Spaces:
Runtime error
Runtime error
File size: 15,730 Bytes
85456ff |
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 |
import sys
import os
import torch
import gradio as gr
import numpy as np
from PIL import Image, ImageOps, ImageDraw, ImageFont, ImageColor
from urllib.request import urlopen
root = os.path.dirname(os.path.abspath(__file__))
static = os.path.join(root, "static")
from diffusers import DiffusionPipeline, DPMSolverMultistepScheduler
from diffusers.pipelines import TextToVideoSDPipeline
from diffusers.utils import export_to_video
from TrailBlazer.Misc import ConfigIO
from TrailBlazer.Misc import Logger as log
from TrailBlazer.Pipeline.TextToVideoSDPipelineCall import (
text_to_video_sd_pipeline_call,
)
from TrailBlazer.Pipeline.UNet3DConditionModelCall import (
unet3d_condition_model_forward,
)
TextToVideoSDPipeline.__call__ = text_to_video_sd_pipeline_call
from diffusers.models.unet_3d_condition import UNet3DConditionModel
unet3d_condition_model_forward_copy = UNet3DConditionModel.forward
UNet3DConditionModel.forward = unet3d_condition_model_forward
from diffusers.utils import export_to_video
model_id = "cerspense/zeroscope_v2_576w"
model_path = model_id
pipe = DiffusionPipeline.from_pretrained(model_path, torch_dtype=torch.float16)
pipe.scheduler = DPMSolverMultistepScheduler.from_config(pipe.scheduler.config)
pipe.enable_model_cpu_offload()
def core(bundle):
generator = torch.Generator().manual_seed(int(bundle["seed"]))
result = pipe(
bundle=bundle,
height=512,
width=512,
generator=generator,
num_inference_steps=40,
)
return result.frames
def clear_btn_fn():
return "", "", "", ""
def gen_btn_fn(
prompts,
bboxes,
frames,
word_prompt_indices,
trailing_length,
n_spatial_steps,
n_temporal_steps,
spatial_strengthen_scale,
spatial_weaken_scale,
temporal_strengthen_scale,
temporal_weaken_scale,
rand_seed,
):
bundle = {}
bundle["trailing_length"] = trailing_length
bundle["num_dd_spatial_steps"] = n_spatial_steps
bundle["num_dd_temporal_steps"] = n_temporal_steps
bundle["num_frames"] = 24
bundle["seed"] = rand_seed
bundle["spatial_strengthen_scale"] = spatial_strengthen_scale
bundle["spatial_weaken_scale"] = spatial_weaken_scale
bundle["temp_strengthen_scale"] = temporal_strengthen_scale
bundle["temp_weaken_scale"] = temporal_weaken_scale
bundle["token_inds"] = [int(v) for v in word_prompt_indices.split(",")]
bundle["keyframe"] = []
frames = frames.split(";")
bboxes = bboxes.split(";")
if ";" in prompts:
prompts = prompts.split(";")
else:
prompts = [prompts for i in range(len(frames))]
assert (
len(frames) == len(bboxes) == len(prompts)
), "Inconsistent number of keyframes in the given inputs."
