GOCR — KSVTRv3-de (deutscher Recognizer), ONNX/CPU · Apache-2.0 (built on OpenOCR + PaddleOCR)

.gitattributes

@@ -0,0 +1,36 @@
+*.7z filter=lfs diff=lfs merge=lfs -text
+*.arrow filter=lfs diff=lfs merge=lfs -text
+*.bin filter=lfs diff=lfs merge=lfs -text
+*.bz2 filter=lfs diff=lfs merge=lfs -text
+*.ckpt filter=lfs diff=lfs merge=lfs -text
+*.ftz filter=lfs diff=lfs merge=lfs -text
+*.gz filter=lfs diff=lfs merge=lfs -text
+*.h5 filter=lfs diff=lfs merge=lfs -text
+*.joblib filter=lfs diff=lfs merge=lfs -text
+*.lfs.* filter=lfs diff=lfs merge=lfs -text
+*.mlmodel filter=lfs diff=lfs merge=lfs -text
+*.model filter=lfs diff=lfs merge=lfs -text
+*.msgpack filter=lfs diff=lfs merge=lfs -text
+*.npy filter=lfs diff=lfs merge=lfs -text
+*.npz filter=lfs diff=lfs merge=lfs -text
+*.onnx filter=lfs diff=lfs merge=lfs -text
+*.ot filter=lfs diff=lfs merge=lfs -text
+*.parquet filter=lfs diff=lfs merge=lfs -text
+*.pb filter=lfs diff=lfs merge=lfs -text
+*.pickle filter=lfs diff=lfs merge=lfs -text
+*.pkl filter=lfs diff=lfs merge=lfs -text
+*.pt filter=lfs diff=lfs merge=lfs -text
+*.pth filter=lfs diff=lfs merge=lfs -text
+*.rar filter=lfs diff=lfs merge=lfs -text
+*.safetensors filter=lfs diff=lfs merge=lfs -text
+saved_model/**/* filter=lfs diff=lfs merge=lfs -text
+*.tar.* filter=lfs diff=lfs merge=lfs -text
+*.tar filter=lfs diff=lfs merge=lfs -text
+*.tflite filter=lfs diff=lfs merge=lfs -text
+*.tgz filter=lfs diff=lfs merge=lfs -text
+*.wasm filter=lfs diff=lfs merge=lfs -text
+*.xz filter=lfs diff=lfs merge=lfs -text
+*.zip filter=lfs diff=lfs merge=lfs -text
+*.zst filter=lfs diff=lfs merge=lfs -text
+*tfevents* filter=lfs diff=lfs merge=lfs -text
+examples/rechnung.png filter=lfs diff=lfs merge=lfs -text

README.md

@@ -0,0 +1,19 @@
+---
+title: GOCR Deutsche OCR
+emoji: 🧾
+colorFrom: blue
+colorTo: indigo
+sdk: gradio
+sdk_version: 6.19.0
+app_file: app.py
+pinned: false
+license: apache-2.0
+short_description: State-of-the-Art OCR-Foundation für deutsche Dokumente (CPU)
+---
+
+# GOCR — State-of-the-Art OCR-Foundation für deutsche Dokumente
+
+Ganzes Dokument → **Text + Position (bbox)**. Schnell, **CPU, kein GPU**.
+Roher, präziser Output — direkt weiterverarbeitbar (z. B. in einem LLM).
+
+→ **[german-ocr.de](https://german-ocr.de)**

