File size: 4,773 Bytes
9d5eb31
 
 
 
 
 
 
 
 
 
 
 
 
cec2dd5
 
9d5eb31
 
 
 
 
 
 
 
 
 
 
 
 
cec2dd5
9d5eb31
 
 
 
cec2dd5
 
 
 
9d5eb31
 
 
 
cec2dd5
 
 
 
 
 
9d5eb31
 
 
 
 
 
cec2dd5
9d5eb31
 
cec2dd5
 
9d5eb31
 
cec2dd5
 
9d5eb31
cec2dd5
 
 
9d5eb31
cec2dd5
9d5eb31
 
cec2dd5
9d5eb31
 
 
 
 
 
 
cec2dd5
9d5eb31
cec2dd5
9d5eb31
 
 
cec2dd5
 
 
 
 
 
 
 
9d5eb31
 
 
 
cec2dd5
9d5eb31
 
 
 
 
 
 
 
 
 
 
cec2dd5
 
 
 
 
 
 
 
 
 
9d5eb31
 
 
cec2dd5
 
 
 
9d5eb31
 
 
 
 
 
cec2dd5
 
 
 
 
9d5eb31
 
cec2dd5
 
 
 
 
 
 
 
 
 
9d5eb31
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
import io
import os
import faiss
import numpy as np
import pandas as pd
from PIL import Image
from fastapi import FastAPI, File, UploadFile, Query
from fastapi.responses import JSONResponse, HTMLResponse
from sentence_transformers import SentenceTransformer
from transformers import BlipProcessor, BlipForConditionalGeneration
import torch

APP_TITLE = "Image → Hadith Similarity (FAISS)"
INDEX_PATH = "hadith_semantic.faiss"
META_PATH  = "hadith_meta.parquet"

SBERT_NAME = "sentence-transformers/paraphrase-multilingual-MiniLM-L12-v2"
BLIP_NAME  = "Salesforce/blip-image-captioning-base"

app = FastAPI(title=APP_TITLE)

index = None
meta = None
sbert = None
blip_processor = None
blip_model = None
device = "cuda" if torch.cuda.is_available() else "cpu"


@app.on_event("startup")
def load_all():
    global index, meta, sbert, blip_processor, blip_model

    if not os.path.exists(INDEX_PATH):
        raise RuntimeError(f"Missing FAISS index: {INDEX_PATH}")
    if not os.path.exists(META_PATH):
        raise RuntimeError(f"Missing meta file: {META_PATH}")

    index = faiss.read_index(INDEX_PATH)
    meta = pd.read_parquet(META_PATH)

    # Basic sanity check
    if len(meta) != index.ntotal:
        # Not always fatal, but usually means mismatch between index build order and meta rows.
        print(f"[WARN] meta rows ({len(meta)}) != index.ntotal ({index.ntotal}). "
              f"Results will use row positions; ensure they align.")

    sbert = SentenceTransformer(SBERT_NAME)

    blip_processor = BlipProcessor.from_pretrained(BLIP_NAME)
    blip_model = BlipForConditionalGeneration.from_pretrained(BLIP_NAME).to(device)
    blip_model.eval()


@app.get("/health")
def health():
    # Try infer dim from index when possible
    dim = getattr(index, "d", None)
    return {
        "ok": True,
        "index_file": INDEX_PATH,
        "meta_file": META_PATH,
        "index_ntotal": int(index.ntotal),
        "meta_rows": int(len(meta)),
        "dim": int(dim) if dim is not None else None,
        "text_model": SBERT_NAME,
        "caption_model": BLIP_NAME,
        "device": device
    }


def caption_image(pil_img: Image.Image) -> str:
    inputs = blip_processor(images=pil_img, return_tensors="pt").to(device)
    with torch.no_grad():
        out = blip_model.generate(**inputs, max_new_tokens=30)
    cap = blip_processor.decode(out[0], skip_special_tokens=True)
    return cap.strip()


def embed_text(text: str) -> np.ndarray:
    # normalize_embeddings => cosine via inner-product
    v = sbert.encode([text], normalize_embeddings=True)
    return v.astype("float32")


def pick_col(row, candidates, default=""):
    for c in candidates:
        if c in row and pd.notna(row[c]):
            return row[c]
    return default


@app.post("/search_image")
async def search_image(
    file: UploadFile = File(...),
    k: int = Query(10, ge=1, le=50),
    format: str = Query("json"),
):
    data = await file.read()
    pil = Image.open(io.BytesIO(data)).convert("RGB")

    cap = caption_image(pil)
    qvec = embed_text(cap)

    scores, idxs = index.search(qvec, k)

    results = []
    for rank, (i, s) in enumerate(zip(idxs[0].tolist(), scores[0].tolist()), start=1):
        if i < 0 or i >= len(meta):
            continue

        row = meta.iloc[i].to_dict()

        hadith_id = pick_col(row, ["hadithID", "hadith_id", "id", "doc_id"], default=i)
        text_ar   = pick_col(row, ["text_ar", "arabic", "ar", "text"], default="")
        text_en   = pick_col(row, ["text_en", "english", "en"], default="")
        source    = pick_col(row, ["source", "book", "collection"], default="")

        results.append({
            "rank": rank,
            "score": float(s),
            "hadithID": int(hadith_id) if str(hadith_id).isdigit() else str(hadith_id),
            "text_ar": str(text_ar),
            "text_en": str(text_en),
            "source": str(source),
        })

    payload = {"caption": cap, "k": k, "results": results}

    if format == "html":
        items = "\n".join([
            f"<li><b>#{r['rank']}</b> score={r['score']:.3f} — hadithID={r['hadithID']}<br>"
            f"<div style='font-family: system-ui; direction: rtl; font-size: 18px'>{r['text_ar']}</div>"
            f"<div style='color:#666; margin-top:6px'>{r['text_en']}</div>"
            f"<div style='color:#999; margin-top:6px'>source: {r['source']}</div>"
            f"</li>"
            for r in results
        ])
        html = f"""
        <html>
        <body style="margin:18px; font-family: system-ui">
          <h3>Caption</h3>
          <p>{cap}</p>
          <h3>Top Results</h3>
          <ol>{items}</ol>
        </body>
        </html>
        """
        return HTMLResponse(html)

    return JSONResponse(payload)