Spaces:
Running
Running
Add Agentic Refinement tab: iterative image→pattern→3D→projection→compare→refine loop
Browse files
app.py
CHANGED
|
@@ -1,5 +1,5 @@
|
|
| 1 |
"""
|
| 2 |
-
Garment Image -> 2D Sewing Pattern + Chat Editing + 3D Preview
|
| 3 |
"""
|
| 4 |
import json, os, re, traceback, copy
|
| 5 |
from typing import Dict, Optional, Tuple, List
|
|
@@ -7,6 +7,7 @@ import gradio as gr
|
|
| 7 |
from PIL import Image
|
| 8 |
from pattern_generator import generate_pattern_from_analysis, get_pattern_pieces
|
| 9 |
from garment_3d import create_3d_figure
|
|
|
|
| 10 |
|
| 11 |
GARMENT_ANALYSIS_PROMPT = """You are a professional fashion pattern maker. Analyze this garment image and extract precise sewing pattern parameters.
|
| 12 |
|
|
@@ -69,32 +70,23 @@ VISION_MODELS = [
|
|
| 69 |
("google/gemma-4-31B-it", "together", "Gemma 4 31B"),
|
| 70 |
("moonshotai/Kimi-K2.5", "together", "Kimi K2.5"),
|
| 71 |
]
|
| 72 |
-
TEXT_MODELS = VISION_MODELS
|
| 73 |
|
| 74 |
def _extract_response_text(message):
|
| 75 |
content = message.get('content', '') or ''
|
| 76 |
reasoning = message.get('reasoning', '') or ''
|
| 77 |
-
|
| 78 |
-
return content.strip()
|
| 79 |
-
if reasoning.strip():
|
| 80 |
-
return reasoning.strip()
|
| 81 |
-
return ''
|
| 82 |
|
| 83 |
def _extract_json_from_text(text):
|
| 84 |
json_match = re.search(r'```(?:json)?\s*([\s\S]*?)\s*```', text)
|
| 85 |
-
if json_match:
|
| 86 |
-
return json_match.group(1)
|
| 87 |
json_match = re.search(r'\{[\s\S]*\}', text)
|
| 88 |
-
if json_match:
|
| 89 |
-
return json_match.group()
|
| 90 |
return None
|
| 91 |
|
| 92 |
def _call_vlm(messages, timeout=180):
|
| 93 |
-
import requests
|
| 94 |
-
from io import BytesIO
|
| 95 |
hf_token = os.environ.get("HF_TOKEN", "")
|
| 96 |
-
if not hf_token:
|
| 97 |
-
return None
|
| 98 |
for model_id, provider, display_name in VISION_MODELS:
|
| 99 |
try:
|
| 100 |
url = f"https://router.huggingface.co/{provider}/v1/chat/completions"
|
|
@@ -105,28 +97,23 @@ def _call_vlm(messages, timeout=180):
|
|
| 105 |
if response.status_code == 200:
|
| 106 |
result = response.json()
|
| 107 |
text = _extract_response_text(result['choices'][0]['message'])
|
| 108 |
-
if not text:
|
| 109 |
-
continue
|
| 110 |
json_str = _extract_json_from_text(text)
|
| 111 |
-
if not json_str:
|
| 112 |
-
continue
|
| 113 |
analysis = json.loads(json_str)
|
| 114 |
-
analysis['_model_used'] =
|
| 115 |
print(f"[VLM] OK: {display_name} detected {analysis.get('garment_type','?')}")
|
| 116 |
return analysis
|
| 117 |
-
else:
|
| 118 |
-
print(f"[VLM] {display_name}: HTTP {response.status_code}")
|
| 119 |
except Exception as e:
|
| 120 |
-
print(f"[VLM] {display_name} failed: {e}")
|
| 121 |
-
continue
|
| 122 |
return None
|
| 123 |
|
| 124 |
def analyze_with_vlm(image):
|
| 125 |
import base64
|
| 126 |
from io import BytesIO
|
| 127 |
hf_token = os.environ.get("HF_TOKEN", "")
|
| 128 |
-
if not hf_token:
|
| 129 |
-
return None
|
| 130 |
max_dim = 1024
|
| 131 |
if max(image.size) > max_dim:
|
| 132 |
ratio = max_dim / max(image.size)
|
|
@@ -155,21 +142,13 @@ def get_default_analysis(garment_type="shirt"):
|
|
| 155 |
_current_analysis = {"data": None}
|
| 156 |
|
| 157 |
def _generate_all_outputs(analysis):
|
| 158 |
-
"""Generate 2D pattern, 3D figure (from pattern pieces), and summary."""
