Delete app.py
Browse files
app.py
DELETED
@@ -1,214 +0,0 @@
|
|
1 |
-
import os
|
2 |
-
import torch
|
3 |
-
from PIL import Image
|
4 |
-
import numpy as np
|
5 |
-
import cv2
|
6 |
-
import random
|
7 |
-
import gradio as gr
|
8 |
-
from gradio.themes import Soft
|
9 |
-
|
10 |
-
from diffusers import StableDiffusionControlNetPipeline, ControlNetModel
|
11 |
-
from diffusers import AutoencoderKL, UNet2DConditionModel, DDPMScheduler
|
12 |
-
from transformers import AutoTokenizer, CLIPTextModel, CLIPFeatureExtractor
|
13 |
-
from transformers import DPTForDepthEstimation, DPTImageProcessor
|
14 |
-
|
15 |
-
|
16 |
-
stable_diffusion_base = "runwayml/stable-diffusion-v1-5"
|
17 |
-
|
18 |
-
finetune_controlnet_path = "controlnet"
|
19 |
-
|
20 |
-
DEVICE = torch.device("cuda" if torch.cuda.is_available() else "cpu")
|
21 |
-
DTYPE = torch.float16 if torch.cuda.is_available() else torch.float32
|
22 |
-
|
23 |
-
pipeline = None
|
24 |
-
depth_estimator_model = None
|
25 |
-
depth_estimator_processor = None
|
26 |
-
|
27 |
-
|
28 |
-
def load_depth_estimator():
|
29 |
-
global depth_estimator_model, depth_estimator_processor
|
30 |
-
if depth_estimator_model is None:
|
31 |
-
model_name = "Intel/dpt-hybrid-midas"
|
32 |
-
depth_estimator_model = DPTForDepthEstimation.from_pretrained(model_name)
|
33 |
-
depth_estimator_processor = DPTImageProcessor.from_pretrained(model_name)
|
34 |
-
depth_estimator_model.to(DEVICE)
|
35 |
-
depth_estimator_model.eval()
|
36 |
-
|
37 |
-
return depth_estimator_model, depth_estimator_processor
|
38 |
-
|
39 |
-
|
40 |
-
|
41 |
-
def load_diffusion_pipeline():
|
42 |
-
global pipeline
|
43 |
-
if pipeline is None:
|
44 |
-
try:
|
45 |
-
if not os.path.exists(finetune_controlnet_path):
|
46 |
-
raise FileNotFoundError(f"ControlNet model not found: {finetune_controlnet_path}")
|
47 |
-
|
48 |
-
# 1. Load individual components of the base Stable Diffusion pipeline from Hugging Face Hub
|
49 |
-
vae = AutoencoderKL.from_pretrained(stable_diffusion_base, subfolder="vae", torch_dtype=DTYPE)
|
50 |
-
tokenizer = AutoTokenizer.from_pretrained(stable_diffusion_base, subfolder="tokenizer")
|
51 |
-
text_encoder = CLIPTextModel.from_pretrained(stable_diffusion_base, subfolder="text_encoder", torch_dtype=DTYPE)
|
52 |
-
unet = UNet2DConditionModel.from_pretrained(stable_diffusion_base, subfolder="unet", torch_dtype=DTYPE)
|
53 |
-
scheduler = DDPMScheduler.from_pretrained(stable_diffusion_base, subfolder="scheduler")
|
54 |
-
feature_extractor = CLIPFeatureExtractor.from_pretrained(stable_diffusion_base, subfolder="feature_extractor")
|
55 |
-
|
56 |
-
controlnet = ControlNetModel.from_pretrained(finetune_controlnet_path, torch_dtype=DTYPE)
|
57 |
-
pipeline = StableDiffusionControlNetPipeline(
|
58 |
-
vae=vae,
|
59 |
-
text_encoder=text_encoder,
|
60 |
-
tokenizer=tokenizer,
|
61 |
-
unet=unet,
|
62 |
-
controlnet=controlnet, # Your fine-tuned ControlNet
|
63 |
-
scheduler=scheduler,
|
64 |
-
safety_checker=None,
|
65 |
-
feature_extractor=feature_extractor,
|
66 |
-
image_encoder=None, # Explicitly set to None as it's not part of this setup
|
67 |
-
requires_safety_checker=False,
|
68 |
-
)
|
69 |
-
|
70 |
-
pipeline.to(DEVICE)
|
71 |
-
if torch.cuda.is_available() and hasattr(pipeline, "enable_xformers_memeory_efficient_attention"):
|
72 |
-
try:
|
73 |
-
pipeline.enable_xformers_memory_efficient_attention()
|
74 |
-
print("xformers memory efficient attention enabled.")
|
75 |
-
except Exception as e:
|
76 |
-
print(f"Could not enable xformers: {e}")
|
77 |
-
|
78 |
-
|
79 |
-
load_depth_estimator()
|
80 |
-
|
81 |
-
except Exception as e:
|
82 |
-
print(f"Error loading pipeline: {e}")
|
83 |
-
pipeline = None
|
84 |
-
raise RuntimeError(f"Failed to load diffusion pipeline: {e}")
|
85 |
-
return pipeline
|
86 |
-
|
87 |
-
|
88 |
-
|
89 |
-
def estimate_depth(pil_image: Image.Image) ->Image.Image:
|
90 |
-
global depth_estimator_model, depth_estimator_processor
|
91 |
-
if depth_estimator_model is None or depth_estimator_processor is None:
|
92 |
-
try:
|
93 |
-
load_depth_estimator()
|
94 |
-
except RuntimeError as e:
|
95 |
-
raise RuntimeError(f"Depth estimator not loaded: {e}")
|
96 |
-
|
97 |
-
input = depth_estimator_processor(pil_image, return_tensors = "pt")
|
98 |
-
input = {k: v.to(DEVICE) for k, v in input.items()}
|
99 |
-
|
100 |
-
|
101 |
-
with torch.no_grad():
|
102 |
-
output = depth_estimator_model(**input)
|
103 |
-
predicted_depth = output.predicted_depth
|
104 |
-
|
105 |
-
depth_numpy = predicted_depth.squeeze().cpu().numpy()
|
106 |
-
|
107 |
-
min_depth = depth_numpy.min()
|
108 |
-
max_depth = depth_numpy.max()
|
109 |
-
normalized_depth = (depth_numpy - min_depth) / (max_depth - min_depth)
|
110 |
-
|
111 |
-
inverted_normalized_depth = 1 - normalized_depth
|
112 |
-
|
113 |
-
depth_image_array = (inverted_normalized_depth * 255).astype(np.uint8)
|
114 |
-
depth_pil_image = Image.fromarray(depth_image_array).convert("RGB")
|
115 |
-
|
116 |
-
print("Depth estimation complete.")
