Spaces:
Runtime error
Runtime error
update app.py
Browse files
app.py
CHANGED
@@ -1,7 +1,269 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
import gradio as gr
|
|
|
|
|
|
|
2 |
|
3 |
-
|
4 |
-
|
|
|
|
|
|
|
|
|
|
|
5 |
|
6 |
-
|
7 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import os
|
2 |
+
import sys
|
3 |
+
sys.path.append(os.getcwd())
|
4 |
+
sys.path.append(os.path.join(os.getcwd(), "annotator/entityseg"))
|
5 |
+
import cv2
|
6 |
+
import spaces
|
7 |
+
import einops
|
8 |
+
import torch
|
9 |
import gradio as gr
|
10 |
+
import numpy as np
|
11 |
+
from pytorch_lightning import seed_everything
|
12 |
+
from PIL import Image
|
13 |
|
14 |
+
from annotator.util import resize_image, HWC3
|
15 |
+
from annotator.canny import CannyDetector
|
16 |
+
from annotator.midas import MidasDetector
|
17 |
+
from annotator.entityseg import EntitysegDetector
|
18 |
+
from annotator.openpose import OpenposeDetector
|
19 |
+
from annotator.content import ContentDetector
|
20 |
+
from annotator.cielab import CIELabDetector
|
21 |
|
22 |
+
from models.util import create_model, load_state_dict
|
23 |
+
from models.ddim_hacked import DDIMSampler
|
24 |
+
|
25 |
+
'''
|
26 |
+
define conditions
|
27 |
+
'''
|
28 |
+
max_conditions = 8
|
29 |
+
condition_types = ["edge", "depth", "seg", "pose", "content", "color"]
|
30 |
+
|
31 |
+
apply_canny = CannyDetector()
|
32 |
+
apply_midas = MidasDetector()
|
33 |
+
apply_seg = EntitysegDetector()
|
34 |
+
apply_openpose = OpenposeDetector()
|
35 |
+
apply_content = ContentDetector()
|
36 |
+
apply_color = CIELabDetector()
|
37 |
+
|
38 |
+
processors = {
|
39 |
+
"edge": apply_canny,
|
40 |
+
"depth": apply_midas,
|
41 |
+
"seg": apply_seg,
|
42 |
+
"pose": apply_openpose,
|
43 |
+
"content": apply_content,
|
44 |
+
"color": apply_color,
|
45 |
+
}
|
46 |
+
|
47 |
+
descriptors = {
|
48 |
+
"edge": "canny",
|
49 |
+
"depth": "depth",
|
50 |
+
"seg": "segmentation",
|
51 |
+
"pose": "openpose",
|
52 |
+
}
|
53 |
+
|
54 |
+
|
55 |
+
@torch.no_grad()
|
56 |
+
def get_unconditional_global(c_global):
|
57 |
+
if isinstance(c_global, dict):
|
58 |
+
return {k:torch.zeros_like(v) for k,v in c_global.items()}
|
59 |
+
elif isinstance(c_global, list):
|
60 |
+
return [torch.zeros_like(c) for c in c_global]
|
61 |
+
else:
|
62 |
+
return torch.zeros_like(c_global)
|
63 |
+
|
64 |
+
|
65 |
+
@spaces.GPU
|
66 |
+
def process(prompt, a_prompt, n_prompt, num_samples, image_resolution, ddim_steps,
|
67 |
+
strength, scale, seed, eta, global_strength, color_strength, local_strength, *args):
|
68 |
+
|
69 |
+
seed_everything(seed)
|
70 |
+
|
71 |
+
conds_and_types = args
|
72 |
+
conds = conds_and_types[0::2]
|
73 |
+
types = conds_and_types[1::2]
|
74 |
+
conds = [c for c in conds if c is not None]
|
75 |
+
types = [t for t in types if t is not None]
|
76 |
+
assert len(conds) == len(types)
|
77 |
+
|
78 |
+
detected_maps = []
|
79 |
+
other_maps = []
|
80 |
+
tasks = []
|
81 |
+
|
82 |
+
# initialize global control
|
83 |
+
global_conditions = dict(clipembedding=np.zeros((1, 768), dtype=np.float32), color=np.zeros((1, 180), dtype=np.float32))