|
| 159 |
garment_type = analysis.get('garment_type', 'shirt')
|
| 160 |
measurements = analysis.get('measurements', {})
|
| 161 |
features = analysis.get('features', {})
|
| 162 |
params = {**measurements, **features}
|
| 163 |
-
|
| 164 |
-
# Generate 2D pattern pieces
|
| 165 |
pattern_pieces = get_pattern_pieces(garment_type, params)
|
| 166 |
-
|
| 167 |
-
# Generate 2D pattern image and summary
|
| 168 |
pattern_image, summary = generate_pattern_from_analysis(analysis)
|
| 169 |
-
|
| 170 |
-
# Generate 3D figure FROM the actual pattern pieces
|
| 171 |
fig_3d = create_3d_figure(analysis, pattern_pieces=pattern_pieces)
|
| 172 |
-
|
| 173 |
display = {k: v for k, v in analysis.items() if k != '_model_used'}
|
| 174 |
model_info = f"\n\n*AI: {analysis.get('_model_used', 'Default')}*" if analysis.get('_model_used') else ""
|
| 175 |
desc = analysis.get('description', 'No description')
|
|
@@ -181,10 +160,8 @@ def process_image(image, garment_type_override="Auto-detect"):
|
|
| 181 |
return None, None, "Please upload a garment image or select a type.", "{}", []
|
| 182 |
analysis = None
|
| 183 |
if image is not None:
|
| 184 |
-
try:
|
| 185 |
-
|
| 186 |
-
except Exception as e:
|
| 187 |
-
print(f"VLM failed: {e}")
|
| 188 |
if analysis is None:
|
| 189 |
gt = garment_type_override.lower() if garment_type_override != "Auto-detect" else "shirt"
|
| 190 |
analysis = get_default_analysis(gt)
|
|
@@ -201,33 +178,23 @@ def process_image(image, garment_type_override="Auto-detect"):
|
|
| 201 |
return None, None, f"Error: {e}", "{}", []
|
| 202 |
|
| 203 |
def process_text(description):
|
| 204 |
-
if not description.strip():
|
| 205 |
-
return None, None, "Enter a description.", "{}", []
|
| 206 |
analysis = None
|
| 207 |
hf_token = os.environ.get("HF_TOKEN", "")
|
| 208 |
if hf_token:
|
| 209 |
-
messages = [{"role": "user", "content": f"
|
| 210 |
-
|
| 211 |
-
Description: {description}
|
| 212 |
-
|
| 213 |
-
Return ONLY JSON with: garment_type, description, measurements (bust, waist, hip, shoulder_width, bodice_length, sleeve_length, skirt_length, pant_length, neckline_depth, neckline_width, bicep, wrist, cap_height, collar_height, flare), features (has_collar, collar_type, has_cuffs, has_pockets, pocket_type, has_hood, fit)."""}]
|
| 214 |
analysis = _call_vlm(messages, timeout=90)
|
| 215 |
if analysis is None:
|
| 216 |
desc_lower = description.lower()
|
| 217 |
for gt in ['hoodie','jacket','coat','blazer','dress','skirt','pants','trousers','jeans','vest','shirt','blouse','top']:
|
| 218 |
if gt in desc_lower:
|
| 219 |
-
analysis = get_default_analysis(gt)
|
| 220 |
-
|
| 221 |
-
break
|
| 222 |
-
if analysis is None:
|
| 223 |
-
analysis = get_default_analysis("shirt")
|
| 224 |
-
analysis['description'] = description
|
| 225 |
_current_analysis["data"] = copy.deepcopy(analysis)
|
| 226 |
try:
|
| 227 |
p2d, p3d, summary, j = _generate_all_outputs(analysis)
|
| 228 |
return p2d, p3d, summary, j, []
|
| 229 |
-
except Exception as e:
|
| 230 |
-
return None, None, f"Error: {e}", "{}", []
|
| 231 |
|
| 232 |
def process_manual(gt,bust,waist,hip,shoulder,bodice,sleeve,skirt,pant,neck,flare_c,collar,ctype,cuffs,pockets,hood,fit):
|
| 233 |
analysis = {"garment_type":gt.lower(),"description":f"Custom {gt.lower()}","measurements":{"bust":bust,"waist":waist,"hip":hip,"shoulder_width":shoulder,"bodice_length":bodice,"sleeve_length":sleeve,"skirt_length":skirt,"pant_length":pant,"neckline_depth":neck,"neckline_width":7,"bicep":30,"wrist":18,"cap_height":14,"collar_height":5,"flare":flare_c},"features":{"has_collar":collar,"collar_type":ctype.lower(),"has_cuffs":cuffs,"has_pockets":pockets,"pocket_type":"patch","has_hood":hood,"fit":fit.lower()}}
|
|
@@ -235,79 +202,133 @@ def process_manual(gt,bust,waist,hip,shoulder,bodice,sleeve,skirt,pant,neck,flar
|
|
| 235 |
try:
|
| 236 |
p2d, p3d, summary, j = _generate_all_outputs(analysis)
|
| 237 |
return p2d, p3d, summary, j, []
|
| 238 |
-
except Exception as e:
|
| 239 |
-
return None, None, f"Error: {e}", "{}", []
|
| 240 |
|
| 241 |
def chat_edit(message, history):
|
| 242 |
-
if not message.strip():
|
| 243 |
-
|
| 244 |
-
|
| 245 |
-
if current is None:
|
| 246 |
-
current = get_default_analysis("shirt")
|
| 247 |
-
_current_analysis["data"] = current
|
| 248 |
current_clean = {k: v for k, v in current.items() if k != '_model_used'}
|
| 249 |
-
edit_prompt = EDIT_PROMPT_TEMPLATE.format(
|
| 250 |
-
current_json=json.dumps(current_clean, indent=2), user_message=message)
|
| 251 |
updated = None
|
| 252 |
-
|
| 253 |
-
|
| 254 |
-
|
| 255 |
-
try:
|
| 256 |
-
updated = _call_vlm(messages, timeout=90)
|
| 257 |
-
except Exception as e:
|
| 258 |
-
print(f"Edit VLM failed: {e}")
|
| 259 |
if updated is None:
|
| 260 |
updated = copy.deepcopy(current)
|
| 261 |
msg_lower = message.lower()
|
| 262 |
-
if "long sleeve" in msg_lower
|
| 263 |
-
|
| 264 |
-
|
| 265 |
-
|
| 266 |
-
if "
|
| 267 |
-
|
| 268 |
-