frames.pop()
bboxes.pop()
prompts.pop()
for i in range(len(frames)):
keyframe = {}
keyframe["bbox_ratios"] = [float(v) for v in bboxes[i].split(",")]
keyframe["frame"] = int(frames[i])
keyframe["prompt"] = prompts[i]
bundle["keyframe"].append(keyframe)
print(bundle)
result = core(bundle)
path = export_to_video(result)
return path
def save_mask(inputs):
layers = inputs["layers"]
if not layers:
return inputs["background"]
mask = layers[0]
new_image = Image.new("RGBA", mask.size, color="white")
new_image.paste(mask, mask=mask)
new_image = new_image.convert("RGB")
print("SAve")
return ImageOps.invert(new_image)
def out_label_cb(im):
layers = im["layers"]
if not isinstance(layers, list):
layers = [layers]
img = None
text = "Bboxes: "
for idx, layer in enumerate(layers):
mask = np.array(layer).sum(axis=-1)
ys, xs = np.where(mask != 0)
h, w = mask.shape
if not list(xs) or not list(ys):
continue
x_min = np.min(xs)
x_max = np.max(xs)
y_min = np.min(ys)
y_max = np.max(ys)
text += "{:.2f},{:.2f},{:.2f},{:.2f}".format(
x_min * 1.0 / w, y_min * 1.0 / h, x_max * 1.0 / w, y_max * 1.0 / h
)
text += ";\n"
return text
def out_board_cb(im):
layers = im["layers"]
if not isinstance(layers, list):
layers = [layers]
img = None
for idx, layer in enumerate(layers):
mask = np.array(layer).sum(axis=-1)
ys, xs = np.where(mask != 0)
if not list(xs) or not list(ys):
continue
h, w = mask.shape
if not img:
img = Image.new("RGBA", (w, h))
x_min = np.min(xs)
x_max = np.max(xs)
y_min = np.min(ys)
y_max = np.max(ys)
# output
shape = [(x_min, y_min), (x_max, y_max)]
colors = list(ImageColor.colormap.keys())
draw = ImageDraw.Draw(img)
draw.rectangle(shape, outline=colors[idx], width=5)
text = "Bbox#{}".format(idx)
font = ImageFont.load_default()
draw.text((x_max - 0.5 * (x_max - x_min), y_max), text, font=font, align="left")
return img
with gr.Blocks(
analytics_enabled=False,
title="TrailBlazer Demo",
) as main:
description = """
<h1 align="center" style="font-size: 48px">TrailBlazer: Trajectory Control for Diffusion-Based Video Generation</h1>
<h4 align="center" style="margin: 0;">If you like our project, please give us a star ✨ at our Huggingface space, and our Github repository.</h4>
<br>
<span align="center" style="font-size: 18px">
[<a href="https://hohonu-vicml.github.io/Trailblazer.Page/" target="_blank">Project Page</a>]
[<a href="http://arxiv.org/abs/2401.00896" target="_blank">Paper</a>]
[<a href="https://github.com/hohonu-vicml/Trailblazer" target="_blank">GitHub</a>]
[<a href="https://www.youtube.com/watch?v=kEN-32wN-xQ" target="_blank">Project Video</a>]
[<a href="https://www.youtube.com/watch?v=P-PSkS7sNco" target="_blank">Result Video</a>]
</span>
</p>
<p>
<strong>Usage:</strong> Our Gradio app is implemented based on our executable script CmdTrailBlazer in our github repository. Please see our general information below for a quick guidance, as well as the hints within the app widgets.
<ul>
<li>Basic: The bounding box (bbox) is the tuple of four floats for the rectangular corners: left, top, right, bottom in the normalized ratio. The Word prompt indices is a list of 1-indexed numbers determining the prompt word.</li>
<li>Advanced Options: We also offer some key parameters to adjust the synthesis result. Please see our paper for more information about the ablations.</li>
</ul>
</p>
"""
gr.HTML(description)
with gr.Row():
with gr.Column(scale=2):
with gr.Row():
with gr.Tab("Main"):
text_prompt_tb = gr.Textbox(
interactive=True, label="Keyframe: Prompt"
)
bboxes_tb = gr.Textbox(interactive=True, label="Keyframe: Bboxes")
frame_tb = gr.Textbox(
interactive=True, label="Keyframe: frame indices"
)
with gr.Row():
word_prompt_indices_tb = gr.Textbox(
interactive=True, label="Word prompt indices:"
)
text = "Hint: Each keyframe ends with <strong>SEMICOLON</strong>, and <strong>COMMA</strong> for separating each value in the keyframe. The prompt field can be a single prompt without semicolon, or multiple prompts ended semicolon. One can use the SketchPadHelper tab to help to design the bboxes field."
gr.HTML(text)
with gr.Row():
clear_btn = gr.Button(value="Clear")
gen_btn = gr.Button(value="Generate")
with gr.Accordion("Advanced Options", open=False):
text = "Hint: This default value should be sufficient for most tasks. However, it's important to note that our approach is currently implemented on ZeroScope, and its performance may be influenced by the model's characteristics. We plan to conduct experiments on different models in the future."