|
84 |
+
global_control = {}
|
85 |
+
for key in global_conditions.keys():
|
86 |
+
global_cond = torch.from_numpy(global_conditions[key]).unsqueeze(0).repeat(num_samples, 1, 1)
|
87 |
+
global_cond = global_cond.cuda().to(memory_format=torch.contiguous_format).float()
|
88 |
+
global_control[key] = global_cond
|
89 |
+
|
90 |
+
# initialize local control
|
91 |
+
anchor_image = HWC3(np.zeros((image_resolution, image_resolution, 3)).astype(np.uint8))
|
92 |
+
oH, oW = anchor_image.shape[:2]
|
93 |
+
H, W, C = resize_image(anchor_image, image_resolution).shape
|
94 |
+
anchor_tensor = ddim_sampler.model.qformer_vis_processor['eval'](Image.fromarray(anchor_image))
|
95 |
+
local_control = torch.tensor(anchor_tensor).cuda().to(memory_format=torch.contiguous_format).half()
|
96 |
+
|
97 |
+
task_prompt = ''
|
98 |
+
|
99 |
+
with torch.no_grad():
|
100 |
+
|
101 |
+
# set up local control
|
102 |
+
for cond, typ in zip(conds, types):
|
103 |
+
if typ in ['edge', 'depth', 'seg', 'pose']:
|
104 |
+
oH, oW = cond.shape[:2]
|
105 |
+
cond_image = HWC3(cv2.resize(cond, (W, H)))
|
106 |
+
cond_detected_map = processors[typ](cond_image)
|
107 |
+
cond_detected_map = HWC3(cond_detected_map)
|
108 |
+
detected_maps.append(cond_detected_map)
|
109 |
+
tasks.append(descriptors[typ])
|
110 |
+
elif typ in ['content']:
|
111 |
+
other_maps.append(cond)
|
112 |
+
content_image = cv2.cvtColor(cond, cv2.COLOR_RGB2BGR)
|
113 |
+
content_emb = apply_content(content_image)
|
114 |
+
global_conditions['clipembedding'] = content_emb
|
115 |
+
elif typ in ['color']:
|
116 |
+
color_hist = apply_color(cond)
|
117 |
+
global_conditions['color'] = color_hist
|
118 |
+
color_palette = apply_color.hist_to_palette(color_hist) # (50, 189, 3)
|
119 |
+
color_palette = cv2.resize(color_palette, (W, H), cv2.INTER_NEAREST)
|
120 |
+
other_maps.append(color_palette)
|
121 |
+
if len(detected_maps) > 0:
|
122 |
+
local_control = torch.cat([ddim_sampler.model.qformer_vis_processor['eval'](Image.fromarray(img)).cuda().unsqueeze(0) for img in detected_maps], dim=1)
|
123 |
+
task_prompt = ' conditioned on ' + ' and '.join(tasks)
|
124 |
+
local_control = local_control.repeat(num_samples, 1, 1, 1)
|
125 |
+
|
126 |
+
# set up global control
|
127 |
+
for key in global_conditions.keys():
|
128 |
+
global_cond = torch.from_numpy(global_conditions[key]).unsqueeze(0).repeat(num_samples, 1, 1)
|
129 |
+
global_cond = global_cond.cuda().to(memory_format=torch.contiguous_format).float()
|
130 |
+
global_control[key] = global_cond
|
131 |
+
|
132 |
+
# set up prompt
|
133 |
+
input_prompt = (prompt + ' ' + task_prompt).strip()
|
134 |
+
|
135 |
+
# set up cfg
|
136 |
+
uc_local_control = local_control
|
137 |
+
uc_global_control = get_unconditional_global(global_control)
|
138 |
+
cond = {
|
139 |
+
"local_control": [local_control],
|
140 |
+
"c_crossattn": [model.get_learned_conditioning([prompt + ', ' + a_prompt] * num_samples)],
|
141 |
+
"global_control": [global_control],
|
142 |
+
"text": [[input_prompt] * num_samples],
|
143 |
+
}
|
144 |
+
un_cond = {
|
145 |
+
"local_control": [uc_local_control],
