ClothIron / app.py
Mihir Mithani
Initial commit: AI Virtual Garment Ironing System
c7159bc
Raw History Blame Contribute Delete
12.8 kB
"""
AI Virtual Garment Ironing System
HuggingFace Spaces Entry Point
"""
import gradio as gr
import numpy as np
from PIL import Image
import torch
from models.segmentation import GarmentSegmenter
from models.wrinkle_detector import WrinkleDetector
from models.ironing_pipeline import IroningPipeline
from utils.image_utils import (
resize_for_processing,
restore_original_size,
create_comparison_image,
overlay_wrinkle_mask,
)
from utils.analysis import analyze_wrinkles, generate_analysis_report
# ── Global model instances (loaded once at startup) ──────────────────────────
segmenter = None
detector = None
iron_pipeline = None
def load_models():
global segmenter, detector, iron_pipeline
print("Loading models…")
segmenter = GarmentSegmenter()
detector = WrinkleDetector()
iron_pipeline = IroningPipeline()
print("All models loaded.")
# ── Core processing function ──────────────────────────────────────────────────
def process_image(
input_image: np.ndarray,
intensity: str,
fabric_type: str,
clothing_type: str,
preserve_folds: bool,
show_mask: bool,
) -> tuple:
if input_image is None:
return None, None, None, "Please upload an image first."
try:
if segmenter is None or detector is None or iron_pipeline is None:
load_models()
pil_img = Image.fromarray(input_image.astype(np.uint8)).convert("RGB")
proc_img, scale_info = resize_for_processing(pil_img, max_size=1024)
seg_result = segmenter.segment(proc_img, clothing_type=clothing_type)
garment_mask = seg_result["garment_mask"]
garment_labels = seg_result["labels"]
if not garment_mask.any():
return (
None, None, None,
"No garment was detected. Try a clearer photo or choose the clothing type.",
)
wrinkle_result = detector.detect(proc_img, garment_mask)
wrinkle_map = wrinkle_result["wrinkle_map"]
wrinkle_score = wrinkle_result["wrinkle_score"]
wrinkle_zones = wrinkle_result["zones"]
intensity_map = {"Light": 0.45, "Medium": 0.8, "Professional Press": 1.0}
intensity_val = intensity_map.get(intensity, 0.8)
ironed_proc = iron_pipeline.iron(
image=proc_img,
garment_mask=garment_mask,
wrinkle_map=wrinkle_map,
intensity=intensity_val,
fabric_type=fabric_type,
preserve_structural_folds=preserve_folds,
)
ironed_full = restore_original_size(ironed_proc, pil_img, scale_info)
ironed_np = np.array(ironed_full)
comparison_np = create_comparison_image(pil_img, ironed_full)
mask_vis_np = None
if show_mask:
mask_vis_np = overlay_wrinkle_mask(pil_img, wrinkle_map, garment_mask)
analysis = analyze_wrinkles(
wrinkle_score=wrinkle_score,
zones=wrinkle_zones,
labels=garment_labels,
fabric=fabric_type,
intensity=intensity,
)
report = generate_analysis_report(analysis)
if seg_result["confidence"] < 0.45:
report = (
"Warning: garment segmentation confidence is low. "
"Use a clearer, well-lit photo for a cleaner result.\n\n" + report
)
return ironed_np, comparison_np, mask_vis_np, report
except Exception as exc:
import traceback
print(traceback.format_exc())
return None, None, None, f"Error during processing:\n{str(exc)}"
# ── Gradio UI ─────────────────────────────────────────────────────────────────
def build_ui() -> gr.Blocks:
css = """
.title-row { text-align: center; padding: 1.5rem 0 0.5rem; }
.title-row h1 { font-size: 2rem; font-weight: 700; color: #1a1a2e; }
.title-row p { color: #555; font-size: 1rem; }
.result-col { background: #f9f9f9; border-radius: 12px; padding: 1rem; }
footer { display: none !important; }
"""
with gr.Blocks(title="AI Virtual Garment Ironing") as demo:
with gr.Row(elem_classes="title-row"):
gr.Markdown(
"""
# πŸ‘” AI Virtual Garment Ironing System
Upload a photo of yourself wearing clothes β€” AI detects wrinkles,
segments your garment, and generates a freshly ironed result.