app.py

@@ -0,0 +1,177 @@
+"""German-OCR · GOCR — Demo (CPU), im german-ocr.de-Look.
+
+Moderner OCR-Viewer: buntes Box-Overlay + Maske + Text-Layout + strukturiertes JSON.
+Nutzt das eigene g_ocr-Paket (from_pretrained). Schnell, CPU, kein GPU.
+"""
+import os
+import json
+import time
+
+import numpy as np
+from PIL import Image, ImageDraw, ImageFont
+import gradio as gr
+import matplotlib.font_manager as fm
+
+import g_ocr
+
+_HERE = os.path.dirname(os.path.abspath(__file__))
+EXAMPLES_DIR = os.path.join(_HERE, "examples")
+_FONT_PATH = fm.findfont("DejaVu Sans")
+WEBSITE = "https://german-ocr.de"
+LOGO = "https://german-ocr.de/icon.png"
+
+ocr = g_ocr.from_pretrained(drop_score=0.4)
+ocr.read(np.zeros((64, 64, 3), np.uint8))  # warmup
+
+# Lebendige, gut unterscheidbare Farben (eine pro Region)
+PALETTE = [
+    (230, 25, 75), (60, 180, 75), (0, 130, 200), (245, 130, 48), (145, 30, 180),
+    (0, 158, 115), (240, 50, 230), (0, 114, 178), (213, 94, 0), (204, 121, 167),
+    (86, 180, 233), (170, 110, 40), (0, 128, 128), (128, 0, 0), (75, 175, 60),
+]
+
+_font_cache = {}
+
+
+def _font(sz):
+    sz = int(max(10, min(42, sz)))
+    if sz not in _font_cache:
+        try:
+            _font_cache[sz] = ImageFont.truetype(_FONT_PATH, sz)
+        except Exception:
+            _font_cache[sz] = ImageFont.load_default()
+    return _font_cache[sz]
+
+
+def _col(i):
+    return PALETTE[i % len(PALETTE)]
+
+
+def _pts(r):
+    return [(int(x), int(y)) for x, y in r["quad"]]
+
+
+def _overlay(rgb, regions):
+    """Original + buntes Box-Overlay (je Region eigene Farbe, halbtransparente Füllung)."""
+    base = Image.fromarray(rgb).convert("RGBA")
+    ov = Image.new("RGBA", base.size, (0, 0, 0, 0))
+    d = ImageDraw.Draw(ov)
+    for i, r in enumerate(regions):
+        c = _col(i)
+        p = _pts(r)
+        d.polygon(p, fill=c + (46,))
+        d.line(p + [p[0]], fill=c + (255,), width=3, joint="curve")
+    return np.array(Image.alpha_composite(base, ov).convert("RGB"))
+
+
+def _mask(shape, regions):
+    """Detektions-Maske: erkannte Regionen weiß auf schwarz."""
+    h, w = shape[:2]
+    im = Image.new("RGB", (w, h), (0, 0, 0))
+    d = ImageDraw.Draw(im)
+    for r in regions:
+        d.polygon(_pts(r), fill=(255, 255, 255))
+    return np.array(im)
+
+
+def _textlayout(shape, regions):
+    """Erkannte Werte farbig an Originalposition (weißes Blatt)."""
+    h, w = shape[:2]
+    im = Image.new("RGB", (w, h), (255, 255, 255))
+    d = ImageDraw.Draw(im)
+    for i, r in enumerate(regions):
+        p = _pts(r)
+        x0 = min(q[0] for q in p)
+        y0 = min(q[1] for q in p)
+        bh = max(q[1] for q in p) - y0
+        d.text((x0 + 1, y0), r["text"][:60], fill=_col(i), font=_font(bh * 0.72))
+    return np.array(im)
+
+
+def _pretty(res):
+    head = {k: res[k] for k in ("engine", "version", "image", "n_regions")}
+    s = json.dumps(head, ensure_ascii=False, indent=2)[:-2]
+    s += ',\n  "text": ' + json.dumps(res["text"], ensure_ascii=False)
+    s += ',\n  "regions": [\n'
+    s += ",\n".join("    " + json.dumps(r, ensure_ascii=False) for r in res["regions"])
+    s += "\n  ]\n}"
+    return s
+
+
+def run(image):
+    if image is None:
+        return None, None, None, "", "{}", "Bitte ein Bild hochladen."
+    t0 = time.time()
+    res = ocr.read(image)
+    ms = (time.time() - t0) * 1000
+    R = res["regions"]
+    boxes = _overlay(image, R)
+    mask = _mask(image.shape, R)
+    tlay = _textlayout(image.shape, R)
+    status = f"✅ {res['n_regions']} Regionen · {ms:.0f} ms (CPU) — Output bereit fürs LLM"
+    return boxes, mask, tlay, res["text"], _pretty(res), status
+
+
+HEADER = f"""
+<div style="text-align:center;padding:24px 16px 4px;">
+  <img src="{LOGO}" alt="German-OCR" style="height:54px;display:block;margin:0 auto;"/>
