Update app.py

app.py

@@ -1,40 +1,77 @@
 import io
 import time
+import numpy as np
 import torch
+import torch.nn.functional as F
 from PIL import Image
-from ultralytics import YOLO
+from huggingface_hub import hf_hub_download
+from torchvision import transforms
 from fastapi import FastAPI, HTTPException, UploadFile, File
 from fastapi.middleware.cors import CORSMiddleware
 from contextlib import asynccontextmanager
 
-# ─── أسماء الـ 17 Keypoints ───────────────────────────────────────
-KEYPOINT_NAMES = [
-    "nose", "left_eye", "right_eye", "left_ear", "right_ear",
-    "left_shoulder", "right_shoulder", "left_elbow", "right_elbow",
-    "left_wrist", "right_wrist", "left_hip", "right_hip",
-    "left_knee", "right_knee", "left_ankle", "right_ankle"
+# ─── 28 Class Labels لـ Sapiens ──────────────────────────────────
+SEG_CLASSES = [
+    "Background", "Apparel", "Face Neck", "Hair", "Left Foot",
+    "Left Hand", "Left Lower Arm", "Left Lower Leg", "Left Shoe",
+    "Left Sock", "Left Upper Arm", "Left Upper Leg", "Lower Clothing",
+    "Right Foot", "Right Hand", "Right Lower Arm", "Right Lower Leg",
+    "Right Shoe", "Right Sock", "Right Upper Arm", "Right Upper Leg",
+    "Torso", "Upper Clothing", "Lower Lip", "Upper Lip", "Lower Teeth",
+    "Upper Teeth", "Tongue"
 ]
 
+# ─── Classes التي تدل على وجود إنسان ─────────────────────────────
+HUMAN_CLASSES = set(range(1, 28))  # كل شيء ما عدا Background (0)
+
 MODEL_DATA = {}
 
+def preprocess(image: Image.Image) -> torch.Tensor:
+    transform = transforms.Compose([
+        transforms.Resize((1024, 768)),
+        transforms.ToTensor(),
+        transforms.Normalize(
+            mean=[0.485, 0.456, 0.406],
+            std=[0.229, 0.224, 0.225]
+        )
+    ])
+    return transform(image).unsqueeze(0)
+
+def postprocess(output: torch.Tensor, original_size: tuple) -> np.ndarray:
+    # output shape: [1, 28, H, W]
+    seg_map = F.interpolate(
+        output, size=original_size, mode="bilinear", align_corners=False
+    )
+    seg_map = torch.argmax(seg_map, dim=1).squeeze(0).cpu().numpy()
+    return seg_map.astype(np.uint8)
+
 @asynccontextmanager
 async def lifespan(app: FastAPI):
-    print("📥 Loading YOLO11x-pose...")
+    print("📥 Downloading sapiens-seg-1b TorchScript model (~4.72 GB)...")
     start = time.time()
-    # يتحمل تلقائياً من ultralytics
-    MODEL_DATA["model"] = YOLO("yolo11x-pose.pt")
-    print(f"✅ YOLO11x-pose ready in {time.time()-start:.1f}s")
+
+    model_path = hf_hub_download(
+        repo_id="facebook/sapiens-seg-1b-torchscript",
+        filename="sapiens_1b_goliath_best_goliath_mIoU_7994_epoch_151_torchscript.pt2"
+    )
+    print(f"✅ Downloaded in {time.time()-start:.1f}s")
+
+    print("📦 Loading model into memory...")
+    load_start = time.time()
+    model = torch.jit.load(model_path, map_location="cpu")
+    model.eval()
+    MODEL_DATA["model"] = model
+    print(f"✅ Model ready in {time.time()-load_start:.1f}s")
     yield
     MODEL_DATA.clear()
 
 app = FastAPI(
-    title="Human Body Detection - YOLO11x-pose",
-    description="Detects human body and keypoints using YOLO11x-pose",
+    title="Human Body Segmentation - Sapiens-1B",
+    description="Meta Sapiens-seg-1b | 28 class body part segmentation",
     version="1.0.0",
     lifespan=lifespan
 )
 