if "
|
| 269 |
-
|
| 270 |
-
if "
|
| 271 |
-
updated['features']['has_hood'] = True
|
| 272 |
-
if "no hood" in msg_lower or "remove hood" in msg_lower:
|
| 273 |
-
updated['features']['has_hood'] = False
|
| 274 |
-
if "add pocket" in msg_lower:
|
| 275 |
-
updated['features']['has_pockets'] = True; updated['features']['pocket_type'] = 'patch'
|
| 276 |
-
if "no pocket" in msg_lower or "remove pocket" in msg_lower:
|
| 277 |
-
updated['features']['has_pockets'] = False
|
| 278 |
-
if "oversized" in msg_lower or "loose" in msg_lower:
|
| 279 |
-
updated['features']['fit'] = 'oversized'
|
| 280 |
-
updated['measurements']['bust'] = updated['measurements'].get('bust', 96) + 10
|
| 281 |
-
if "fitted" in msg_lower or "slim" in msg_lower:
|
| 282 |
-
updated['features']['fit'] = 'fitted'
|
| 283 |
-
if "flare" in msg_lower or "a-line" in msg_lower:
|
| 284 |
-
updated['measurements']['flare'] = max(updated['measurements'].get('flare', 0), 8)
|
| 285 |
-
if "straight" in msg_lower: updated['measurements']['flare'] = 0
|
| 286 |
-
if "mini" in msg_lower: updated['measurements']['skirt_length'] = 30
|
| 287 |
-
if "midi" in msg_lower: updated['measurements']['skirt_length'] = 80
|
| 288 |
-
if "maxi" in msg_lower: updated['measurements']['skirt_length'] = 110
|
| 289 |
updated['_model_used'] = 'Rule-based edit'
|
| 290 |
-
if 'garment_type' not in updated:
|
| 291 |
-
updated['garment_type'] = current.get('garment_type', 'shirt')
|
| 292 |
_current_analysis["data"] = copy.deepcopy(updated)
|
| 293 |
-
try:
|
| 294 |
-
|
| 295 |
-
|
| 296 |
-
p2d, p3d, summary, j = None, None, f"Error: {e}", "{}"
|
| 297 |
-
ai_msg = f"Applied edit: {message}\n\n"
|
| 298 |
changes = []
|
| 299 |
for k in set(list(current.get('measurements',{}).keys()) + list(updated.get('measurements',{}).keys())):
|
| 300 |
ov, nv = current.get('measurements',{}).get(k), updated.get('measurements',{}).get(k)
|
| 301 |
-
if ov != nv and ov is not None and nv is not None:
|
| 302 |
-
changes.append(f" {k}: {ov} -> {nv}")
|
| 303 |
for k in set(list(current.get('features',{}).keys()) + list(updated.get('features',{}).keys())):
|
| 304 |
ov, nv = current.get('features',{}).get(k), updated.get('features',{}).get(k)
|
| 305 |
-
if ov != nv: changes.append(f" {k}: {ov}
|
| 306 |
-
ai_msg += ("
|
| 307 |
-
history = history or []
|
| 308 |
-
history.append((message, ai_msg))
|
| 309 |
return history, p2d, p3d, summary, j
|
| 310 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 311 |
CSS = """
|
| 312 |
.main-header { text-align: center; margin-bottom: 20px; }
|
| 313 |
.info-box { padding: 15px; border-radius: 10px; background: #f0f7ff; border: 1px solid #cce0ff; margin: 10px 0; }
|
|
@@ -315,18 +336,17 @@ CSS = """
|
|
| 315 |
"""
|
| 316 |
|
| 317 |
with gr.Blocks(title="Garment Pattern Studio") as demo:
|
| 318 |
-
gr.HTML("""
|
| 319 |
-
<div class="main-header">
|
| 320 |
<h1>🧵 Garment Pattern Studio</h1>
|
| 321 |
-
<p style="font-size: 1.1em; color: #555;">Analyze garments, edit with chat, preview in 3D</p>
|
| 322 |
</div>
|
| 323 |
<div class="info-box">
|
| 324 |
-
<b>Powered by:</b> Qwen 3.5
|
| 325 |
<a href="https://huggingface.co/docs/inference-providers">HF Inference Providers</a>
|
| 326 |
-
| <b>3D view
|
| 327 |
</div>""")
|
| 328 |
|
| 329 |
-
with gr.Tab("From Image"):
|
| 330 |
with gr.Row():
|
| 331 |
with gr.Column(scale=1):
|
| 332 |
input_image = gr.Image(type="pil", label="Upload Garment Image", height=350)
|
|
@@ -337,10 +357,10 @@ with gr.Blocks(title="Garment Pattern Studio") as demo:
|
|
| 337 |
with gr.Column(): out_pattern_2d = gr.Image(label="2D Sewing Pattern", height=400)
|
| 338 |
with gr.Column(): out_3d = gr.Plot(label="3D Garment Preview")
|
| 339 |
out_summary = gr.Markdown(label="Pattern Summary")
|
| 340 |
-
with gr.Accordion("Raw JSON", open=False): out_json = gr.Code(language="json"
|
| 341 |
analyze_btn.click(process_image, inputs=[input_image, garment_override], outputs=[out_pattern_2d, out_3d, out_summary, out_json])
|
| 342 |
|
| 343 |
-
with gr.Tab("From Text"):
|
| 344 |
with gr.Row():
|
| 345 |
with gr.Column(scale=1):
|
| 346 |
text_input = gr.Textbox(label="Describe the garment", placeholder="e.g., A fitted A-line dress with cap sleeves", lines=3)
|
|
@@ -361,36 +381,20 @@ with gr.Blocks(title="Garment Pattern Studio") as demo:
|
|
| 361 |
with gr.Accordion("Raw JSON", open=False): txt_json = gr.Code(language="json")
|
| 362 |
text_btn.click(process_text, inputs=[text_input], outputs=[txt_pattern_2d, txt_3d, txt_summary, txt_json])
|
| 363 |
|
| 364 |
-
with gr.Tab("
|
| 365 |
with gr.Row():
|
| 366 |
with gr.Column(scale=1):
|
| 367 |
m_type = gr.Dropdown(choices=["Shirt","Dress","Skirt","Pants","Jacket","Hoodie","Vest"], value="Shirt", label="Garment Type")
|
| 368 |
gr.Markdown("### Measurements (cm)")
|
| 369 |
-
with gr.Row():
|
| 370 |
-
|
| 371 |
-
|
| 372 |
-
with gr.Row():
|
| 373 |
-
|
| 374 |
-
m_shoulder = gr.Slider(35,55,value=42,step=1,label="Shoulder")