"""
)
with gr.Row():
# ── Left: inputs ──────────────────────────────────────────────────
with gr.Column(scale=1):
input_image = gr.Image(
label="Upload photo",
type="numpy",
sources=["upload", "webcam"],
height=380,
)
with gr.Group():
gr.Markdown("### βš™οΈ Ironing settings")
intensity = gr.Radio(
choices=["Light", "Medium", "Professional Press"],
value="Medium",
label="Ironing intensity",
)
with gr.Row():
fabric_type = gr.Dropdown(
choices=[
"Auto-detect", "Cotton", "Linen", "Silk",
"Denim", "Polyester", "Wool", "Synthetic blend",
],
value="Auto-detect",
label="Fabric type",
)
clothing_type = gr.Dropdown(
choices=[
"Auto-detect", "Shirt / Blouse", "T-Shirt",
"Trousers", "Dress", "Jacket / Blazer",
"Traditional wear", "Saree / Kurta",
],
value="Auto-detect",
label="Clothing type",
)
with gr.Row():
preserve_folds = gr.Checkbox(
value=True,
label="Preserve natural body folds",
info="Keep folds caused by posture & movement",
)
show_mask = gr.Checkbox(
value=False,
label="Show wrinkle heatmap",
)
iron_btn = gr.Button(
"✨ Iron this garment",
variant="primary",
size="lg",
)
gr.Examples(
examples=[
["assets/example_shirt.jpg", "Medium", "Cotton", "Shirt / Blouse", True, False],
["assets/example_tshirt.jpg", "Light", "Polyester", "T-Shirt", True, False],
["assets/example_dress.jpg", "Professional Press", "Linen", "Dress", True, True ],
["assets/example_trousers.jpg", "Medium", "Denim", "Trousers", False, False],
],
inputs=[input_image, intensity, fabric_type, clothing_type, preserve_folds, show_mask],
label="Try an example",
cache_examples=False,
)
# ── Right: outputs ────────────────────────────────────────────────
with gr.Column(scale=1, elem_classes="result-col"):
gr.Markdown("### πŸ“Έ Results")
with gr.Tabs():
with gr.TabItem("Ironed result"):
output_ironed = gr.Image(
label="Ironed garment",
type="numpy",
height=380,
)
with gr.TabItem("Before / After"):
output_comparison = gr.Image(
label="Side-by-side comparison",
type="numpy",
height=380,
)
with gr.TabItem("Wrinkle map"):
output_mask = gr.Image(
label="Wrinkle heatmap overlay",
type="numpy",
height=380,
)
with gr.TabItem("Analysis"):
output_analysis = gr.Textbox(
label="Garment analysis report",
lines=18,
max_lines=30,
)
# ── Batch processing ──────────────────────────────────────────────────
with gr.Accordion("πŸ“¦ Batch processing (multiple images)", open=False):
gr.Markdown(
"Upload multiple images for batch ironing. "
"All images will use the settings chosen above."
)
batch_files = gr.File(file_count="multiple", label="Upload images")
batch_btn = gr.Button("Process batch", variant="secondary")
batch_gallery = gr.Gallery(label="Batch results", columns=3, height=400)
# ── How it works ──────────────────────────────────────────────────────
with gr.Accordion("ℹ️ How it works", open=False):
gr.Markdown(
"""
**Processing pipeline**
1. **Garment segmentation** β€” SegFormer isolates clothing pixels
2. **Wrinkle detection** β€” multi-scale edge + CNN maps crease density
3. **Structural fold preservation** β€” body-movement folds are kept
4. **Fabric-aware smoothing** β€” targeted luminance correction
5. **Texture reconstruction** β€” logos, patterns, buttons preserved
6. **Lighting / shadow blending** β€” composited back onto original scene
**Tips for best results**
- Use a well-lit, clear photo
- Ensure the full garment is visible
- Higher resolution input β†’ sharper output
"""
)
# ── Event handlers ────────────────────────────────────────────────────
iron_btn.click(
fn=process_image,
inputs=[input_image, intensity, fabric_type, clothing_type, preserve_folds, show_mask],
outputs=[output_ironed, output_comparison, output_mask, output_analysis],
)
def process_batch(files, intensity_val, fabric, clothing, preserve, mask):
if not files:
return []
results = []
for f in files:
img = np.array(Image.open(f.name).convert("RGB"))
ironed, _, _, _ = process_image(img, intensity_val, fabric, clothing, preserve, mask)
if ironed is not None:
results.append(ironed)
return results
batch_btn.click(
fn=process_batch,
inputs=[batch_files, intensity, fabric_type, clothing_type, preserve_folds, show_mask],
outputs=[batch_gallery],
)
return demo
# ── Entry point ───────────────────────────────────────────────────────────────
if __name__ == "__main__":
load_models()
demo = build_ui()
demo.launch(
server_name="0.0.0.0",
server_port=7860,
share=False,
show_error=True,
theme=gr.themes.Soft(
primary_hue="teal",
secondary_hue="slate",
font=[gr.themes.GoogleFont("Inter"), "sans-serif"],
),
)