|
146 |
+
"c_crossattn": [model.get_learned_conditioning([n_prompt] * num_samples)],
|
147 |
+
'global_control': [uc_global_control],
|
148 |
+
"text": [[input_prompt] * num_samples],
|
149 |
+
}
|
150 |
+
shape = (4, H // 8, W // 8)
|
151 |
+
|
152 |
+
model.control_scales = [strength] * 13
|
153 |
+
samples, _ = ddim_sampler.sample(ddim_steps, num_samples,
|
154 |
+
shape, cond, verbose=False, eta=eta,
|
155 |
+
unconditional_guidance_scale=scale,
|
156 |
+
unconditional_conditioning=un_cond,
|
157 |
+
global_strength=global_strength,
|
158 |
+
color_strength=color_strength,
|
159 |
+
local_strength=local_strength)
|
160 |
+
|
161 |
+
x_samples = model.decode_first_stage(samples)
|
162 |
+
x_samples = (einops.rearrange(x_samples, 'b c h w -> b h w c') * 127.5 + 127.5).cpu().numpy().clip(0, 255).astype(np.uint8)
|
163 |
+
results = [x_samples[i] for i in range(num_samples)]
|
164 |
+
|
165 |
+
results = [cv2.resize(res, (oW, oH)) for res in results]
|
166 |
+
detected_maps = [cv2.resize(maps, (oW, oH)) for maps in detected_maps]
|
167 |
+
return [results, detected_maps+other_maps]
|
168 |
+
|
169 |
+
|
170 |
+
def variable_image_outputs(k):
|
171 |
+
if k is None:
|
172 |
+
k = 1
|
173 |
+
k = int(k)
|
174 |
+
imageboxes = []
|
175 |
+
for i in range(max_conditions):
|
176 |
+
if i<k:
|
177 |
+
with gr.Row(visible=True):
|
178 |
+
img = gr.Image(sources=['upload'], type="numpy", label=f'Condition {i+1}', visible=True, interactive=True, scale=3, height=200)
|
179 |
+
typ = gr.Dropdown(condition_types, visible=True, interactive=True, label="type", scale=1)
|
180 |
+
else:
|
181 |
+
with gr.Row(visible=False):
|
182 |
+
img = gr.Image(sources=['upload'], type="numpy", label=f'Condition {i+1}', visible=False, scale=3, height=200)
|
183 |
+
typ = gr.Dropdown(condition_types, visible=False, interactive=True, label="type", scale=1)
|
184 |
+
imageboxes.append(img)
|
185 |
+
imageboxes.append(typ)
|
186 |
+
return imageboxes
|
187 |
+
|
188 |
+
|
189 |
+
'''
|
190 |
+
define model
|
191 |
+
'''
|
192 |
+
config_file = "configs/anycontrol.yaml"
|
193 |
+
model_file = "ckpts/anycontrol_15.ckpt"
|
194 |
+
model = create_model(config_file).cpu()
|
195 |
+
model.load_state_dict(load_state_dict(model_file, location='cuda'))
|
196 |
+
model = model.cuda()
|
197 |
+
ddim_sampler = DDIMSampler(model)
|
198 |
+
|
199 |
+
|
200 |
+
|
201 |
+
block = gr.Blocks(theme='bethecloud/storj_theme').queue()
|
202 |
+
with block:
|
203 |
+
with gr.Row():
|
204 |
+
gr.Markdown("## AnyControl Demo")
|
205 |
+
gr.Markdown("---")
|
206 |
+
with gr.Row():
|
207 |
+
with gr.Column(scale=1):
|
208 |
+
with gr.Blocks():
|
209 |
+
s = gr.Slider(1, max_conditions, value=1, step=1, label="How many conditions to upload:")
|
210 |
+
imageboxes = []
|
211 |
+
for i in range(max_conditions):
|
212 |
+
if i==0:
|
213 |
+
with gr.Row():
|
214 |
+
img = gr.Image(visible=True, sources=['upload'], type="numpy", label='Condition 1', interactive=True, scale=3, height=200)
|
215 |
+
typ = gr.Dropdown(condition_types, visible=True, interactive=True, label="type", scale=1)
|
216 |
+
else:
|
217 |
+
with gr.Row():
|
218 |
+
img = gr.Image(visible=False, sources=['upload'], type="numpy", label=f'Condition {i+1}', scale=3, height=200)
|
219 |
+
typ = gr.Dropdown(condition_types, visible=False, interactive=True, label="type", scale=1)
|
220 |
+
imageboxes.append(img)
|
221 |
+
imageboxes.append(typ)
|
222 |
+
s.change(variable_image_outputs, s, imageboxes)
|
223 |
+
with gr.Column(scale=2):
|
224 |
+
with gr.Row():
|
225 |
+
prompt = gr.Textbox(label="Prompt")
|
226 |
+
with gr.Row():
|
227 |
+
with gr.Column():
|
228 |
+
with gr.Accordion("Advanced options", open=False):
|
229 |
+
num_samples = gr.Slider(label="Images", minimum=1, maximum=12, value=4, step=1)
|
230 |
+
image_resolution = gr.Slider(label="Image Resolution", minimum=256, maximum=768, value=512, step=64)
|
231 |
+
strength = gr.Slider(label="Control Strength", minimum=0.0, maximum=2.0, value=1, step=0.01)
|
232 |
+
|
233 |
+
local_strength = gr.Slider(label="Local Strength", minimum=0, maximum=2, value=1, step=0.01)
|
234 |
+
global_strength = gr.Slider(label="Global Strength", minimum=0, maximum=2, value=1, step=0.01)
|
235 |
+
color_strength = gr.Slider(label="Color Strength", minimum=0, maximum=2, value=1, step=0.01)
|
236 |
+
|
237 |
+
ddim_steps = gr.Slider(label="Steps", minimum=1, maximum=100, value=50, step=1)
|
238 |
+
scale = gr.Slider(label="Guidance Scale", minimum=0.1, maximum=30.0, value=7.5, step=0.1)
|
239 |
+
seed = gr.Slider(label="Seed", minimum=-1, maximum=2147483647, value=42, step=1)
|
240 |
+
eta = gr.Number(label="Eta (DDIM)", value=0.0)
|
241 |
+
|
242 |
+
a_prompt = gr.Textbox(label="Added Prompt", value='best quality, extremely detailed')
|
243 |
+
n_prompt = gr.Textbox(label="Negative Prompt",
|
244 |
+
value='longbody, lowres, bad anatomy, bad hands, missing fingers, extra digit, fewer digits, cropped, worst quality, low quality')
|
245 |
+
|
246 |
+
|
247 |
+
with gr.Row():
|
248 |
+
run_button = gr.Button(value="Run")
|
249 |
+
with gr.Row():
|
250 |
+
image_gallery = gr.Gallery(label='Generation', show_label=True, elem_id="gallery", columns=[4], rows=[1], height='auto', interactive=False)
|
251 |
+
with gr.Row():
|
252 |
+
cond_gallery = gr.Gallery(label='Condition', show_label=True, elem_id="gallery", columns=[4], rows=[1], height='auto', interactive=False)
|
253 |
+
|
254 |
+
inputs = [prompt, a_prompt, n_prompt, num_samples, image_resolution, ddim_steps,
|
255 |
+
strength, scale, seed, eta, local_strength, global_strength, color_strength] + imageboxes
|
256 |
+
run_button.click(fn=process, inputs=inputs, outputs=[image_gallery, cond_gallery])
|
257 |
+
|
258 |
+
|
259 |
+
# uncomment this block in case you need it
|
260 |
+
# os.environ['http_proxy'] = ''
|
261 |
+
# os.environ['https_proxy'] = ''
|
262 |
+
# os.environ['no_proxy'] = 'localhost,127.0.0.0/8,127.0.1.1'
|
263 |
+
# os.environ['HTTP_PROXY'] = ''
|
264 |
+
# os.environ['HTTPS_PROXY'] = ''
|
265 |
+
# os.environ['NO_PROXY'] = 'localhost,127.0.0.0/8,127.0.1.1'
|
266 |
+
# os.environ['TMPDIR'] = './tmpfiles'
|
267 |
+
|
268 |
+
|
269 |
+
block.launch(server_name='0.0.0.0', allowed_paths=["."], share=False)
|