|
| 375 |
-
with gr.Row():
|
| 376 |
-
m_bodice = gr.Slider(30,80,value=42,step=1,label="Bodice Length")
|
| 377 |
-
m_sleeve = gr.Slider(10,75,value=60,step=1,label="Sleeve Length")
|
| 378 |
-
with gr.Row():
|
| 379 |
-
m_skirt = gr.Slider(25,120,value=55,step=1,label="Skirt Length")
|
| 380 |
-
m_pant = gr.Slider(25,115,value=100,step=1,label="Pant Length")
|
| 381 |
-
with gr.Row():
|
| 382 |
-
m_neck = gr.Slider(3,25,value=8,step=1,label="Neckline Depth")
|
| 383 |
-
m_flare = gr.Slider(0,20,value=0,step=1,label="Hem Flare")
|
| 384 |
gr.Markdown("### Features")
|
| 385 |
-
with gr.Row():
|
| 386 |
-
|
| 387 |
-
|
| 388 |
-
with gr.Row():
|
| 389 |
-
m_cuffs = gr.Checkbox(value=True,label="Cuffs")
|
| 390 |
-
m_pockets = gr.Checkbox(value=False,label="Pockets")
|
| 391 |
-
with gr.Row():
|
| 392 |
-
m_hood = gr.Checkbox(value=False,label="Hood")
|
| 393 |
-
m_fit = gr.Dropdown(["Fitted","Regular","Oversized","Loose"],value="Regular",label="Fit")
|
| 394 |
manual_btn = gr.Button("Generate Pattern", variant="primary", size="lg")
|
| 395 |
with gr.Column(scale=2):
|
| 396 |
with gr.Row():
|
|
@@ -400,31 +404,60 @@ with gr.Blocks(title="Garment Pattern Studio") as demo:
|
|
| 400 |
with gr.Accordion("Raw JSON", open=False): man_json = gr.Code(language="json")
|
| 401 |
manual_btn.click(process_manual, inputs=[m_type,m_bust,m_waist,m_hip,m_shoulder,m_bodice,m_sleeve,m_skirt,m_pant,m_neck,m_flare,m_collar,m_ctype,m_cuffs,m_pockets,m_hood,m_fit], outputs=[man_pattern_2d, man_3d, man_summary, man_json])
|
| 402 |
|
| 403 |
-
with gr.Tab("Chat & Edit"):
|
| 404 |
-
gr.Markdown("### Edit
|
| 405 |
with gr.Row():
|
| 406 |
with gr.Column(scale=1):
|
| 407 |
chatbot = gr.Chatbot(label="Pattern Editor", height=400)
|
| 408 |
-
chat_input = gr.Textbox(label="Edit instruction", placeholder="e.g., Make
|
| 409 |
-
with gr.Row():
|
| 410 |
-
chat_send = gr.Button("Apply Edit", variant="primary")
|
| 411 |
-
chat_clear = gr.Button("Clear Chat")
|
| 412 |
with gr.Column(scale=2):
|
| 413 |
with gr.Row():
|
| 414 |
-
with gr.Column(): chat_pattern_2d = gr.Image(label="Updated 2D
|
| 415 |
-
with gr.Column(): chat_3d = gr.Plot(label="Updated 3D
|
| 416 |
chat_summary = gr.Markdown()
|
| 417 |
-
with gr.Accordion("
|
| 418 |
def clear_chat(): return [], None, None, "", "{}"
|
| 419 |
chat_send.click(chat_edit, inputs=[chat_input, chatbot], outputs=[chatbot, chat_pattern_2d, chat_3d, chat_summary, chat_json])
|
| 420 |
chat_input.submit(chat_edit, inputs=[chat_input, chatbot], outputs=[chatbot, chat_pattern_2d, chat_3d, chat_summary, chat_json])
|
| 421 |
chat_clear.click(clear_chat, outputs=[chatbot, chat_pattern_2d, chat_3d, chat_summary, chat_json])
|
| 422 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 423 |
gr.HTML("""<div class="ref-box" style="margin-top: 20px;"><h4>Research References</h4><ul>
|
| 424 |
-
<li><b>ChatGarment</b> (2024) [<a href="https://arxiv.org/abs/2412.17811">Paper</a>]
|
| 425 |
-
<li><b>NGL-Prompter</b> (2025) [<a href="https://arxiv.org/abs/2602.20700">Paper</a>]</li>
|
| 426 |
-
<li><b>
|
| 427 |
-
<li><b>
|
| 428 |
</ul></div>""")
|
| 429 |
|
| 430 |
if __name__ == "__main__":
|
|
|
|
| 1 |
"""
|
| 2 |
+
Garment Image -> 2D Sewing Pattern + Chat Editing + 3D Preview + Agentic Refinement
|
| 3 |
"""
|
| 4 |
import json, os, re, traceback, copy
|
| 5 |
from typing import Dict, Optional, Tuple, List
|
|
|
|
| 7 |
from PIL import Image
|
| 8 |
from pattern_generator import generate_pattern_from_analysis, get_pattern_pieces
|
| 9 |
from garment_3d import create_3d_figure
|
| 10 |
+
from refinement_loop import refinement_loop, render_3d_to_image
|
| 11 |
|
| 12 |
GARMENT_ANALYSIS_PROMPT = """You are a professional fashion pattern maker. Analyze this garment image and extract precise sewing pattern parameters.
|
| 13 |
|
|
|
|
| 70 |
("google/gemma-4-31B-it", "together", "Gemma 4 31B"),
|
| 71 |
("moonshotai/Kimi-K2.5", "together", "Kimi K2.5"),
|
| 72 |
]
|
|
|
|
| 73 |
|
| 74 |
def _extract_response_text(message):
|
| 75 |
content = message.get('content', '') or ''
|
| 76 |
reasoning = message.get('reasoning', '') or ''
|
| 77 |
+
return content.strip() or reasoning.strip() or ''
|
|
|
|
|
|
|
|
|
|
|
|
|
| 78 |
|
| 79 |
def _extract_json_from_text(text):
|
| 80 |
json_match = re.search(r'```(?:json)?\s*([\s\S]*?)\s*```', text)
|
| 81 |
+
if json_match: return json_match.group(1)
|
|
|
|
| 82 |
json_match = re.search(r'\{[\s\S]*\}', text)
|
| 83 |
+
if json_match: return json_match.group()
|
|
|
|
| 84 |
return None
|
| 85 |
|
| 86 |
def _call_vlm(messages, timeout=180):
|
| 87 |
+
import requests
|
|
|
|
| 88 |
hf_token = os.environ.get("HF_TOKEN", "")
|
| 89 |
+
if not hf_token: return None
|
|
|
|
| 90 |
for model_id, provider, display_name in VISION_MODELS:
|
| 91 |
try:
|
| 92 |
url = f"https://router.huggingface.co/{provider}/v1/chat/completions"
|
|
|
|
| 97 |
if response.status_code == 200:
|
| 98 |
