server.py

"""
Hugging Face Spaces deployment for SAM2 Auto Annotation API.
This file serves as the entry point for the FastAPI application on Hugging Face Spaces.
"""
import sys
import os

# Add sam2 folder to path to import from local sam2 directory
_current_dir = os.path.dirname(os.path.abspath(__file__))
_sam2_dir = os.path.join(_current_dir, "sam2")
# Add sam2 directory to sys.path if not already there
abs_sam2_dir = os.path.abspath(_sam2_dir)
if abs_sam2_dir not in sys.path:
    sys.path.insert(0, abs_sam2_dir)

from fastapi import FastAPI, HTTPException
from fastapi.middleware.cors import CORSMiddleware
import cv2
import numpy as np
import torch
import psutil
import PIL.Image
from requests.exceptions import Timeout, RequestException

# Import sam2 from local folder
from sam2.automatic_mask_generator import SAM2AutomaticMaskGenerator
from model.sam_model import predict_polygon, predict_polygon_from_point
from model.utils import load_image_from_url, mask_to_polygon
from model.sam2_detection_function import SAM2AutoAnnotation, create_sam2_auto_annotation

# Hugging Face model ID for SAM2.1 Hiera Large model
HUGGINGFACE_MODEL_ID = "facebook/sam2.1-hiera-large"
device = "cuda" if torch.cuda.is_available() else "cpu"

# Global SAM2 auto annotation (initialized once)
sam2_auto_annotation_global = None

app = FastAPI(
    title="SAM Auto Annotation API (BBox ➜ Polygon)",
    description="AI-powered auto-annotation API using Meta's Segment Anything Model (SAM)",
    version="1.0.0"
)

# Add CORS middleware to handle preflight OPTIONS requests
app.add_middleware(
    CORSMiddleware,
    allow_origins=["*"],  # Allows all origins
    allow_credentials=True,
    allow_methods=["*"],  # Allows all methods including OPTIONS
    allow_headers=["*"],  # Allows all headers
)


@app.get("/")
def root():
    """Root endpoint - API information."""
    return {
        "status": "Service is up and running!",
        "message": "Backend service is active",
        "api": "SAM Auto Annotation API",
        "version": "1.0.0"
    }


@app.get("/health")
def health_check():
    """Health check endpoint."""
    return {"status": "healthy", "service": "same model segmenticAPI"}


@app.post("/segment")
def segment(data: dict):
    """
    Segment image using SAM2 model to convert bounding box to polygon (CVAT-style).
    Bbox is used as a prompt to identify the object, not as a constraint.
    
    **Input:**
    ```json
    {
      "imageUrl": "https://example.com/image.jpg",
      "bbox": {"x": 494.97, "y": 187.22, "width": 137.99, "height": 98.00, "label": "Object"},
      "imageSize": {"width": 663.07, "height": 442}
    }
    ```
    
    OR
    
    ```json
    {
      "imageUrl": "https://example.com/image.jpg",
      "bbox": [494.97, 187.22, 137.99, 98.00],  // [x, y, width, height]
      "imageSize": [663.07, 442]  // [width, height]
    }
    ```
    
    **Output:**
    ```json
    {
      "polygon": [x1, y1, x2, y2, x3, y3, ...],  // CVAT format: flattened coordinates
      "confidence": 0.96
    }
    ```
    """
    try:
        # Validate input
        if "imageUrl" not in data:
            raise HTTPException(status_code=400, detail="Missing required field: imageUrl")
        if "bbox" not in data:
            raise HTTPException(status_code=400, detail="Missing required field: bbox")
        
        image_url = data["imageUrl"]
        bbox = data["bbox"]
        image_size = data.get("imageSize")  # Optional: for coordinate scaling
        
        # Validate bbox format
        if isinstance(bbox, dict):
            required_keys = ["x", "y", "width", "height"]
            if not all(key in bbox for key in required_keys):
                raise HTTPException(
                    status_code=400,
                    detail=f"bbox dict must contain: {required_keys}"
                )
        elif isinstance(bbox, list):
            if len(bbox) != 4:
                raise HTTPException(
                    status_code=400,
                    detail="bbox list must contain exactly 4 values: [x, y, width, height]"
                )
        else:
            raise HTTPException(
                status_code=400,
                detail="bbox must be either a dict or a list"
            )
        
        # Validate imageSize format if provided
        if image_size is not None:
            if isinstance(image_size, dict):
                if not ("width" in image_size and "height" in image_size):
                    raise HTTPException(
                        status_code=400,
                        detail="imageSize dict must contain 'width' and 'height'"
                    )
            elif isinstance(image_size, list):
                if len(image_size) != 2:
                    raise HTTPException(
                        status_code=400,
                        detail="imageSize list must contain exactly 2 values: [width, height]"
                    )
            else:
                raise HTTPException(
                    status_code=400,
                    detail="imageSize must be either a dict or a list"
                )
        
        # Load image from URL
        img_bgr = load_image_from_url(image_url)
        img_rgb = cv2.cvtColor(img_bgr, cv2.COLOR_BGR2RGB)

        # Predict polygon using SAM2 (bbox as prompt, CVAT-style)
        mask, confidence, scale_factors = predict_polygon(img_rgb, bbox, image_size)
        
        # Convert mask to polygon (CVAT-style)
        polygon = mask_to_polygon(mask, scale_factors)
        
        if not polygon:
            raise HTTPException(status_code=400, detail="No polygon found in mask")

        return {
            "polygon": polygon,  # CVAT format: flattened coordinates
            "confidence": confidence
        }
    except KeyError as e:
        raise HTTPException(status_code=400, detail=f"Missing required field: {str(e)}")
    except ValueError as e:
        raise HTTPException(status_code=400, detail=str(e))
    except FileNotFoundError as e:
        raise HTTPException(status_code=500, detail=str(e))
    except ImportError as e:
        raise HTTPException(
            status_code=500,
            detail=f"Segment Anything library not installed. Please run: pip install -e . in segment-anything directory"
        )
    except Timeout as e:
        raise HTTPException(
            status_code=504,
            detail=f"Image download timeout: {str(e)}. The image server may be slow or unreachable. Please try again or use a different image URL."
        )
    except RequestException as e:
        raise HTTPException(
            status_code=502,
            detail=f"Failed to fetch image from URL: {str(e)}. Please check the image URL and try again."
        )
    except HTTPException:
        raise
    except Exception as e:
        raise HTTPException(status_code=500, detail=f"Internal server error: {str(e)}")


@app.post("/segment/point")
def segment_from_point(data: dict):
    """
    Segment image using SAM2 model with a point click to select object.
    The point identifies which object to segment.
    
    **Input:**
    ```json
    {
      "imageUrl": "https://example.com/image.jpg",
      "point": {"x": 494.97, "y": 187.22},
      "imageSize": {"width": 663.07, "height": 442}
    }
    ```
    
    OR
    
    ```json
    {
      "imageUrl": "https://example.com/image.jpg",
      "point": [494.97, 187.22],  // [x, y]
      "imageSize": [663.07, 442]  // [width, height]
    }
    ```
    
    **Output:**
    ```json
    {
      "polygon": [x1, y1, x2, y2, x3, y3, ...],  // CVAT format: flattened coordinates
      "confidence": 0.96
    }
    ```
    """
    try:
        # Validate input
        if "imageUrl" not in data:
            raise HTTPException(status_code=400, detail="Missing required field: imageUrl")
        if "point" not in data:
            raise HTTPException(status_code=400, detail="Missing required field: point")
        
        image_url = data["imageUrl"]
        point = data["point"]
        image_size = data.get("imageSize")  # Optional: for coordinate scaling
        
        # Validate point format
        if isinstance(point, dict):
            required_keys = ["x", "y"]
            if not all(key in point for key in required_keys):
                raise HTTPException(
                    status_code=400,
                    detail=f"point dict must contain: {required_keys}"
                )
        elif isinstance(point, list):
            if len(point) != 2:
                raise HTTPException(
                    status_code=400,
                    detail="point list must contain exactly 2 values: [x, y]"
                )
        else:
            raise HTTPException(
                status_code=400,
                detail="point must be either a dict or a list"
            )
        
        # Validate imageSize format if provided
        if image_size is not None:
            if isinstance(image_size, dict):
                if not ("width" in image_size and "height" in image_size):
                    raise HTTPException(
                        status_code=400,
                        detail="imageSize dict must contain 'width' and 'height'"
                    )
            elif isinstance(image_size, list):
                if len(image_size) != 2:
                    raise HTTPException(
                        status_code=400,
                        detail="imageSize list must contain exactly 2 values: [width, height]"
                    )
            else:
                raise HTTPException(
                    status_code=400,
                    detail="imageSize must be either a dict or a list"
                )
        
        # Load image from URL
        img_bgr = load_image_from_url(image_url)
        img_rgb = cv2.cvtColor(img_bgr, cv2.COLOR_BGR2RGB)

        # Predict polygon using SAM2 (point click as prompt)
        mask, confidence, scale_factors = predict_polygon_from_point(img_rgb, point, image_size)
        
        # Convert mask to polygon (CVAT-style)
        polygon = mask_to_polygon(mask, scale_factors)
        
        if not polygon:
            raise HTTPException(status_code=400, detail="No polygon found in mask. Try clicking on a different point.")

        return {
            "polygon": polygon,  # CVAT format: flattened coordinates
            "confidence": confidence
        }
    except KeyError as e:
        raise HTTPException(status_code=400, detail=f"Missing required field: {str(e)}")
    except ValueError as e:
        raise HTTPException(status_code=400, detail=str(e))
    except FileNotFoundError as e:
        raise HTTPException(status_code=500, detail=str(e))
    except ImportError as e:
        raise HTTPException(
            status_code=500,
            detail=f"Segment Anything library not installed. Please run: pip install -e . in segment-anything directory"
        )
    except Timeout as e:
        raise HTTPException(
            status_code=504,
            detail=f"Image download timeout: {str(e)}. The image server may be slow or unreachable. Please try again or use a different image URL."
        )
    except RequestException as e:
        raise HTTPException(
            status_code=502,
            detail=f"Failed to fetch image from URL: {str(e)}. Please check the image URL and try again."
        )
    except HTTPException:
        raise
    except Exception as e:
        raise HTTPException(status_code=500, detail=f"Internal server error: {str(e)}")


@app.post("/auto-annotate")
def auto_annotate(data: dict):
    """
    Automatically detect and segment all objects in an image using SAM2 from Hugging Face.
    Uses SAM2AutomaticMaskGenerator (facebook/sam2.1-hiera-large) to detect all objects without requiring prompts (bbox or points).
    
    **Input:**
    ```json
    {
      "imageUrl": "https://example.com/image.jpg",
      "imageSize": {"width": 663.07, "height": 442},
      "minArea": 100,
      "minConfidence": 0.5,
      "maxImageDimension": 1024,
      "pointsPerSide": 32,
      "pointsPerBatch": 64,
      "filterObjectsOnly": true
    }
    ```
    
    **Output:**
    ```json
    {
      "masks": [
        {
          "polygon": [x1, y1, x2, y2, x3, y3, ...],
          "confidence": 0.93,
          "area": 12345
        },
        ...
      ],
      "count": 10,
      "memoryInfo": {
        "before_mb": 512.5,
        "after_mb": 1024.3,
        "peak_mb": 1024.3,
        "estimated_mb": 800.0,
        "memory_used_mb": 511.8
      },
      "imageInfo": {
        "wasResized": true,
        "originalSize": [1920, 1080],
        "processedSize": [1024, 576],
        "resizeScale": [1.875, 1.875]
      }
    }
    ```
    """
    try:
        # Validate input
        if "imageUrl" not in data:
            raise HTTPException(status_code=400, detail="Missing required field: imageUrl")
        
        image_url = data["imageUrl"]
        image_size = data.get("imageSize")  # Optional: for coordinate scaling
        min_area = data.get("minArea", 100)  # Optional: minimum mask area
        min_confidence = data.get("minConfidence", 0.5)  # Optional: minimum confidence
        max_image_dimension = data.get("maxImageDimension", 1024)  # Optional: max dimension before resizing
        # Lower default values for faster processing
        points_per_side = data.get("pointsPerSide", 32)  # Optional: points per side (lower = faster)
        points_per_batch = data.get("pointsPerBatch", 64)  # Optional: points per batch (lower = faster)
        filter_objects_only = data.get("filterObjectsOnly", False)  # Optional: filter out background masks
        
        # Validate imageSize format if provided
        if image_size is not None:
            if isinstance(image_size, dict):
                if not ("width" in image_size and "height" in image_size):
                    raise HTTPException(
                        status_code=400,
                        detail="imageSize dict must contain 'width' and 'height'"
                    )
            elif isinstance(image_size, list):
                if len(image_size) != 2:
                    raise HTTPException(
                        status_code=400,
                        detail="imageSize list must contain exactly 2 values: [width, height]"
                    )
            else:
                raise HTTPException(
                    status_code=400,
                    detail="imageSize must be either a dict or a list"
                )
        