gr.HTML(text)
with gr.Row():
trailing_length = gr.Slider(
minimum=0,
maximum=30,
step=1,
value=13,
interactive=True,
label="#Trailing",
)
n_spatial_steps = gr.Slider(
minimum=0,
maximum=30,
step=1,
value=5,
interactive=True,
label="#Spatial edits",
)
n_temporal_steps = gr.Slider(
minimum=0,
maximum=30,
step=1,
value=5,
interactive=True,
label="#Temporal edits",
)
with gr.Row():
spatial_strengthen_scale = gr.Slider(
minimum=0,
maximum=2,
step=0.01,
value=0.15,
interactive=True,
label="Spatial Strengthen Scale",
)
spatial_weaken_scale = gr.Slider(
minimum=0,
maximum=1,
step=0.01,
value=0.001,
interactive=True,
label="Spatial Weaken Scale",
)
temporal_strengthen_scale = gr.Slider(
minimum=0,
maximum=2,
step=0.01,
value=0.15,
interactive=True,
label="Temporal Strengthen Scale",
)
temporal_weaken_scale = gr.Slider(
minimum=0,
maximum=1,
step=0.01,
value=0.001,
interactive=True,
label="Temporal Weaken Scale",
)
with gr.Row():
guidance_scale = gr.Slider(
minimum=0,
maximum=50,
step=0.5,
value=7.5,
interactive=True,
label="Guidance Scale",
)
rand_seed = gr.Slider(
minimum=0,
maximum=523451232531,
step=1,
value=0,
interactive=True,
label="Seed",
)
with gr.Tab("SketchPadHelper"):
with gr.Row():
user_board = gr.ImageMask(type="pil", label="Draw me")
out_board = gr.Image(type="pil", label="Processed bbox")
user_board.change(
out_board_cb, inputs=[user_board], outputs=[out_board]
)
with gr.Row():
text = "Hint: Utilize a black pen with the Draw Button to create a ``rough'' bbox. When you press the green ``Save Changes'' Button, the app calculates the minimum and maximum boundaries. Each ``Layer'', located at the bottom left of the pad, corresponds to one bounding box. Copy the returned value to the bbox textfield in the main tab."
gr.HTML(text)
with gr.Row():
out_label = gr.Label(label="Converted bboxes string")
user_board.change(
out_label_cb, inputs=[user_board], outputs=[out_label]
)
with gr.Column(scale=1):
gr.HTML(
'<span style="font-size: 20px; font-weight: bold">Generated Images</span>'
)
with gr.Row():
out_gen_1 = gr.Video(visible=True, show_label=False)
with gr.Row():
gr.Examples(
examples=[
[
"A clown fish swimming in a coral reef",
"0.5,0.35,1.0,0.65; 0.0,0.35,0.5,0.65;",
"0; 24;",
"1,2,3",
"123451232531",
"assets/gradio/fish-RL.mp4",
],
[
"A cat is running on the grass",
"0.0,0.35,0.4,0.65; 0.6,0.35,1.0,0.65; 0.0,0.35,0.4,0.65;"
"0.6,0.35,1.0,0.65; 0.0,0.35,0.4,0.65;",
"0; 6; 12; 18; 24;",
"1,2",
"123451232530",
"assets/gradio/cat-LRLR.mp4",
],
[
"A fish swimming in the ocean",
"0.0,0.0,0.1,0.1; 0.5,0.5,1.0,1.0;",
"0; 24;",
"1, 2",
"0",
"assets/gradio/fish-TL2BR.mp4"
],
[
"A tiger walking alone down the street",
"0.0,0.0,0.1,0.1; 0.5,0.5,1.0,1.0;",
"0; 24;",
"1, 2",
"0",
"assets/gradio/tiger-TL2BR.mp4"
],
[
"A white cat walking on the grass; A yellow dog walking on the grass;",
"0.7,0.4,1.0,0.65; 0.0,0.4,0.3,0.65;",
"0; 24;",
"1,2,3",
"123451232531",
"assets/gradio/Cat2Dog.mp4",
],
],
inputs=[text_prompt_tb, bboxes_tb, frame_tb, word_prompt_indices_tb, rand_seed,out_gen_1],
outputs=None,
fn=None,
cache_examples=False,
)
clear_btn.click(
clear_btn_fn,
inputs=[],
outputs=[text_prompt_tb, bboxes_tb, frame_tb, word_prompt_indices_tb],
queue=False,
)
gen_btn.click(
gen_btn_fn,
inputs=[
text_prompt_tb,
bboxes_tb,
frame_tb,
word_prompt_indices_tb,
trailing_length,
n_spatial_steps,
n_temporal_steps,
spatial_strengthen_scale,
spatial_weaken_scale,
temporal_strengthen_scale,
temporal_weaken_scale,
rand_seed,
],
outputs=[out_gen_1],
queue=False,
)
if __name__ == "__main__":
main.launch(share=False)
|