result = response.json()
|
| 99 |
text = _extract_response_text(result['choices'][0]['message'])
|
| 100 |
+
if not text: continue
|
|
|
|
| 101 |
json_str = _extract_json_from_text(text)
|
| 102 |
+
if not json_str: continue
|
|
|
|
| 103 |
analysis = json.loads(json_str)
|
| 104 |
+
analysis['_model_used'] = display_name
|
| 105 |
print(f"[VLM] OK: {display_name} detected {analysis.get('garment_type','?')}")
|
| 106 |
return analysis
|
| 107 |
+
else: print(f"[VLM] {display_name}: HTTP {response.status_code}")
|
|
|
|
| 108 |
except Exception as e:
|
| 109 |
+
print(f"[VLM] {display_name} failed: {e}"); continue
|
|
|
|
| 110 |
return None
|
| 111 |
|
| 112 |
def analyze_with_vlm(image):
|
| 113 |
import base64
|
| 114 |
from io import BytesIO
|
| 115 |
hf_token = os.environ.get("HF_TOKEN", "")
|
| 116 |
+
if not hf_token: return None
|
|
|
|
| 117 |
max_dim = 1024
|
| 118 |
if max(image.size) > max_dim:
|
| 119 |
ratio = max_dim / max(image.size)
|
|
|
|
| 142 |
_current_analysis = {"data": None}
|
| 143 |
|
| 144 |
def _generate_all_outputs(analysis):
|
|
|
|
| 145 |
garment_type = analysis.get('garment_type', 'shirt')
|
| 146 |
measurements = analysis.get('measurements', {})
|
| 147 |
features = analysis.get('features', {})
|
| 148 |
params = {**measurements, **features}
|
|
|
|
|
|
|
| 149 |
pattern_pieces = get_pattern_pieces(garment_type, params)
|
|
|
|
|
|
|
| 150 |
pattern_image, summary = generate_pattern_from_analysis(analysis)
|
|
|
|
|
|
|
| 151 |
fig_3d = create_3d_figure(analysis, pattern_pieces=pattern_pieces)
|
|
|
|
| 152 |
display = {k: v for k, v in analysis.items() if k != '_model_used'}
|
| 153 |
model_info = f"\n\n*AI: {analysis.get('_model_used', 'Default')}*" if analysis.get('_model_used') else ""
|
| 154 |
desc = analysis.get('description', 'No description')
|
|
|
|
| 160 |
return None, None, "Please upload a garment image or select a type.", "{}", []
|
| 161 |
analysis = None
|
| 162 |
if image is not None:
|
| 163 |
+
try: analysis = analyze_with_vlm(image)
|
| 164 |
+
except Exception as e: print(f"VLM failed: {e}")
|
|
|
|
|
|
|
| 165 |
if analysis is None:
|
| 166 |
gt = garment_type_override.lower() if garment_type_override != "Auto-detect" else "shirt"
|
| 167 |
analysis = get_default_analysis(gt)
|
|
|
|
| 178 |
return None, None, f"Error: {e}", "{}", []
|
| 179 |
|
| 180 |
def process_text(description):
|
| 181 |
+
if not description.strip(): return None, None, "Enter a description.", "{}", []
|
|
|
|
| 182 |
analysis = None
|
| 183 |
hf_token = os.environ.get("HF_TOKEN", "")
|
| 184 |
if hf_token:
|
| 185 |
+
messages = [{"role": "user", "content": f"Based on this garment description, extract sewing pattern parameters.\n\nDescription: {description}\n\nReturn ONLY JSON with: garment_type, description, measurements (bust, waist, hip, shoulder_width, bodice_length, sleeve_length, skirt_length, pant_length, neckline_depth, neckline_width, bicep, wrist, cap_height, collar_height, flare), features (has_collar, collar_type, has_cuffs, has_pockets, pocket_type, has_hood, fit)."}]
|
|
|
|
|
|
|
|
|
|
|
|
|
| 186 |
analysis = _call_vlm(messages, timeout=90)
|
| 187 |
if analysis is None:
|
| 188 |
desc_lower = description.lower()
|
| 189 |
for gt in ['hoodie','jacket','coat','blazer','dress','skirt','pants','trousers','jeans','vest','shirt','blouse','top']:
|
| 190 |
if gt in desc_lower:
|
| 191 |
+
analysis = get_default_analysis(gt); analysis['description'] = description; break
|
| 192 |
+
if analysis is None: analysis = get_default_analysis("shirt"); analysis['description'] = description
|
|
|
|
|
|
|
|
|
|
|
|
|
| 193 |
_current_analysis["data"] = copy.deepcopy(analysis)
|
| 194 |
try:
|
| 195 |
p2d, p3d, summary, j = _generate_all_outputs(analysis)
|
| 196 |
return p2d, p3d, summary, j, []
|
| 197 |
+
except Exception as e: return None, None, f"Error: {e}", "{}", []
|
|
|
|
| 198 |
|
| 199 |
def process_manual(gt,bust,waist,hip,shoulder,bodice,sleeve,skirt,pant,neck,flare_c,collar,ctype,cuffs,pockets,hood,fit):
|
| 200 |
analysis = {"garment_type":gt.lower(),"description":f"Custom {gt.lower()}","measurements":{"bust":bust,"waist":waist,"hip":hip,"shoulder_width":shoulder,"bodice_length":bodice,"sleeve_length":sleeve,"skirt_length":skirt,"pant_length":pant,"neckline_depth":neck,"neckline_width":7,"bicep":30,"wrist":18,"cap_height":14,"collar_height":5,"flare":flare_c},"features":{"has_collar":collar,"collar_type":ctype.lower(),"has_cuffs":cuffs,"has_pockets":pockets,"pocket_type":"patch","has_hood":hood,"fit":fit.lower()}}
|
|
|
|
| 202 |
try:
|
| 203 |
p2d, p3d, summary, j = _generate_all_outputs(analysis)
|
| 204 |
return p2d, p3d, summary, j, []
|
| 205 |
+
except Exception as e: return None, None, f"Error: {e}", "{}", []
|
|
|
|
| 206 |
|
| 207 |
def chat_edit(message, history):
|
| 208 |
+
if not message.strip(): return history, None, None, "Please enter an edit request.", "{}"
|
| 209 |
+