        # Validate minArea and minConfidence
        try:
            min_area = int(min_area)
            if min_area < 0:
                raise HTTPException(status_code=400, detail="minArea must be >= 0")
        except (ValueError, TypeError):
            raise HTTPException(status_code=400, detail="minArea must be an integer")
        
        try:
            min_confidence = float(min_confidence)
            if not (0.0 <= min_confidence <= 1.0):
                raise HTTPException(status_code=400, detail="minConfidence must be between 0.0 and 1.0")
        except (ValueError, TypeError):
            raise HTTPException(status_code=400, detail="minConfidence must be a float between 0.0 and 1.0")
        
        # Validate maxImageDimension
        try:
            max_image_dimension = int(max_image_dimension)
            if max_image_dimension < 256:
                raise HTTPException(status_code=400, detail="maxImageDimension must be >= 256")
            if max_image_dimension > 4096:
                raise HTTPException(status_code=400, detail="maxImageDimension must be <= 4096")
        except (ValueError, TypeError):
            raise HTTPException(status_code=400, detail="maxImageDimension must be an integer between 256 and 4096")
        
        # Validate pointsPerSide
        try:
            points_per_side = int(points_per_side)
            if points_per_side < 8:
                raise HTTPException(status_code=400, detail="pointsPerSide must be >= 8")
            if points_per_side > 128:
                raise HTTPException(status_code=400, detail="pointsPerSide must be <= 128")
        except (ValueError, TypeError):
            raise HTTPException(status_code=400, detail="pointsPerSide must be an integer between 8 and 128")
        
        # Validate pointsPerBatch
        try:
            points_per_batch = int(points_per_batch)
            if points_per_batch < 16:
                raise HTTPException(status_code=400, detail="pointsPerBatch must be >= 16")
            if points_per_batch > 256:
                raise HTTPException(status_code=400, detail="pointsPerBatch must be <= 256")
        except (ValueError, TypeError):
            raise HTTPException(status_code=400, detail="pointsPerBatch must be an integer between 16 and 256")
        
        # Get memory before processing
        process = psutil.Process(os.getpid())
        memory_before = process.memory_info().rss / (1024 * 1024)  # MB
        
        # Load image from URL
        img_bgr = load_image_from_url(image_url)
        img_rgb = cv2.cvtColor(img_bgr, cv2.COLOR_BGR2RGB)
        
        # Resize image if needed to reduce memory usage
        original_h, original_w = img_rgb.shape[:2]
        original_size = [original_w, original_h]
        
        processed_image = img_rgb
        resize_scale = [1.0, 1.0]
        was_resized = False
        
        if max(original_h, original_w) > max_image_dimension:
            was_resized = True
            if original_h > original_w:
                new_h = max_image_dimension
                new_w = int(original_w * (max_image_dimension / original_h))
            else:
                new_w = max_image_dimension
                new_h = int(original_h * (max_image_dimension / original_w))
            processed_image = cv2.resize(img_rgb, (new_w, new_h), interpolation=cv2.INTER_LINEAR)
            resize_scale = [original_w / new_w, original_h / new_h]
        
        processed_h, processed_w = processed_image.shape[:2]
        processed_size = [processed_w, processed_h]
        
        # Estimate memory requirements
        estimated_mb = ((processed_w * processed_h * 3 * 4) + (processed_w * processed_h * 256 * 4) + (processed_w * processed_h * 100 * 1)) / (1024 * 1024)
        
        # Calculate scale factors for coordinate scaling (matching predict_polygon_from_point logic)
        # We need to scale FROM processed image TO display size (imageSize)
        # mask_to_polygon expects scale_factors that represent: FROM processed TO display
        # It divides by these factors, so we pass (processed_w/display_w, processed_h/display_h)
        scale_factor_x, scale_factor_y = 1.0, 1.0
        
        if image_size is not None:
            if isinstance(image_size, dict):
                display_w = float(image_size.get("width", processed_w))
                display_h = float(image_size.get("height", processed_h))
            else:
                display_w, display_h = float(image_size[0]), float(image_size[1])
            
            # Calculate scale factors: FROM processed image TO display size
            # These will be used in mask_to_polygon: polygon / scale_factor = display coords
            scale_factor_x = processed_w / display_w if display_w > 0 else 1.0
            scale_factor_y = processed_h / display_h if display_h > 0 else 1.0
        
