ocr-poetry-pkg 0.1.0__tar.gz

ocr_poetry_pkg-0.1.0/PKG-INFO

@@ -0,0 +1,16 @@
+Metadata-Version: 2.1
+Name: ocr-poetry-pkg
+Version: 0.1.0
+Summary:  OCR CLI for Arabic images using unsloth/surya
+Author: alka gupta
+Author-email: alka.gupta@ksolves.com
+Requires-Python: >=3.12,<4.0
+Classifier: Programming Language :: Python :: 3
+Classifier: Programming Language :: Python :: 3.12
+Requires-Dist: click (>=8.3.0,<9.0.0)
+Requires-Dist: pillow (>=12.0.0,<13.0.0)
+Description-Content-Type: text/markdown
+
+OCR CLI for Arabic images using unsloth/surya
+
+

ocr_poetry_pkg-0.1.0/README.md

@@ -0,0 +1,2 @@
+OCR CLI for Arabic images using unsloth/surya
+

ocr_poetry_pkg-0.1.0/pyproject.toml

@@ -0,0 +1,19 @@
+[tool.poetry]
+name = "ocr-poetry-pkg"
+version = "0.1.0"
+description = " OCR CLI for Arabic images using unsloth/surya"
+authors = ["alka gupta <alka.gupta@ksolves.com>"]
+readme = "README.md"
+packages = [{ include = "ocr_poetry_pkg", from = "src" }]
+
+[tool.poetry.dependencies]
+python = "^3.12"
+pillow = "^12.0.0"
+click = "^8.3.0"
+
+[tool.poetry.scripts]
+ocr-batch = "ocr_poetry_pkg.cli:main"
+
+[build-system]
+requires = ["poetry-core>=1.0.0"]
+build-backend = "poetry.core.masonry.api"

ocr_poetry_pkg-0.1.0/src/ocr_poetry_pkg/__init__.py

@@ -0,0 +1 @@
+__version__ = "0.1.0"

ocr_poetry_pkg-0.1.0/src/ocr_poetry_pkg/cli.py

@@ -0,0 +1,110 @@
+import json
+import math
+import traceback
+from pathlib import Path
+from concurrent.futures import ThreadPoolExecutor, as_completed
+from PIL import Image
+import torch
+from unsloth import FastVisionModel
+
+# ---------------- Model Loading ----------------
+model, tokenizer = FastVisionModel.from_pretrained(
+    "AhmedZaky1/DIMI-Arabic-OCR-v2",
+    load_in_4bit=True,
+    device_map={"": 0},
+)
+FastVisionModel.for_inference(model)
+
+device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
+model.to(device)
+
+# ---------------- Config ----------------.
+IMAGES_DIR = Path("/home/alkaks1309/Desktop/ocr_poetry_pkg/images")  # folder containing page images
+JSON_BASE = Path("/home/alkaks1309/Desktop/ocr_poetry_pkg/detections/results")  # base folder for detection jsons
+OUT_DIR = Path("/home/alkaks1309/Desktop/ocr_poetry_pkg/output")
+OUT_DIR.mkdir(parents=True, exist_ok=True)
+
+        # shrink from 2048 for speed
+
+
+# ---------------- Helpers ----------------
+def clamp(v, lo, hi): return max(lo, min(v, hi))
+
+def axis_aligned_from_polygon(poly, img_w, img_h):
+    xs = [p[0] for p in poly]; ys = [p[1] for p in poly]
+    x0, y0 = clamp(int(min(xs)), 0, img_w - 1), clamp(int(min(ys)), 0, img_h - 1)
+    x1, y1 = clamp(int(max(xs)), 0, img_w), clamp(int(max(ys)), 0, img_h)
+    return [x0, y0, x1, y1]
+
+INSTRUCTION = (
+    "Extract only the Arabic text visible in this image. Ignore Urdu or Persian. "
+    "Preserve diacritics and punctuation exactly. Do not translate. Output Arabic only."
+)
+
+# ---------------- Core Function ----------------
+def process_image(image_path: Path):
+    try:
+        img = Image.open(image_path).convert("RGB")
+        img_w, img_h = img.size
+
+        # Construct the detection JSON path like:
+        # /kaggle/input/json-folder/results/surya/{stem}/results.json
+        json_path = JSON_BASE / image_path.stem / "results.json"
+        if not json_path.exists():
+            return {"image": str(image_path), "error": f"missing detection JSON: {json_path}"}
+
+        result_data = json.loads(json_path.read_text(encoding="utf-8"))
+        if isinstance(result_data, dict):
+            page_entries = next(iter(result_data.values()))
+        elif isinstance(result_data, list):
+            page_entries = result_data
+        else:
+            return {"image": str(image_path), "error": "invalid JSON structure"}
+
+        metadata = []
+        for p in page_entries:
+            for bb in p.get("bboxes", []):
+                poly = bb.get("polygon")
+                if not poly:
+                    continue
+                x0, y0, x1, y1 = axis_aligned_from_polygon(poly[:4], img_w, img_h)
+                crop = img.crop((x0, y0, x1, y1))
+                
+                messages = [
+                    {"role": "user", "content": [
+                        {"type": "image", "image": crop},
+                        {"type": "text", "text": INSTRUCTION},
+                    ]}
+                ]
+                text = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
+                inputs = tokenizer(text=[text], images=[crop], return_tensors="pt", padding=True)
+                inputs = {k: v.to(device) if hasattr(v, "to") else v for k, v in inputs.items()}
+
+                with torch.inference_mode():
+                    outputs = model.generate(**inputs, max_new_tokens=2048)
+                input_len = inputs["input_ids"].shape[1]
+                preds = outputs[:, input_len:]
+                decoded = tokenizer.batch_decode(preds, skip_special_tokens=True)
+                prediction = decoded[0] if decoded else ""
+
+                metadata.append({
+                    "bbox": bb,
+                    "prediction": prediction,
+                })
+
+        out_file = OUT_DIR / f"{image_path.stem}.json"
+        out_file.write_text(json.dumps({"image": str(image_path), "results": metadata}, ensure_ascii=False, indent=2))
+        return {"image": str(image_path), "status": "done"}
+
+    except Exception as e:
+        return {"image": str(image_path), "error": str(e), "traceback": traceback.format_exc()}
+
+# ---------------- Parallel Runner ----------------
+def main():
+    images = sorted([p for p in IMAGES_DIR.iterdir() if p.suffix.lower() in {".jpg", ".jpeg", ".png"}])
+    for img in images:
+        result = process_image(img)
+        print(json.dumps(result, ensure_ascii=False))
+
+if __name__ == "__main__":
+    main()

