bb2 |
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 |
import gradio as gr
import numpy as np
import mlxu
import os
import re
import torch
from io import BytesIO
from natsort import natsorted
from PIL import Image
from inference import LocalInferenceModel
FLAGS, _ = mlxu.define_flags_with_default(
host='0.0.0.0',
port=5000,
dtype='float16',
checkpoint='Emma02/LVM_ckpts',
torch_devices='',
context_frames=16,
)
def natural_sort_key(s):
return [int(text) if text.isdigit() else text.lower() for text in re.split('([0-9]+)', s)]
def load_example_image_groups(directory):
example_groups = {}
for subdir in os.listdir(directory):
subdir_path = os.path.join(directory, subdir)
if os.path.isdir(subdir_path):
example_groups[subdir] = []
images = [f for f in os.listdir(subdir_path) if f.lower().endswith(('.png', '.jpg', '.jpeg'))]
images = natsorted(images, key=natural_sort_key)
for filename in images:
img = Image.open(os.path.join(subdir_path, filename))
example_groups[subdir].append(img)
return example_groups
def main(_):
assert FLAGS.checkpoint != ''
model = LocalInferenceModel(
checkpoint=FLAGS.checkpoint,
torch_device=torch.device("cuda"),
dtype=FLAGS.dtype,
context_frames=FLAGS.context_frames,
use_lock=False,
)
checkerboard_r1 = np.concatenate([np.zeros((8, 8, 3)), np.ones((8, 8, 3)), np.zeros((8, 8, 3))], axis=1)
checkerboard_r2 = np.concatenate([np.ones((8, 8, 3)), np.zeros((8, 8, 3)), np.ones((8, 8, 3))], axis=1)
checkerboard = np.concatenate([checkerboard_r1, checkerboard_r2] * 16, axis=0).astype(np.float32)
def generate_images(input_images, n_new_frames, n_candidates, temperature=1.0, top_p=0.9):
assert len(input_images) > 0
input_images = [
np.array(img.convert('RGB').resize((256, 256)), dtype=np.float32) / 255.0
for img in input_images
]
input_images = np.stack(input_images, axis=0)
output_images = model([input_images], n_new_frames, n_candidates, temperature, top_p)[0]
generated_images = []
for candidate in output_images:
concatenated_image = []
for i, img in enumerate(candidate):
concatenated_image.append(img)
if i < len(candidate) - 1:
concatenated_image.append(checkerboard)
generated_images.append(
Image.fromarray(
(np.concatenate(concatenated_image, axis=1) * 255).astype(np.uint8)
)
)
return generated_images
with gr.Blocks(css="""
.small-button {
padding: 5px 10px;
min-width: 80px;
}
.large-gallery img {
width: 100%;
height: auto;
max-height: 150px;
}
""") as demo:
with gr.Column():
image_list = gr.State([])
gr.Markdown('# LVM Demo')
gr.Markdown(f'Serving model: {FLAGS.checkpoint}')
gr.Markdown('**We recommend using the model by prompting it with few shots: describe the task with pairs of (x, y) inputs where x is the input image and y the "annotated" image. Alternatively, one can also input a sequence of continuous frames and let the model generate the next one. Please refer to the default examples below.**')
gr.Markdown('## Inputs')
with gr.Row():
upload_drag = gr.File(
type='binary',
file_types=['image'],
file_count='multiple',
)
with gr.Column():
gen_length_slider = gr.Slider(
label='Generation length',
minimum=1,
maximum=32,
value=1,
step=1,
interactive=True,
)
n_candidates_slider = gr.Slider(
label='Number of candidates',
minimum=1,
maximum=10,
value=4,
step=1,
interactive=True,
)
temp_slider = gr.Slider(
label='Temperature',
minimum=0,
maximum=2.0,
value=1.0,
interactive=True,
)
top_p_slider = gr.Slider(
label='Top p',
minimum=0,
maximum=1.0,
value=0.9,
interactive=True,
)
clear_btn = gr.Button(
value='Clear',
elem_classes=['small-button'],
)
generate_btn = gr.Button(
value='Generate',
interactive=False,
elem_classes=['small-button'],
)
input_gallery = gr.Gallery(
columns=7,
rows=1,
object_fit='scale-down',
label="Input image sequence"
)
gr.Markdown('## Outputs (multi candidates)')
output_gallery = gr.Gallery(
columns=4,
object_fit='scale-down',
label="Output image"
)
def upload_image_fn(files, images):
for file in files:
images.append(Image.open(BytesIO(file)))
return {
upload_drag: None,
image_list: images,
input_gallery: images,
generate_btn: gr.update(interactive=True),
}
def clear_fn():
return {
image_list: [],
input_gallery: [],
generate_btn: gr.update(interactive=False),
output_gallery: [],
}
def disable_generate_btn():
return {
generate_btn: gr.update(interactive=False),
}
def generate_fn(images, n_candidates, gen_length, temperature, top_p):
new_images = generate_images(
images,
gen_length,
n_candidates=n_candidates,
temperature=temperature,
top_p=top_p,
)
return {
output_gallery: new_images,
generate_btn: gr.update(interactive=True),
}
upload_drag.upload(
upload_image_fn,
inputs=[upload_drag, image_list],
outputs=[upload_drag, image_list, input_gallery, generate_btn],
)
clear_btn.click(
clear_fn,
inputs=None,
outputs=[image_list, input_gallery, generate_btn, output_gallery],
)
generate_btn.click(
disable_generate_btn,
inputs=None,
outputs=[generate_btn],
).then(
generate_fn,
inputs=[image_list, n_candidates_slider, gen_length_slider, temp_slider, top_p_slider],
outputs=[output_gallery, generate_btn],
)
example_groups = load_example_image_groups('prompts')
def add_image_group_fn(group_name, images):
new_images = images + example_groups[group_name]
return {
image_list: new_images,
input_gallery: new_images,
generate_btn: gr.update(interactive=True),
}
gr.Markdown('## Default examples')
for group_name, group_images in example_groups.items():
with gr.Row():
with gr.Column(scale=3):
add_button = gr.Button(value=f'Add {group_name}', elem_classes=['small-button'])
with gr.Column(scale=7):
group_gallery = gr.Gallery(
value=[Image.fromarray(np.array(img)) for img in group_images],
columns=5,
rows=1,
object_fit='scale-down',
label=group_name,
elem_classes=['large-gallery'],
)
add_button.click(
add_image_group_fn,
inputs=[gr.State(group_name), image_list],
outputs=[image_list, input_gallery, generate_btn],
)
demo.launch()
if __name__ == "__main__":
mlxu.run(main)
|