|
117 |
-
return depth_pil_image
|
118 |
-
|
119 |
-
|
120 |
-
def generate_image_for_gradio(
|
121 |
-
prompt: str,
|
122 |
-
input_image_for_depth: Image.Image,
|
123 |
-
num_inference_step: int,
|
124 |
-
guidance_scale: float,
|
125 |
-
|
126 |
-
) -> Image.Image:
|
127 |
-
|
128 |
-
global pipeline
|
129 |
-
if pipeline is None:
|
130 |
-
try:
|
131 |
-
load_diffusion_pipeline()
|
132 |
-
except RuntimeError as e:
|
133 |
-
return gr.Error(f"Model not loaded: {e}")
|
134 |
-
|
135 |
-
try:
|
136 |
-
depth_map_pil = estimate_depth(input_image_for_depth)
|
137 |
-
except Exception as e:
|
138 |
-
return gr.Error(f"Error during depth estimation: {e}")
|
139 |
-
|
140 |
-
print(f"Generating image for prompt: '{prompt}'")
|
141 |
-
|
142 |
-
negative_prompt = "lowres, watermark, banner, logo, watermark, contactinfo, text, deformed, blurry, blur, out of focus, out of frame, surreal, ugly"
|
143 |
-
control_image = depth_map_pil.convert("RGB")
|
144 |
-
control_image = control_image.resize((512, 512), Image.LANCZOS)
|
145 |
-
|
146 |
-
input_image_for_pipeline = [control_image]
|
147 |
-
|
148 |
-
generator = None
|
149 |
-
# if seed is None:
|
150 |
-
seed = random.randint(0, 100000)
|
151 |
-
generator = torch.Generator(device=DEVICE).manual_seed(seed)
|
152 |
-
|
153 |
-
|
154 |
-
with torch.no_grad():
|
155 |
-
generated_images = pipeline(
|
156 |
-
prompt,
|
157 |
-
image=input_image_for_pipeline,
|
158 |
-
num_inference_steps=num_inference_step,
|
159 |
-
guidance_scale = guidance_scale,
|
160 |
-
generator=generator,
|
161 |
-
).images
|
162 |
-
|
163 |
-
|
164 |
-
print(f"Image generation complete (seed: {seed}).")
|
165 |
-
return generated_images[0]
|
166 |
-
|
167 |
-
|
168 |
-
|
169 |
-
iface = gr.Interface(
|
170 |
-
fn=generate_image_for_gradio,
|
171 |
-
inputs=[
|
172 |
-
gr.Textbox(label="Prompt", value="a high-quality photo of a modern interior design"),
|
173 |
-
gr.Image(type="pil", label="Input Image (for Depth Estimation)"),
|
174 |
-
gr.Slider(minimum=10, maximum=100, value=25, step=1, label="Inference Steps"),
|
175 |
-
gr.Slider(minimum=1.0, maximum=20.0, value=8.0, step=0.5, label="Guidance Scale"),
|
176 |
-
# gr.Number(label="Seed (optional, leave blank for random)", value=None),
|
177 |
-
# gr.Number(label="Resolution", value=512, interactive=False)
|
178 |
-
],
|
179 |
-
outputs=gr.Image(type="pil", label="Generated Image"),
|
180 |
-
title="Stable Diffusion ControlNet Depth Demo (with Depth Estimation)",
|
181 |
-
description="Upload an input image, and the app will estimate its depth map, then use it with your prompt to generate a new image. This allows for structural guidance from your input photo.",
|
182 |
-
allow_flagging="never",
|
183 |
-
live=False,
|
184 |
-
theme=Soft(),
|
185 |
-
css="""
|
186 |
-
/* Target the upload icon within the Image component */
|
187 |
-
.gr-image .icon-lg {
|
188 |
-
font-size: 2em !important; /* Adjust size as needed, e.g., 2em, 3em */
|
189 |
-
max-width: 50px; /* Max width to prevent it from filling the container */
|
190 |
-
max-height: 50px; /* Max height */
|
191 |
-
}
|
192 |
-
/* Target the image placeholder icon (if it's different) */
|
193 |
-
.gr-image .gr-image-placeholder {
|
194 |
-
max-width: 100px; /* Adjust size as needed */
|
195 |
-
max-height: 100px;
|
196 |
-
object-fit: contain; /* Ensures the icon scales down without distortion */
|
197 |
-
}
|
198 |
-
/* General styling for the image input area to ensure it has space */
|
199 |
-
.gr-image-container {
|
200 |
-
min-height: 200px; /* Give the image input area a minimum height */
|
201 |
-
display: flex;
|
202 |
-
align-items: center;
|
203 |
-
justify-content: center;
|
204 |
-
}
|
205 |
-
"""
|
206 |
-
)
|
207 |
-
|
208 |
-
|
209 |
-
|
210 |
-
load_diffusion_pipeline()
|
211 |
-
|
212 |
-
|
213 |
-
if __name__ == "__main__":
|
214 |
-
iface.launch()
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|