-# ─── CORS ────────────────────────────────────────────────────────
 app.add_middleware(
     CORSMiddleware,
     allow_origins=["*"],
@@ -60,58 +97,48 @@ async def analyze_image(file: UploadFile = File(...)):
 
     try:
         model = MODEL_DATA["model"]
-        start_time = time.time()
-
-        # ─── تشغيل YOLO ──────────────────────────────────────────
-        results = model(image, verbose=False)
-        elapsed = round(time.time() - start_time, 2)
-
-        persons = []
+        original_size = (image.height, image.width)
 
-        for r in results:
-            num_persons = len(r.boxes) if r.boxes is not None else 0
+        # ─── Preprocessing ────────────────────────────────────────
+        tensor = preprocess(image)
 
-            for i in range(num_persons):
-                person = {"id": i + 1}
-
-                # ─── Bounding Box ─────────────────────────────────
-                if r.boxes is not None and i < len(r.boxes):
-                    box = r.boxes.xyxy[i].tolist()
-                    conf = float(r.boxes.conf[i])
-                    person["bbox"]       = [round(x, 1) for x in box]
-                    person["confidence"] = round(conf, 3)
-
-                # ─── Keypoints ────────────────────────────────────
-                if r.keypoints is not None and i < len(r.keypoints.xy):
-                    kp_xy   = r.keypoints.xy[i].tolist()
-                    kp_conf = r.keypoints.conf[i].tolist() if r.keypoints.conf is not None else [1.0] * 17
-
-                    visible_keypoints = {}
-                    for name, (x, y), c in zip(KEYPOINT_NAMES, kp_xy, kp_conf):
-                        if x > 0 and y > 0 and c > 0.3:
-                            visible_keypoints[name] = {
-                                "x": round(x, 1),
-                                "y": round(y, 1),
-                                "confidence": round(c, 3)
-                            }
-
-                    person["visible_keypoints"]       = visible_keypoints
-                    person["visible_keypoints_count"] = len(visible_keypoints)
-                    person["visible_keypoints_names"] = list(visible_keypoints.keys())
-
-                persons.append(person)
+        # ─── Inference ────────────────────────────────────────────
+        start_time = time.time()
+        with torch.inference_mode():
+            output = model(tensor)
+        elapsed = round(time.time() - start_time, 2)
 
-        # ─── القرار النهائي ───────────────────────────────────────
-        human_detected = len(persons) > 0
-        summary = f"yes detected human: {len(persons)} person(s)" if human_detected else "no detected human body"
+        # ─── Postprocessing ───────────────────────────────────────
+        seg_map = postprocess(output, original_size)
+
+        # ─── استخراج الـ classes الموجودة ─────────────────────────
+        unique_classes = set(np.unique(seg_map).tolist())
+        detected_parts = [
+            SEG_CLASSES[c] for c in sorted(unique_classes)
+            if c in HUMAN_CLASSES
+        ]
+
+        # ─── حساب نسبة كل جزء من الصورة ──────────────────────────
+        total_pixels = seg_map.size
+        parts_coverage = {}
+        for c in sorted(unique_classes):
+            if c in HUMAN_CLASSES:
+                count = int(np.sum(seg_map == c))
+                percentage = round((count / total_pixels) * 100, 2)
+                if percentage > 0.1:  # تجاهل أقل من 0.1%
+                    parts_coverage[SEG_CLASSES[c]] = f"{percentage}%"
+
+        human_detected = len(detected_parts) > 0
+        summary = f"yes detected human body parts: {', '.join(detected_parts)}" \
+                  if human_detected else "no detected human body"
 
         return {
-            "summary":        summary,
-            "detected":       human_detected,
-            "persons_count":  len(persons),
-            "persons":        persons,
-            "execution_time": elapsed,
-            "status":         "success"
+            "summary":         summary,
+            "detected":        human_detected,
+            "parts_found":     detected_parts,
+            "parts_coverage":  parts_coverage,
+            "execution_time":  elapsed,
+            "status":          "success"
         }
 
     except Exception as e: