New logic
Browse files
app.py
CHANGED
@@ -1,20 +1,26 @@
|
|
1 |
import gradio as gr
|
2 |
import torch
|
3 |
import jax
|
|
|
4 |
import numpy as np
|
5 |
from PIL import Image
|
6 |
import pickle
|
|
|
7 |
from huggingface_hub import hf_hub_download
|
8 |
from diffusers import StableDiffusionXLImg2ImgPipeline
|
9 |
-
from transformers import
|
10 |
-
from model import build_thera
|
|
|
|
|
|
|
|
|
11 |
|
12 |
# Configurar dispositivos
|
13 |
JAX_DEVICE = jax.devices("cpu")[0]
|
14 |
TORCH_DEVICE = "cpu"
|
15 |
|
16 |
|
17 |
-
# 1. Carregar modelos do Thera
|
18 |
def load_thera_model(repo_id, filename):
|
19 |
model_path = hf_hub_download(repo_id=repo_id, filename=filename)
|
20 |
with open(model_path, 'rb') as fh:
|
@@ -27,7 +33,7 @@ def load_thera_model(repo_id, filename):
|
|
27 |
print("Carregando Thera EDSR...")
|
28 |
model_edsr, params_edsr = load_thera_model("prs-eth/thera-edsr-pro", "model.pkl")
|
29 |
|
30 |
-
# 2. Carregar SDXL + LoRA
|
31 |
print("Carregando SDXL + LoRA...")
|
32 |
pipe = StableDiffusionXLImg2ImgPipeline.from_pretrained(
|
33 |
"stabilityai/stable-diffusion-xl-base-1.0",
|
@@ -35,33 +41,36 @@ pipe = StableDiffusionXLImg2ImgPipeline.from_pretrained(
|
|
35 |
).to(TORCH_DEVICE)
|
36 |
pipe.load_lora_weights("KappaNeuro/bas-relief", weight_name="BAS-RELIEF.safetensors")
|
37 |
|
38 |
-
# 3. Carregar modelo de profundidade
|
39 |
print("Carregando DPT Depth...")
|
40 |
-
feature_extractor =
|
41 |
depth_model = DPTForDepthEstimation.from_pretrained("Intel/dpt-large").to(TORCH_DEVICE)
|
42 |
|
43 |
|
44 |
-
# Pipeline principal
|
45 |
def full_pipeline(image, prompt, scale_factor=2.0):
|
46 |
try:
|
47 |
# 1. Super Resolução com Thera
|
48 |
-
|
|
|
49 |
target_shape = (int(image.height * scale_factor), int(image.width * scale_factor))
|
50 |
|
51 |
-
#
|
52 |
source_jax = jax.device_put(source, JAX_DEVICE)
|
|
|
53 |
|
54 |
# Processar com Thera
|
55 |
upscaled = model_edsr.apply(
|
56 |
params_edsr,
|
57 |
source_jax,
|
|
|
58 |
target_shape,
|
59 |
do_ensemble=True
|
60 |
)
|
61 |
upscaled_pil = Image.fromarray((np.array(upscaled) * 255).astype(np.uint8))
|
62 |
|
63 |
# 2. Gerar Bas-Relief
|
64 |
-
full_prompt = f"BAS-RELIEF {prompt},
|
65 |
bas_relief = pipe(
|
66 |
prompt=full_prompt,
|
67 |
image=upscaled_pil,
|
@@ -82,26 +91,32 @@ def full_pipeline(image, prompt, scale_factor=2.0):
|
|
82 |
mode="bicubic"
|
83 |
).squeeze().cpu().numpy()
|
84 |
|
85 |
-
|
|
|
|
|
|
|
86 |
|
87 |
except Exception as e:
|
88 |
raise gr.Error(f"Erro no processamento: {str(e)}")
|
89 |
|
90 |
|
91 |
-
# Interface Gradio
|
92 |
with gr.Blocks(title="Super Res + Bas-Relief") as app:
|
93 |
gr.Markdown("## 🔍 Super Resolução + 🗿 Bas-Relief + 🗺️ Profundidade")
|
94 |
|
95 |
with gr.Row():
|
96 |
with gr.Column():
|
97 |
img_input = gr.Image(type="pil", label="Imagem de Entrada")
|
98 |
-
prompt = gr.Textbox(
|
|
|
|
|
|
|
99 |
scale = gr.Slider(1.0, 4.0, value=2.0, label="Fator de Escala")
|
100 |
btn = gr.Button("Processar")
|
101 |
|
102 |
with gr.Column():
|
103 |
-
img_upscaled = gr.Image(label="Super Resolvida")
|
104 |
-
img_basrelief = gr.Image(label="Bas-Relief")
|
105 |
img_depth = gr.Image(label="Mapa de Profundidade")
|
106 |
|
107 |
btn.click(
|
|
|
1 |
import gradio as gr
|
2 |
import torch
|
3 |
import jax
|
4 |
+
import jax.numpy as jnp
|
5 |
import numpy as np
|
6 |
from PIL import Image
|
7 |
import pickle
|
8 |
+
import warnings
|
9 |
from huggingface_hub import hf_hub_download
|
10 |
from diffusers import StableDiffusionXLImg2ImgPipeline
|
11 |
+
from transformers import DPTImageProcessor, DPTForDepthEstimation
|
12 |
+
from model import build_thera
|
13 |
+
|
14 |
+
# Configurações e supressão de avisos
|
15 |
+
warnings.filterwarnings("ignore", category=FutureWarning)
|
16 |
+
warnings.filterwarnings("ignore", category=UserWarning)
|
17 |
|
18 |
# Configurar dispositivos
|
19 |
JAX_DEVICE = jax.devices("cpu")[0]
|
20 |
TORCH_DEVICE = "cpu"
|
21 |
|
22 |
|
23 |
+
# 1. Carregar modelos do Thera ----------------------------------------------------------------
|
24 |
def load_thera_model(repo_id, filename):
|
25 |
model_path = hf_hub_download(repo_id=repo_id, filename=filename)
|
26 |
with open(model_path, 'rb') as fh:
|
|
|
33 |
print("Carregando Thera EDSR...")