        # Get image dimensions for filtering
        total_image_area = processed_w * processed_h
        
        # Initialize SAM2 Auto Annotation
        # This uses facebook/sam2.1-hiera-large model from Hugging Face
        # Cache the annotation instance globally to avoid reloading on every request
        global sam2_auto_annotation_global
        
        if sam2_auto_annotation_global is None:
            try:
                sam2_auto_annotation_global = create_sam2_auto_annotation(
                    points_per_side=points_per_side,
                    points_per_batch=points_per_batch,
                    pred_iou_thresh=0.88,
                    stability_score_thresh=0.95,
                    min_mask_region_area=min_area,
                )
            except ImportError as e:
                raise HTTPException(
                    status_code=500,
                    detail=f"Failed to import required modules. Please ensure 'sam2' and 'huggingface_hub' are installed. Error: {str(e)}"
                )
            except Exception as e:
                raise HTTPException(
                    status_code=500,
                    detail=f"Failed to load SAM2 Auto Annotation from Hugging Face ({HUGGINGFACE_MODEL_ID}). Error: {str(e)}"
                )
        
        # Generate masks using SAM2AutoAnnotation with proper scaling (matching predict_polygon_from_point)
        # Pass scale_factors to scale FROM processed image TO display size
        mask_results = sam2_auto_annotation_global.generate_masks(
            image=processed_image,
            min_confidence=min_confidence,
            min_area=min_area,
            filter_blank_regions=True,
            scale_factors=(scale_factor_x, scale_factor_y)
        )
        
        # Get memory after processing
        memory_after = process.memory_info().rss / (1024 * 1024)  # MB
        memory_used = memory_after - memory_before
        
        # Process mask results (polygons are already scaled to display size by generate_masks)
        results = []
        
        for mask_result in mask_results:
            # Extract mask information
            polygon = mask_result.get("polygon")
            score = mask_result.get("confidence")
            area = mask_result.get("area")
            
            # Early filtering: Skip masks that don't meet basic criteria
            if area < min_area or score < min_confidence:
                continue
            
            # Filter out background masks if filterObjectsOnly is True
            if filter_objects_only:
                coverage_ratio = area / total_image_area if total_image_area > 0 else 0
                if coverage_ratio >= 0.8:  # Skip masks covering >80% (likely background)
                    continue
            
            # Polygon is already scaled to display size by generate_masks (using mask_to_polygon with scale_factors)
            # Return polygon in flattened format [x1, y1, x2, y2, ...]
            if polygon and len(polygon) >= 6:  # At least 3 points
                mask_obj = {
                    "polygon": polygon  # Already in flattened format and scaled to display size
                }
                if score is not None:
                    mask_obj["confidence"] = score
                if area is not None:
                    mask_obj["area"] = area
                results.append(mask_obj)
        
        # Build response with all required fields
        response = {
            "masks": results,
            "count": len(results),
            "memoryInfo": {
                "before_mb": round(memory_before, 2),
                "after_mb": round(memory_after, 2),
                "peak_mb": round(memory_after, 2),
                "estimated_mb": round(estimated_mb, 2),
                "memory_used_mb": round(memory_used, 2)
            },
            "imageInfo": {
                "wasResized": was_resized,
                "originalSize": original_size,
                "processedSize": processed_size,
                "resizeScale": resize_scale
            }
        }
        
        return response
        
    except KeyError as e:
        raise HTTPException(status_code=400, detail=f"Missing required field: {str(e)}")
    except ValueError as e:
        raise HTTPException(status_code=400, detail=str(e))
    except FileNotFoundError as e:
        raise HTTPException(status_code=500, detail=str(e))
    except ImportError as e:
        raise HTTPException(
            status_code=500,
            detail=f"Segment Anything library not installed. Please ensure 'sam2' and 'huggingface_hub' are installed."
        )
    except Timeout as e:
        raise HTTPException(
            status_code=504,
            detail=f"Image download timeout: {str(e)}. The image server may be slow or unreachable. Please try again or use a different image URL."
        )
    except RequestException as e:
        raise HTTPException(
            status_code=502,
            detail=f"Failed to fetch image from URL: {str(e)}. Please check the image URL and try again."
        )
    except HTTPException:
        raise
    except Exception as e:
        raise HTTPException(status_code=500, detail=f"Internal server error: {str(e)}")