Spaces:
Running
on
Zero
Running
on
Zero
Update app.py
Browse files
app.py
CHANGED
@@ -24,6 +24,73 @@ generator = None
|
|
24 |
accelerator = None
|
25 |
model_path = None
|
26 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
27 |
@spaces.GPU
|
28 |
def initialize_models():
|
29 |
global pipeline, generator, accelerator, model_path
|
@@ -41,62 +108,8 @@ def initialize_models():
|
|
41 |
token=os.environ['Read2']
|
42 |
)
|
43 |
|
44 |
-
# Load
|
45 |
-
|
46 |
-
os.path.join(model_path, "stable-diffusion-2-1-base/scheduler")
|
47 |
-
)
|
48 |
-
|
49 |
-
text_encoder = CLIPTextModel.from_pretrained(
|
50 |
-
os.path.join(model_path, "stable-diffusion-2-1-base/text_encoder")
|
51 |
-
)
|
52 |
-
|
53 |
-
tokenizer = CLIPTokenizer.from_pretrained(
|
54 |
-
os.path.join(model_path, "stable-diffusion-2-1-base/tokenizer")
|
55 |
-
)
|
56 |
-
|
57 |
-
feature_extractor = CLIPImageProcessor.from_pretrained(
|
58 |
-
os.path.join(model_path, "stable-diffusion-2-1-base/feature_extractor")
|
59 |
-
)
|
60 |
-
|
61 |
-
unet = UNet2DConditionModel.from_pretrained(
|
62 |
-
os.path.join(model_path, "stable-diffusion-2-1-base/unet")
|
63 |
-
)
|
64 |
-
|
65 |
-
controlnet = ControlNetModel.from_pretrained(
|
66 |
-
os.path.join(model_path, "Controlnet")
|
67 |
-
)
|
68 |
-
|
69 |
-
vae = AutoencoderKL.from_pretrained(
|
70 |
-
os.path.join(model_path, "vae")
|
71 |
-
)
|
72 |
-
|
73 |
-
# Freeze models
|
74 |
-
for model in [vae, text_encoder, unet, controlnet]:
|
75 |
-
model.requires_grad_(False)
|
76 |
-
|
77 |
-
# Initialize pipeline
|
78 |
-
pipeline = StableDiffusionControlNetPipeline(
|
79 |
-
vae=vae,
|
80 |
-
text_encoder=text_encoder,
|
81 |
-
tokenizer=tokenizer,
|
82 |
-
feature_extractor=feature_extractor,
|
83 |
-
unet=unet,
|
84 |
-
controlnet=controlnet,
|
85 |
-
scheduler=scheduler,
|
86 |
-
safety_checker=None,
|
87 |
-
requires_safety_checker=False,
|
88 |
-
)
|
89 |
-
|
90 |
-
# Get weight dtype based on mixed precision
|
91 |
-
weight_dtype = torch.float32
|
92 |
-
if accelerator.mixed_precision == "fp16":
|
93 |
-
weight_dtype = torch.float16
|
94 |
-
elif accelerator.mixed_precision == "bf16":
|
95 |
-
weight_dtype = torch.bfloat16
|
96 |
-
|
97 |
-
# Move models to device with appropriate dtype
|
98 |
-
for model in [text_encoder, vae, unet, controlnet]:
|
99 |
-
model.to(accelerator.device, dtype=weight_dtype)
|
100 |
|
101 |
# Initialize generator
|
102 |
generator = torch.Generator(device=accelerator.device)
|
@@ -149,6 +162,8 @@ def process_image(
|
|
149 |
t_max=0.6666,
|
150 |
t_min=0.0,
|
151 |
tile_diffusion=False,
|
|
|
|
|
152 |
added_prompt=prompt,
|
153 |
image=input_pil,
|
154 |
num_inference_steps=num_inference_steps,
|
@@ -158,6 +173,9 @@ def process_image(
|
|
158 |
guidance_scale=guidance_scale,
|
159 |
negative_prompt=negative_prompt,
|
160 |
conditioning_scale=conditioning_scale,
|
|
|
|
|
|
|
161 |
)
|
162 |
|
163 |
generated_image = output.images[0]
|
@@ -193,11 +211,7 @@ iface = gr.Interface(
|
|
193 |
],
|
194 |
outputs=gr.Image(label="Generated Image"),
|
195 |
title="Controllable Conditional Super-Resolution",
|
196 |
-
description="Upload an image to enhance its resolution using CCSR."