current = _current_analysis.get("data") or get_default_analysis("shirt")
|
| 210 |
+
_current_analysis["data"] = current
|
|
|
|
|
|
|
|
|
|
| 211 |
current_clean = {k: v for k, v in current.items() if k != '_model_used'}
|
| 212 |
+
edit_prompt = EDIT_PROMPT_TEMPLATE.format(current_json=json.dumps(current_clean, indent=2), user_message=message)
|
|
|
|
| 213 |
updated = None
|
| 214 |
+
if os.environ.get("HF_TOKEN", ""):
|
| 215 |
+
try: updated = _call_vlm([{"role": "user", "content": edit_prompt}], timeout=90)
|
| 216 |
+
except: pass
|
|
|
|
|
|
|
|
|
|
|
|
|
| 217 |
if updated is None:
|
| 218 |
updated = copy.deepcopy(current)
|
| 219 |
msg_lower = message.lower()
|
| 220 |
+
if "long sleeve" in msg_lower: updated['measurements']['sleeve_length'] = 65
|
| 221 |
+
elif "short sleeve" in msg_lower: updated['measurements']['sleeve_length'] = 25
|
| 222 |
+
if "no collar" in msg_lower: updated['features']['has_collar'] = False; updated['features']['collar_type'] = 'none'
|
| 223 |
+
if "add collar" in msg_lower: updated['features']['has_collar'] = True; updated['features']['collar_type'] = 'standard'
|
| 224 |
+
if "add hood" in msg_lower: updated['features']['has_hood'] = True
|
| 225 |
+
if "no hood" in msg_lower: updated['features']['has_hood'] = False
|
| 226 |
+
if "oversized" in msg_lower: updated['features']['fit'] = 'oversized'; updated['measurements']['bust'] = updated['measurements'].get('bust', 96) + 10
|
| 227 |
+
if "fitted" in msg_lower: updated['features']['fit'] = 'fitted'
|
| 228 |
+
if "flare" in msg_lower: updated['measurements']['flare'] = max(updated['measurements'].get('flare', 0), 8)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 229 |
updated['_model_used'] = 'Rule-based edit'
|
| 230 |
+
if 'garment_type' not in updated: updated['garment_type'] = current.get('garment_type', 'shirt')
|
|
|
|
| 231 |
_current_analysis["data"] = copy.deepcopy(updated)
|
| 232 |
+
try: p2d, p3d, summary, j = _generate_all_outputs(updated)
|
| 233 |
+
except Exception as e: p2d, p3d, summary, j = None, None, f"Error: {e}", "{}"
|
| 234 |
+
ai_msg = f"Applied: {message}\n"
|
|
|
|
|
|
|
| 235 |
changes = []
|
| 236 |
for k in set(list(current.get('measurements',{}).keys()) + list(updated.get('measurements',{}).keys())):
|
| 237 |
ov, nv = current.get('measurements',{}).get(k), updated.get('measurements',{}).get(k)
|
| 238 |
+
if ov != nv and ov is not None and nv is not None: changes.append(f" {k}: {ov} → {nv}")
|
|
|
|
| 239 |
for k in set(list(current.get('features',{}).keys()) + list(updated.get('features',{}).keys())):
|
| 240 |
ov, nv = current.get('features',{}).get(k), updated.get('features',{}).get(k)
|
| 241 |
+
if ov != nv: changes.append(f" {k}: {ov} → {nv}")
|
| 242 |
+
ai_msg += ("\n".join(changes)) if changes else "No changes."
|
| 243 |
+
history = history or []; history.append((message, ai_msg))
|
|
|
|
| 244 |
return history, p2d, p3d, summary, j
|
| 245 |
|
| 246 |
+
# ── Agentic Refinement ──────────────────────────────────────────────────────
|
| 247 |
+
def run_refinement(image, garment_type_override, max_iters):
|
| 248 |
+
"""Run the agentic refinement loop."""
|
| 249 |
+
if image is None:
|
| 250 |
+
yield None, None, None, "Please upload a garment image.", "{}", None
|
| 251 |
+
return
|
| 252 |
+
|
| 253 |
+
# Step 1: Initial VLM analysis
|
| 254 |
+
analysis = None
|
| 255 |
+
try:
|
| 256 |
+
analysis = analyze_with_vlm(image)
|
| 257 |
+
except Exception as e:
|
| 258 |
+
print(f"VLM failed: {e}")
|
| 259 |
+
|
| 260 |
+
if analysis is None:
|
| 261 |
+
gt = garment_type_override.lower() if garment_type_override != "Auto-detect" else "shirt"
|
| 262 |
+
analysis = get_default_analysis(gt)
|
| 263 |
+
|
| 264 |
+
if garment_type_override != "Auto-detect":
|
| 265 |
+
analysis['garment_type'] = garment_type_override.lower()
|
| 266 |
+
|
| 267 |
+
# Generate function for the loop
|
| 268 |
+
def gen_fn(a):
|
| 269 |
+
return _generate_all_outputs(a)
|
| 270 |
+
|
| 271 |
+
# Run refinement loop
|
| 272 |
+
max_iters = int(max_iters)
|
| 273 |
+
result = refinement_loop(
|
| 274 |
+
original_image=image,
|
| 275 |
+
initial_analysis=analysis,
|
| 276 |
+
generate_fn=gen_fn,
|
| 277 |
+
max_iterations=max_iters,
|
| 278 |
+
target_composite=0.82,
|
| 279 |
+
plateau_patience=3,
|
| 280 |
+
lr=0.7,
|
| 281 |
+
)
|
| 282 |
+
|
| 283 |
+
# Build log markdown
|
| 284 |
+
log_lines = [f"## Refinement Results\n"]
|
| 285 |
+
log_lines.append(f"**Converged:** {'✅ Yes' if result['converged'] else '❌ No'}")
|
| 286 |
+
log_lines.append(f"**Iterations:** {result['total_iterations']}")
|
| 287 |
+
log_lines.append(f"**Best Score:** {result['best_score']:.4f}")
|
| 288 |
+
if result['scores']:
|
| 289 |
+
log_lines.append(f"**Score progression:** {' → '.join(f'{s:.3f}' for s in result['scores'])}")
|
| 290 |
+
log_lines.append("")
|
| 291 |
+
|
| 292 |
+
for step in result['history']:
|
| 293 |
+
it = step['iteration']
|
| 294 |
+
status = step.get('status', '?')
|
| 295 |
+
metrics = step.get('metrics', {})
|
| 296 |
+
log_lines.append(f"### Iteration {it} — {status}")
|
| 297 |
+
if metrics:
|
| 298 |
+
log_lines.append(f"SSIM={metrics.get('ssim',0):.3f} | Edge={metrics.get('edge_ssim',0):.3f} | Composite={metrics.get('composite',0):.3f}")
|
| 299 |
+
if step.get('new_best'):
|
| 300 |
+
log_lines.append("⭐ **New best!**")
|
| 301 |
+
diffs = step.get('vlm_differences', [])
|
| 302 |
+
if diffs:
|
| 303 |
+
log_lines.append("**Differences:** " + "; ".join(diffs[:3]))
|
| 304 |
+
adj = step.get('adjustments', {})
|
| 305 |
+
if adj:
|
| 306 |
+
log_lines.append("**Adjustments:** " + ", ".join(f"{k}={v}" for k, v in adj.items()))
|
| 307 |
+
reason = step.get('reason', '')
|
| 308 |
+
if reason:
|
| 309 |
+
log_lines.append(f"*{reason}*")
|
| 310 |
+
log_lines.append("")
|
| 311 |
+
|
| 312 |
+
log_md = "\n".join(log_lines)
|
| 313 |
+
|
| 314 |
+
# Get best outputs
|
| 315 |
+
best = result['best_analysis']
|
| 316 |
+
_current_analysis["data"] = copy.deepcopy(best)
|
| 317 |
+
try:
|
| 318 |
+
p2d, p3d, summary, j = _generate_all_outputs(best)
|
| 319 |
+
except:
|
| 320 |
+
p2d, p3d, summary, j = None, None, "Error generating final outputs", "{}"
|
| 321 |
+
|
| 322 |
+
# Get last projection
|
| 323 |
+
last_proj = None
|
| 324 |
+
for step in reversed(result['history']):
|
| 325 |
+
if 'projection' in step:
|
| 326 |
+
last_proj = step['projection']
|
| 327 |
+
break
|
| 328 |
+
|
| 329 |
+
yield p2d, p3d, last_proj, log_md, j, summary
|
| 330 |
+
|
| 331 |
+
# ── UI ──────────────────────────────────────────────────────────────────────
|
| 332 |
CSS = """
|
| 333 |
.main-header { text-align: center; margin-bottom: 20px; }
|
| 334 |
.info-box { padding: 15px; border-radius: 10px; background: #f0f7ff; border: 1px solid #cce0ff; margin: 10px 0; }
|
|
|
|
| 336 |
"""
|
| 337 |
|
| 338 |
with gr.Blocks(title="Garment Pattern Studio") as demo:
|
| 339 |
+
gr.HTML("""<div class="main-header">
|
|
|
|
| 340 |
<h1>🧵 Garment Pattern Studio</h1>
|
| 341 |
+
<p style="font-size: 1.1em; color: #555;">Analyze garments, edit with chat, preview in 3D, refine with AI agent</p>
|
| 342 |
</div>
|
| 343 |
<div class="info-box">
|
| 344 |
+
<b>Powered by:</b> Qwen 3.5 · Gemma 4 · Kimi K2.5 via
|
| 345 |
<a href="https://huggingface.co/docs/inference-providers">HF Inference Providers</a>
|
| 346 |
+
| <b>3D view built from actual 2D pattern pieces</b>
|
| 347 |
</div>""")
|
| 348 |
|
| 349 |
+
with gr.Tab("📸 From Image"):
|
| 350 |
with gr.Row():
|
| 351 |
with gr.Column(scale=1):
|
| 352 |
input_image = gr.Image(type="pil", label="Upload Garment Image", height=350)
|
|
|
|
| 357 |
with gr.Column(): out_pattern_2d = gr.Image(label="2D Sewing Pattern", height=400)
|
| 358 |
with gr.Column(): out_3d = gr.Plot(label="3D Garment Preview")
|
| 359 |
out_summary = gr.Markdown(label="Pattern Summary")
|
| 360 |
+
with gr.Accordion("Raw JSON", open=False): out_json = gr.Code(language="json")
|
| 361 |
analyze_btn.click(process_image, inputs=[input_image, garment_override], outputs=[out_pattern_2d, out_3d, out_summary, out_json])
|
| 362 |
|
| 363 |
+
with gr.Tab("✍️ From Text"):
|
| 364 |
with gr.Row():
|
| 365 |
with gr.Column(scale=1):
|
| 366 |
text_input = gr.Textbox(label="Describe the garment", placeholder="e.g., A fitted A-line dress with cap sleeves", lines=3)
|
|
|
|
| 381 |
with gr.Accordion("Raw JSON", open=False): txt_json = gr.Code(language="json")
|
| 382 |
text_btn.click(process_text, inputs=[text_input], outputs=[txt_pattern_2d, txt_3d, txt_summary, txt_json])
|
| 383 |
|
| 384 |
+
with gr.Tab("📐 Manual"):
|
| 385 |
with gr.Row():
|
| 386 |
with gr.Column(scale=1):
|
| 387 |
m_type = gr.Dropdown(choices=["Shirt","Dress","Skirt","Pants","Jacket","Hoodie","Vest"], value="Shirt", label="Garment Type")
|
| 388 |
gr.Markdown("### Measurements (cm)")
|
| 389 |
+
with gr.Row(): m_bust = gr.Slider(70,130,value=92,step=1,label="Bust"); m_waist = gr.Slider(55,110,value=74,step=1,label="Waist")
|
| 390 |
+
with gr.Row(): m_hip = gr.Slider(75,130,value=96,step=1,label="Hip"); m_shoulder = gr.Slider(35,55,value=42,step=1,label="Shoulder")
|
| 391 |
+
with gr.Row(): m_bodice = gr.Slider(30,80,value=42,step=1,label="Bodice Length"); m_sleeve = gr.Slider(10,75,value=60,step=1,label="Sleeve Length")
|
| 392 |
+
with gr.Row(): m_skirt = gr.Slider(25,120,value=55,step=1,label="Skirt Length"); m_pant = gr.Slider(25,115,value=100,step=1,label="Pant Length")
|
| 393 |
+
with gr.Row(): m_neck = gr.Slider(3,25,value=8,step=1,label="Neckline Depth"); m_flare = gr.Slider(0,20,value=0,step=1,label="Hem Flare")
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 394 |
gr.Markdown("### Features")
|
| 395 |
+
with gr.Row(): m_collar = gr.Checkbox(value=True,label="Collar"); m_ctype = gr.Dropdown(["Standard","Mandarin","Peter_pan"],value="Standard",label="Collar Type")
|
| 396 |
+
with gr.Row(): m_cuffs = gr.Checkbox(value=True,label="Cuffs"); m_pockets = gr.Checkbox(value=False,label="Pockets")
|
| 397 |
+
with gr.Row(): m_hood = gr.Checkbox(value=False,label="Hood"); m_fit = gr.Dropdown(["Fitted","Regular","Oversized","Loose"],value="Regular",label="Fit")
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 398 |
manual_btn = gr.Button("Generate Pattern", variant="primary", size="lg")
|
| 399 |
with gr.Column(scale=2):
|
| 400 |
with gr.Row():
|
|
|
|
| 404 |
with gr.Accordion("Raw JSON", open=False): man_json = gr.Code(language="json")
|
| 405 |
manual_btn.click(process_manual, inputs=[m_type,m_bust,m_waist,m_hip,m_shoulder,m_bodice,m_sleeve,m_skirt,m_pant,m_neck,m_flare,m_collar,m_ctype,m_cuffs,m_pockets,m_hood,m_fit], outputs=[man_pattern_2d, man_3d, man_summary, man_json])
|
| 406 |
|
| 407 |
+
with gr.Tab("💬 Chat & Edit"):
|
| 408 |
+
gr.Markdown("### Edit pattern with natural language\nGenerate a pattern first, then refine here.")
|
| 409 |
with gr.Row():
|
| 410 |
with gr.Column(scale=1):
|
| 411 |
chatbot = gr.Chatbot(label="Pattern Editor", height=400)
|
| 412 |
+
chat_input = gr.Textbox(label="Edit instruction", placeholder="e.g., Make sleeves longer, Add a hood", lines=2)
|
| 413 |
+
with gr.Row(): chat_send = gr.Button("Apply Edit", variant="primary"); chat_clear = gr.Button("Clear")
|
|
|
|
|
|
|
| 414 |
with gr.Column(scale=2):
|
| 415 |
with gr.Row():
|
| 416 |
+
with gr.Column(): chat_pattern_2d = gr.Image(label="Updated 2D", height=400)
|
| 417 |
+
with gr.Column(): chat_3d = gr.Plot(label="Updated 3D")
|
| 418 |
chat_summary = gr.Markdown()
|
| 419 |
+
with gr.Accordion("JSON", open=False): chat_json = gr.Code(language="json")
|
| 420 |
def clear_chat(): return [], None, None, "", "{}"
|
| 421 |
chat_send.click(chat_edit, inputs=[chat_input, chatbot], outputs=[chatbot, chat_pattern_2d, chat_3d, chat_summary, chat_json])
|
| 422 |
chat_input.submit(chat_edit, inputs=[chat_input, chatbot], outputs=[chatbot, chat_pattern_2d, chat_3d, chat_summary, chat_json])
|
| 423 |
chat_clear.click(clear_chat, outputs=[chatbot, chat_pattern_2d, chat_3d, chat_summary, chat_json])
|
| 424 |
|
| 425 |
+
with gr.Tab("🔄 Agentic Refinement"):
|
| 426 |
+
gr.Markdown("""### Iterative Refinement Loop
|
| 427 |
+
Upload a garment image. The AI agent will:
|
| 428 |
+
1. **Analyze** → extract initial pattern parameters
|
| 429 |
+
2. **Generate** → create 2D pattern + 3D garment
|
| 430 |
+
3. **Project** → render 3D to 2D front view
|
| 431 |
+
4. **Compare** → measure similarity (SSIM + VLM visual comparison)
|
| 432 |
+
5. **Refine** → VLM suggests parameter adjustments
|
| 433 |
+
6. **Repeat** until convergence or max iterations
|
| 434 |
+
|
| 435 |
+
*Requires HF_TOKEN for VLM-powered refinement.*""")
|
| 436 |
+
with gr.Row():
|
| 437 |
+
with gr.Column(scale=1):
|
| 438 |
+
refine_image = gr.Image(type="pil", label="Upload Garment Image", height=300)
|
| 439 |
+
refine_type = gr.Dropdown(choices=["Auto-detect","Shirt","Dress","Skirt","Pants","Jacket","Hoodie","Vest"], value="Auto-detect", label="Garment Type")
|
| 440 |
+
refine_iters = gr.Slider(1, 15, value=5, step=1, label="Max Iterations")
|
| 441 |
+
refine_btn = gr.Button("🚀 Start Refinement", variant="primary", size="lg")
|
| 442 |
+
with gr.Column(scale=2):
|
| 443 |
+
with gr.Row():
|
| 444 |
+
with gr.Column(): refine_2d = gr.Image(label="Best 2D Pattern", height=350)
|
| 445 |
+
with gr.Column(): refine_proj = gr.Image(label="3D→2D Projection", height=350)
|
| 446 |
+
with gr.Row():
|
| 447 |
+
with gr.Column(): refine_3d = gr.Plot(label="Best 3D Preview")
|
| 448 |
+
with gr.Column(): refine_log = gr.Markdown(label="Refinement Log")
|
| 449 |
+
refine_summary = gr.Markdown()
|
| 450 |
+
with gr.Accordion("Best Parameters JSON", open=False): refine_json = gr.Code(language="json")
|
| 451 |
+
|
| 452 |
+
refine_btn.click(run_refinement,
|
| 453 |
+
inputs=[refine_image, refine_type, refine_iters],
|
| 454 |
+
outputs=[refine_2d, refine_3d, refine_proj, refine_log, refine_json, refine_summary])
|
| 455 |
+
|
| 456 |
gr.HTML("""<div class="ref-box" style="margin-top: 20px;"><h4>Research References</h4><ul>
|
| 457 |
+
<li><b>ChatGarment</b> (2024) — VLM + dialogue for garment editing [<a href="https://arxiv.org/abs/2412.17811">Paper</a>]</li>
|
| 458 |
+
<li><b>NGL-Prompter</b> (2025) — Training-free VLM pattern estimation [<a href="https://arxiv.org/abs/2602.20700">Paper</a>]</li>
|
| 459 |
+
<li><b>RRVF</b> (2025) — Render-compare visual feedback loops [<a href="https://arxiv.org/abs/2507.20766">Paper</a>]</li>
|
| 460 |
+
<li><b>SceneAssistant</b> (2026) — Agentic VLM scene refinement [<a href="https://arxiv.org/abs/2603.12238">Paper</a>]</li>
|
| 461 |
</ul></div>""")
|
| 462 |
|
| 463 |
if __name__ == "__main__":
|