ocr_poetry_pkg-0.1.0/src/ocr_poetry_pkg/layout.py

@@ -0,0 +1,43 @@
+import os
+import subprocess
+from pathlib import Path
+from concurrent.futures import ThreadPoolExecutor, as_completed
+
+IMAGES_DIR = Path("/home/alkaks1309/Desktop/ocr_poetry_pkg/images")
+VALID_EXTS = {".jpg", ".jpeg", ".png", ".tif", ".tiff"}
+
+# tune this: number of parallel processes to run
+MAX_WORKERS = min(8, (os.cpu_count() or 4))  # example: up to 8 or CPU count
+
+def run_layout(image_path: Path):
+    cmd = ["surya_layout", str(image_path)]
+    try:
+        proc = subprocess.run(cmd, capture_output=True, text=True, check=True)
+        return (image_path.name, True, proc.stdout, proc.stderr)
+    except subprocess.CalledProcessError as e:
+        return (image_path.name, False, e.stdout or "", e.stderr or str(e))
+
+def get_images(folder: Path):
+    return [p for p in folder.iterdir() if p.suffix.lower() in VALID_EXTS]
+
+def main():
+    images = get_images(IMAGES_DIR)
+    if not images:
+        print("No images found.")
+        return
+
+    with ThreadPoolExecutor(max_workers=MAX_WORKERS) as ex:
+        futures = {ex.submit(run_layout, img): img for img in images}
+        for fut in as_completed(futures):
+            name, ok, out, err = fut.result()
+            if ok:
+                print(f"{name} finished.")
+                if out.strip():
+                    print(f"  stdout: {out.strip()}")
+            else:
+                print(f"{name} failed.")
+                if err:
+                    print(f"  stderr: {err.strip()}")
+
+if __name__ == "__main__":
+    main()

ocr_poetry_pkg-0.1.0/src/ocr_poetry_pkg/pdf2img.py

@@ -0,0 +1,15 @@
+from pdf2image import convert_from_path
+from pathlib import Path
+
+PDF = "/path/to/input.pdf"
+OUT_DIR = Path("out_images")
+OUT_DIR.mkdir(parents=True, exist_ok=True)
+
+# convert all pages
+pages = convert_from_path(PDF, dpi=300)  # increase dpi for better quality
+for i, page in enumerate(pages, start=1):
+    out_path = OUT_DIR / f"page_{i:03d}.jpg"
+    page.save(out_path, "JPEG", quality=95)
+
+# Convert a page range or single page:
+# pages = convert_from_path(PDF, dpi=300, first_page=2, last_page=5)