|
197 |
-
examples=[
|
198 |
-
["example1.jpg", "clean, sharp, detailed", "blurry, noise", 1.0, 1.0, 20, 42, 2, "adain"],
|
199 |
-
["example2.jpg", "high-resolution, pristine", "artifacts, pixelated", 1.5, 1.0, 30, 123, 2, "wavelet"],
|
200 |
-
]
|
201 |
)
|
202 |
|
203 |
if __name__ == "__main__":
|
|
|
24 |
accelerator = None
|
25 |
model_path = None
|
26 |
|
27 |
+
def load_pipeline(accelerator, model_path):
|
28 |
+
# Load scheduler
|
29 |
+
scheduler = DDPMScheduler.from_pretrained(
|
30 |
+
model_path,
|
31 |
+
subfolder="stable-diffusion-2-1-base/scheduler"
|
32 |
+
)
|
33 |
+
|
34 |
+
# Load models
|
35 |
+
text_encoder = CLIPTextModel.from_pretrained(
|
36 |
+
model_path,
|
37 |
+
subfolder="stable-diffusion-2-1-base/text_encoder"
|
38 |
+
)
|
39 |
+
|
40 |
+
tokenizer = CLIPTokenizer.from_pretrained(
|
41 |
+
model_path,
|
42 |
+
subfolder="stable-diffusion-2-1-base/tokenizer"
|
43 |
+
)
|
44 |
+
|
45 |
+
feature_extractor = CLIPImageProcessor.from_pretrained(
|
46 |
+
os.path.join(model_path, "stable-diffusion-2-1-base/feature_extractor")
|
47 |
+
)
|
48 |
+
|
49 |
+
unet = UNet2DConditionModel.from_pretrained(
|
50 |
+
model_path,
|
51 |
+
subfolder="stable-diffusion-2-1-base/unet"
|
52 |
+
)
|
53 |
+
|
54 |
+
controlnet = ControlNetModel.from_pretrained(
|
55 |
+
model_path,
|
56 |
+
subfolder="Controlnet"
|
57 |
+
)
|
58 |
+
|
59 |
+
vae = AutoencoderKL.from_pretrained(
|
60 |
+
model_path,
|
61 |
+
subfolder="vae"
|
62 |
+
)
|
63 |
+
|
64 |
+
# Freeze models
|
65 |
+
for model in [vae, text_encoder, unet, controlnet]:
|
66 |
+
model.requires_grad_(False)
|
67 |
+
|
68 |
+
# Initialize pipeline
|
69 |
+
pipeline = StableDiffusionControlNetPipeline(
|
70 |
+
vae=vae,
|
71 |
+
text_encoder=text_encoder,
|
72 |
+
tokenizer=tokenizer,
|
73 |
+
feature_extractor=feature_extractor,
|
74 |
+
unet=unet,
|
75 |
+
controlnet=controlnet,
|
76 |
+
scheduler=scheduler,
|
77 |
+
safety_checker=None,
|
78 |
+
requires_safety_checker=False,
|
79 |
+
)
|
80 |
+
|
81 |
+
# Set weight dtype based on mixed precision
|
82 |
+
weight_dtype = torch.float32
|
83 |
+
if accelerator.mixed_precision == "fp16":
|
84 |
+
weight_dtype = torch.float16
|
85 |
+
elif accelerator.mixed_precision == "bf16":
|
86 |
+
weight_dtype = torch.bfloat16
|
87 |
+
|
88 |
+
# Move models to accelerator device with appropriate dtype
|
89 |
+
for model in [text_encoder, vae, unet, controlnet]:
|
90 |
+
model.to(accelerator.device, dtype=weight_dtype)
|
91 |
+
|
92 |
+
return pipeline
|
93 |
+
|
94 |
@spaces.GPU
|
95 |
def initialize_models():
|
96 |
global pipeline, generator, accelerator, model_path
|
|
|
108 |
token=os.environ['Read2']
|
109 |
)
|
110 |
|
111 |
+
# Load pipeline using the original loading function
|
112 |
+
pipeline = load_pipeline(accelerator, model_path)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
113 |
|
114 |
# Initialize generator
|
115 |
generator = torch.Generator(device=accelerator.device)
|
|
|
162 |
t_max=0.6666,
|
163 |
t_min=0.0,
|
164 |
tile_diffusion=False,
|
165 |
+
tile_diffusion_size=512,
|
166 |
+
tile_diffusion_stride=256,
|
167 |
added_prompt=prompt,
|
168 |
image=input_pil,
|
169 |
num_inference_steps=num_inference_steps,
|
|
|
173 |
guidance_scale=guidance_scale,
|
174 |
negative_prompt=negative_prompt,
|
175 |
conditioning_scale=conditioning_scale,
|
176 |
+
start_steps=999,
|
177 |
+
start_point='lr',
|
178 |
+
use_vae_encode_condition=False
|
179 |
)
|
180 |
|
181 |
generated_image = output.images[0]
|
|
|
211 |
],
|
212 |
outputs=gr.Image(label="Generated Image"),
|
213 |
title="Controllable Conditional Super-Resolution",
|
214 |
+
description="Upload an image to enhance its resolution using CCSR."
|
|
|
|
|
|
|
|
|
215 |
)
|
216 |
|
217 |
if __name__ == "__main__":
|