|
34 |
model_edsr, params_edsr = load_thera_model("prs-eth/thera-edsr-pro", "model.pkl")
|
35 |
|
36 |
+
# 2. Carregar SDXL + LoRA ---------------------------------------------------------------------
|
37 |
print("Carregando SDXL + LoRA...")
|
38 |
pipe = StableDiffusionXLImg2ImgPipeline.from_pretrained(
|
39 |
"stabilityai/stable-diffusion-xl-base-1.0",
|
|
|
41 |
).to(TORCH_DEVICE)
|
42 |
pipe.load_lora_weights("KappaNeuro/bas-relief", weight_name="BAS-RELIEF.safetensors")
|
43 |
|
44 |
+
# 3. Carregar modelo de profundidade ----------------------------------------------------------
|
45 |
print("Carregando DPT Depth...")
|
46 |
+
feature_extractor = DPTImageProcessor.from_pretrained("Intel/dpt-large")
|
47 |
depth_model = DPTForDepthEstimation.from_pretrained("Intel/dpt-large").to(TORCH_DEVICE)
|
48 |
|
49 |
|
50 |
+
# Pipeline principal --------------------------------------------------------------------------
|
51 |
def full_pipeline(image, prompt, scale_factor=2.0):
|
52 |
try:
|
53 |
# 1. Super Resolução com Thera
|
54 |
+
image = image.convert("RGB")
|
55 |
+
source = np.array(image) / 255.0
|
56 |
target_shape = (int(image.height * scale_factor), int(image.width * scale_factor))
|
57 |
|
58 |
+
# Preparar parâmetros para JAX
|
59 |
source_jax = jax.device_put(source, JAX_DEVICE)
|
60 |
+
t = jnp.array([1.0 / (scale_factor ** 2)], dtype=jnp.float32)
|
61 |
|
62 |
# Processar com Thera
|
63 |
upscaled = model_edsr.apply(
|
64 |
params_edsr,
|
65 |
source_jax,
|
66 |
+
t,
|
67 |
target_shape,
|
68 |
do_ensemble=True
|
69 |
)
|
70 |
upscaled_pil = Image.fromarray((np.array(upscaled) * 255).astype(np.uint8))
|
71 |
|
72 |
# 2. Gerar Bas-Relief
|
73 |
+
full_prompt = f"BAS-RELIEF {prompt}, insanely detailed and complex engraving relief, ultra-high definition, rich in detail, 16K resolution"
|
74 |
bas_relief = pipe(
|
75 |
prompt=full_prompt,
|
76 |
image=upscaled_pil,
|
|
|
91 |
mode="bicubic"
|
92 |
).squeeze().cpu().numpy()
|
93 |
|
94 |
+
depth_normalized = (depth_map - depth_map.min()) / (depth_map.max() - depth_map.min())
|
95 |
+
depth_pil = Image.fromarray((depth_normalized * 255).astype(np.uint8))
|
96 |
+
|
97 |
+
return upscaled_pil, bas_relief, depth_pil
|
98 |
|
99 |
except Exception as e:
|
100 |
raise gr.Error(f"Erro no processamento: {str(e)}")
|
101 |
|
102 |
|
103 |
+
# Interface Gradio ----------------------------------------------------------------------------
|
104 |
with gr.Blocks(title="Super Res + Bas-Relief") as app:
|
105 |
gr.Markdown("## 🔍 Super Resolução + 🗿 Bas-Relief + 🗺️ Profundidade")
|
106 |
|
107 |
with gr.Row():
|
108 |
with gr.Column():
|
109 |
img_input = gr.Image(type="pil", label="Imagem de Entrada")
|
110 |
+
prompt = gr.Textbox(
|
111 |
+
label="Descrição do Relevo",
|
112 |
+
value="insanely detailed and complex engraving relief, ultra-high definition, rich in detail, and 16K resolution."
|
113 |
+
)
|
114 |
scale = gr.Slider(1.0, 4.0, value=2.0, label="Fator de Escala")
|
115 |
btn = gr.Button("Processar")
|
116 |
|
117 |
with gr.Column():
|
118 |
+
img_upscaled = gr.Image(label="Imagem Super Resolvida")
|
119 |
+
img_basrelief = gr.Image(label="Resultado Bas-Relief")
|
120 |
img_depth = gr.Image(label="Mapa de Profundidade")
|
121 |
|
122 |
btn.click(
|