import gradio as gr from collections import Counter from transformers import AutoImageProcessor, AutoModelForImageClassification from PIL import Image import torch import math # ============================== (همان پارامترها و توابع قبلی) material_params = { "brick": {"alpha": 0.3, "eps": 0.9, "I": 1600}, "stone": {"alpha": 0.25, "eps": 0.92, "I": 2000}, "polishedstone": {"alpha": 0.2, "eps": 0.9, "I": 2100}, "concrete": {"alpha": 0.35, "eps": 0.9, "I": 1800}, "metal": {"alpha": 0.5, "eps": 0.2, "I": 4000}, "glass": {"alpha": 0.1, "eps": 0.85, "I": 1500}, "wood": {"alpha": 0.35, "eps": 0.9, "I": 800}, "tile": {"alpha": 0.4, "eps": 0.9, "I": 1200}, "ceramic": {"alpha": 0.45, "eps": 0.92, "I": 1300}, "painted": {"alpha": 0.3, "eps": 0.9, "I": 1000}, "plastic": {"alpha": 0.1, "eps": 0.95, "I": 800}, "paper": {"alpha": 0.6, "eps": 0.95, "I": 500}, "mirror": {"alpha": 0.7, "eps": 0.1, "I": 2000}, "foliage": {"alpha": 0.25, "eps": 0.98, "I": 900}, "water": {"alpha": 0.06, "eps": 0.98, "I": 4200}, } material_categories = { "facade": {"members": ["brick", "stone", "polishedstone", "concrete", "tile", "ceramic", "painted"], "candidates": ["brick", "stone", "polishedstone", "concrete", "tile", "ceramic", "painted"]}, "glazing": {"members": ["glass", "mirror"], "candidates": ["glass", "mirror"]}, "metallic": {"members": ["metal"], "candidates": ["metal"]}, "coverings": {"members": ["plastic", "paper", "fabric"], "candidates": ["plastic", "paper", "fabric"]}, "wood_elements": {"members": ["wood"], "candidates": ["wood"]}, "vegetation": {"members": ["foliage"], "candidates": ["foliage"]}, "water_bodies": {"members": ["water"], "candidates": ["water"]}, } replacement_text = { "facade": {"brick": "آجر روشن یا نمای سرامیکی/تایل روشن با پوشش بازتابی (cool coating)", "stone": "سنگ روشن یا سنگ با پوشش بازتابی", "polishedstone": "سنگ مات روشن یا سرامیک نما روشن", "concrete": "بتن روشن با پوشش بازتابی یا موزاییک نما روشن", "tile": "کاشی/سرامیک روشن یا متخلخل", "ceramic": "سرامیک روشن با نمای بازتابی", "painted": "رنگ بازتابی (cool paint) یا پوشش نانو بازتابی"}, "glazing": {"glass": "شیشه دو جداره با پوشش Low-E یا شیشه بازتابی کنترل‌شده", "mirror": "شیشه مات یا شیشه Low-E با فریم عایق"}, "metallic": {"metal": "آلومینیوم رنگ روشن یا پوشش پودری با بازتاب بالا"}, "coverings": {"plastic": "سنگ سبک یا چوب روکش‌دار روشن (بسته به کاربرد)", "paper": "در نما کاربرد معمول ندارد - بررسی بهینه‌سازی طراحی", "fabric": "پارچه با روکش بازتابی یا سایه‌انداز طبیعی"}, "wood_elements": {"wood": "چوب رنگ روشن یا چوب با روکش بازتابی/محافظ"}, "vegetation": {"foliage": None}, "water_bodies": {"water": None}, } # ============================== (توابع کمکی) def ET_proxy(T, RH): es = 0.6108 * math.exp((17.27 * T) / (T + 237.3)) return es * (1 - RH / 100.0) def calc_deltaT(material, T_air, RH=40, u=2, S=700): if material not in material_params: return 0.0 alpha, eps, I = material_params[material]["alpha"], material_params[material]["eps"], material_params[material]["I"] A, B, C, D = 1.0, 0.4, 0.8, 0.015 h_c = 5.8 + 4.1 * u if material == "foliage": C_m = A * (1 - alpha) - D * ET_proxy(T_air, RH) else: C_m = A * (1 - alpha) + B * (1 - eps) + (C / math.sqrt(max(I, 1))) gamma = S / max(h_c, 1e-6) return gamma * C_m / 1000.0 # ============================== (بارگذاری مدل) model_id = "prithivMLmods/Minc-Materials-23" processor = AutoImageProcessor.from_pretrained(model_id) model = AutoModelForImageClassification.from_pretrained(model_id) patch_size = 224 def get_patches(image, size=224, stride=100): patches = [] w, h = image.size for scale in [1.0, 0.75, 0.5]: scaled_w, scaled_h = int(w * scale), int(h * scale) if min(scaled_w, scaled_h) < size: continue scaled_img = image.resize((scaled_w, scaled_h), Image.Resampling.LANCZOS) for i in range(0, scaled_w, stride): for j in range(0, scaled_h, stride): box = (i, j, min(i+size, scaled_w), min(j+size, scaled_h)) patch = scaled_img.crop(box) if patch.size[0] >= size and patch.size[1] >= size: patches.append(patch) return patches # ============================== (تابع اصلی Gradio) def analyze_image(image, T_air=32.0, RH=40, u=2.0, S=700): patches = get_patches(image, size=patch_size) all_predictions = [] for patch in patches: inputs = processor(images=patch, return_tensors="pt") with torch.no_grad(): outputs = model(**inputs) probs = torch.nn.functional.softmax(outputs.logits, dim=-1) top1 = torch.argmax(probs[0]).item() label = model.config.id2label[top1] all_predictions.append(label) counter = Counter(all_predictions) total_patches = len(patches) MIN_COUNT = 3 ignore_classes = ["food", "skin", "other", "wallpaper", "carpet","sky"] materials_found = {label for label, count in counter.items() if count >= MIN_COUNT and label not in ignore_classes} if len(materials_found) == 0: return "هیچ مصالح معتبرِ کافی در تصویر شناسایی نشد (حداقل تکرار MIN_COUNT رعایت نمی‌شود)." material_info = {} for label in sorted(materials_found): count = counter[label] share = count / total_patches dT = calc_deltaT(label, T_air, RH, u, S) material_info[label] = {"count": count, "share": share, "deltaT": dT} # مقایسه درون‌دسته‌ای و توصیه IMPROVEMENT_THRESHOLD = 0.02 SHARE_IMPORTANCE_THRESHOLD = 0.03 recommendations = [] candidate_delta_cache = {} for cat, info in material_categories.items(): for candidate in info["candidates"]: if candidate not in candidate_delta_cache: candidate_delta_cache[candidate] = calc_deltaT(candidate, T_air, RH, u, S) for label, info in material_info.items(): found_category = None for cat, cinfo in material_categories.items(): if label in cinfo["members"]: found_category = cat break if found_category is None: recommendations.append(f"{label}: در دسته‌های پیش‌تعریف قرار ندارد.") continue candidates = material_categories[found_category]["candidates"] cand_list = [(c, candidate_delta_cache.get(c, calc_deltaT(c, T_air, RH, u, S))) for c in candidates] cand_list.sort(key=lambda x: x[1]) current_dT = info["deltaT"] best_candidate, best_dT = cand_list[0] improvement = current_dT - best_dT share_pct = info["share"] * 100 if improvement >= IMPROVEMENT_THRESHOLD and best_candidate != label: importance = "High" if info["share"] >= SHARE_IMPORTANCE_THRESHOLD else "Optional" suggestion_text = replacement_text.get(found_category, {}).get(best_candidate, f"Consider replacing with {best_candidate}") recommendations.append( f"{label} ({found_category}): ΔT={current_dT:+.2f}°C → جایگزین: {best_candidate} (ΔT={best_dT:+.2f}°C) | بهبود: {improvement:+.2f}°C | اهمیت: {importance} | پیشنهاد: {suggestion_text}" ) else: recommendations.append(f"{label}: ΔT={current_dT:+.2f}°C → نیازی به جایگزینی ندارد.") scene_deltaT = sum([info["share"] * info["deltaT"] for info in material_info.values()]) recommendations.append(f"ΔT میانگین وزنی کل صحنه: {scene_deltaT:+.2f}°C") recommendations.append(f"دمای مؤثر سطح: {T_air + scene_deltaT:.2f}°C") return "\n".join(recommendations) # ============================== (راه‌اندازی رابط Gradio) iface = gr.Interface( fn=analyze_image, inputs=[ gr.Image(type="pil", label="آپلود تصویر"), gr.Number(value=32.0, label="دمای هوا T_air (°C)"), gr.Number(value=40, label="رطوبت نسبی RH (%)"), gr.Number(value=2.0, label="سرعت باد u (m/s)"), gr.Number(value=700, label="تابش خورشیدی S (W/m²)") ], outputs=gr.Textbox(label="خروجی ΔT و توصیه‌ها"), title="تحلیل مصالح و ΔT سطحی", description="آپلود تصویر ساختمان/محیط → نمایش ΔT مصالح و توصیه جایگزینی منطقی." ) iface.launch()