1

Dockerfile

@@ -0,0 +1,42 @@
+# ────────────────────────────────────────────────────────────────────────────────
+# Grounded‑SAM CPU Docker image with Flask API
+# ────────────────────────────────────────────────────────────────────────────────
+FROM python:3.10-slim
+
+ENV DEBIAN_FRONTEND=noninteractive \
+    PYTHONUNBUFFERED=1 \
+    AM_I_DOCKER=True \
+    BUILD_WITH_CUDA=False \
+    # ↓ Hugging Face cache inside the container (optional)
+    HF_HOME=/opt/hf_cache
+
+# ––– OS packages –––
+RUN apt-get update && \
+    apt-get install -y --no-install-recommends \
+        git wget ffmpeg libgl1 && \
+    apt-get clean && rm -rf /var/lib/apt/lists/*
+
+# ––– Code –––
+WORKDIR /workspace
+COPY requirements.txt ./requirements.txt
+RUN pip install --no-cache-dir -r requirements.txt
+
+# Segment‑Anything & GroundingDINO in editable mode
+RUN git clone --depth 1 https://github.com/facebookresearch/segment-anything.git && \
+    pip install -e segment-anything
+
+RUN git clone --depth 1 https://github.com/IDEA-Research/GroundingDINO.git && \
+    pip install --no-build-isolation -e GroundingDINO
+
+# Flask API
+COPY app.py ./app.py
+
+# Download pretrained checkpoints at build time (comment out to download on first run)
+RUN mkdir -p weights && \
+    wget -q -O weights/sam_vit_h_4b8939.pth \
+      https://dl.fbaipublicfiles.com/segment_anything/sam_vit_h_4b8939.pth && \
+    wget -q -O weights/groundingdino_swint_ogc.pth \
+      https://github.com/IDEA-Research/GroundingDINO/releases/download/v0.1.0-alpha/groundingdino_swint_ogc.pth
+
+EXPOSE 7860
+ENTRYPOINT ["python", "app.py", "--host", "0.0.0.0", "--port", "7860"]

app.py

@@ -0,0 +1,85 @@
+"""
+Grounded‑SAM Flask API (CPU only)
+
+POST /segment
+Body (multipart/form-data):
+  - image: the house photo
+  - prompt: text prompt, e.g. "roof sheet"
+Query params:
+  - overlay (bool, default=false): if true, returns a PNG overlay instead
+Returns:
+  - image/png mask (single channel)  OR  overlay
+"""
+import io
+import os
+import argparse
+import numpy as np
+from PIL import Image
+from flask import Flask, request, send_file
+from flask_cors import CORS
+
+import torch
+from groundingdino.util.inference import Model as GroundingModel
+from segment_anything import sam_model_registry, SamPredictor
+
+# ─── Load models once ───────────────────────────────────────────────────────────
+device = torch.device("cpu")
+
+DINO_CKPT = "weights/groundingdino_swint_ogc.pth"
+SAM_CKPT  = "weights/sam_vit_h_4b8939.pth"
+
+grounder = GroundingModel(DINO_CKPT, device=device)
+sam = sam_model_registry["vit_h"](checkpoint=SAM_CKPT).to(device)
+predictor = SamPredictor(sam)
+
+# ─── Flask app ──────────────────────────────────────────────────────────────────
+app = Flask(__name__)
+CORS(app)
+
+def segment(image_pil: Image.Image, prompt: str):
+    # 1) Run GroundingDINO to get boxes for the prompt
+    boxes, _, _ = grounder.predict(image_pil, prompt=prompt, box_threshold=0.3, text_threshold=0.25)
+    if boxes.size == 0:
+        raise ValueError("No boxes found for prompt.")
+
+    # 2) Largest box → mask via SAM
+    box = boxes[np.argmax((boxes[:,2]-boxes[:,0])*(boxes[:,3]-boxes[:,1]))]
+    predictor.set_image(np.array(image_pil))
+    masks, _, _ = predictor.predict(box=box)
+    mask = masks[0]  # boolean HxW
+
+    return mask
+
+@app.route("/segment", methods=["POST"])
+def segment_endpoint():
+    if "image" not in request.files or "prompt" not in request.form:
+        return {"error": "image file and prompt are required."}, 400
+
+    prompt = request.form["prompt"]
+    image = Image.open(request.files["image"].stream).convert("RGB")
+
+    try:
+        mask = segment(image, prompt)
+    except ValueError as e:
+        return {"error": str(e)}, 422
+
+    overlay = request.args.get("overlay", "false").lower() == "true"
+    if overlay:
+        colored = np.array(image).copy()
+        colored[mask] = [255, 0, 0]  # red overlay
+        out_img = Image.fromarray(colored)
+    else:
+        out_img = Image.fromarray((mask * 255).astype(np.uint8))
+
+    buf = io.BytesIO()
+    out_img.save(buf, format="PNG")
+    buf.seek(0)
+    return send_file(buf, mimetype="image/png")
+
+# ─── CLI ────────────────────────────────────────────────────────────────────────
+if __name__ == "__main__":
+    parser = argparse.ArgumentParser()
+    parser.add_argument("--host", default="127.0.0.1")
+    parser.add_argument("--port", default=7860, type=int)
+    args = parser.parse_args()
+    app.run(host=args.host, port=args.port)

requirements.txt

@@ -0,0 +1,12 @@
+# Core ML
+torch==2.3.0       # CPU build from PyPI
+torchvision==0.18.0
+# Grounded‑SAM additional deps
+diffusers==0.27.0
+opencv-python-headless==4.10.0.82
+Pillow>=10.1
+pycocotools==2.0.7
+matplotlib==3.9.0
+# API
+flask==3.0.3
+flask-cors==4.0.0

spaces.yaml

@@ -0,0 +1 @@
+sdk : "docker"