+  <div style="font-size:1.95em;font-weight:800;color:#0f172a;margin-top:8px;letter-spacing:-.5px;">
+     German-OCR <span style="color:#2563eb;">GOCR</span></div>
+  <div style="font-size:1.12em;color:#334155;margin-top:2px;">Deutsche OCR-Engine — ganzes Dokument → Text + Position (bbox)</div>
+  <div style="color:#64748b;margin-top:4px;font-size:.98em;">Schnell · CPU, kein GPU · Output direkt fürs LLM</div>
+  <div style="margin-top:12px;display:flex;gap:8px;justify-content:center;flex-wrap:wrap;">
+    <span style="background:#eff6ff;color:#1d4ed8;padding:4px 12px;border-radius:999px;font-size:.85em;font-weight:600;">🇩🇪 Deutsch</span>
+    <span style="background:#f1f5f9;color:#334155;padding:4px 12px;border-radius:999px;font-size:.85em;font-weight:600;">⚡ CPU, kein GPU</span>
+    <span style="background:#f1f5f9;color:#334155;padding:4px 12px;border-radius:999px;font-size:.85em;font-weight:600;">📦 ~38 MB</span>
+    <span style="background:#f1f5f9;color:#334155;padding:4px 12px;border-radius:999px;font-size:.85em;font-weight:600;">🎯 bbox + JSON</span>
+  </div>
+</div>
+"""
+
+FOOTER = f"""
+<div style="text-align:center;margin:20px auto 6px;padding:18px;border-radius:14px;
+     background:#2563eb;color:#fff;max-width:1080px;">
+  <div style="font-size:1.2em;font-weight:800;">Das KI-Büro für Ihre Dokumente</div>
+  <div style="opacity:.92;margin:6px 0 12px;">OCR · Vision · KI-Assistent · REST-API — on-prem, DSGVO (Frankfurt).</div>
+  <a href="{WEBSITE}" target="_blank" style="background:#fff;color:#2563eb;font-weight:800;
+     padding:9px 20px;border-radius:8px;text-decoration:none;">→ german-ocr.de</a>
+</div>
+"""
+
+
+def build():
+    theme = gr.themes.Soft(primary_hue="blue", neutral_hue="slate")
+    with gr.Blocks(title="German-OCR · GOCR", theme=theme) as demo:
+        gr.HTML(HEADER)
+        with gr.Row():
+            with gr.Column(scale=1):
+                inp = gr.Image(label="Dokument", type="numpy", height=420)
+                btn = gr.Button("🚀 Dokument erkennen", variant="primary", size="lg")
+                status = gr.Textbox(label="Status", interactive=False)
+                if os.path.isdir(EXAMPLES_DIR) and os.listdir(EXAMPLES_DIR):
+                    gr.Examples(examples=[os.path.join(EXAMPLES_DIR, f)
+                                          for f in sorted(os.listdir(EXAMPLES_DIR))],
+                                inputs=inp, label="Beispiele")
+            with gr.Column(scale=1):
+                with gr.Tab("📦 Boxen"):
+                    out_boxes = gr.Image(label="Erkannte Regionen (Farbe je Box)",
+                                         interactive=False, height=420)
+                with gr.Tab("🎭 Maske"):
+                    out_mask = gr.Image(label="Detektions-Maske", interactive=False, height=420)
+                with gr.Tab("🔤 Text-Layout"):
+                    out_tlay = gr.Image(label="Erkannte Werte an Position",
+                                        interactive=False, height=420)
+                with gr.Tab("📋 Text"):
+                    out_txt = gr.Textbox(label="Volltext (Lesereihenfolge)", lines=15)
+                with gr.Tab("🔧 JSON"):
+                    out_json = gr.Code(label="Strukturiertes JSON — direkt fürs LLM/Backend",
+                                       language="json")
+        gr.HTML(FOOTER)
+        btn.click(run, inputs=inp,
+                  outputs=[out_boxes, out_mask, out_tlay, out_txt, out_json, status])
+    return demo
+
+
+if __name__ == "__main__":
+    build().launch(server_name="0.0.0.0", server_port=7860,
+                   allowed_paths=[EXAMPLES_DIR, os.path.expanduser("~/.cache")])

requirements.txt

@@ -0,0 +1,2 @@
+git+https://github.com/Keyvanhardani/g-ocr
+matplotlib