Suya OCR API — vLLM-backed, OpenAI-compatible OCR service

FastAPI service wrapping the Surya-OCR-2 model (datalab-to) served through vLLM: legacy /v1/api/ai/* endpoints, an OpenAI-compatible /v1/chat/completions endpoint, a coalescing request batcher, a local OCR CLI, Docker packaging, multilingual example outputs, and quantization/concurrency benchmarks. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-06-17 10:20:02 +04:00
commit 1a585693be
147 changed files with 13827 additions and 0 deletions
@@ -0,0 +1,59 @@
+import click
+import copy
+import json
+import time
+from collections import defaultdict
+
+from surya.detection import DetectionPredictor
+from surya.debug.draw import draw_polys_on_image
+from surya.logging import configure_logging, get_logger
+from surya.scripts.config import CLILoader
+import os
+
+configure_logging()
+logger = get_logger()
+
+
+@click.command(help="Detect bboxes in an input file or folder (PDFs or image).")
+@CLILoader.common_options
+def detect_text_cli(input_path: str, **kwargs):
+    loader = CLILoader(input_path, kwargs)
+
+    det_predictor = DetectionPredictor()
+
+    start = time.time()
+    predictions = det_predictor(loader.images, include_maps=loader.debug)
+    end = time.time()
+    if loader.debug:
+        logger.debug(f"Detection took {end - start} seconds")
+
+    if loader.save_images:
+        for idx, (image, pred, name) in enumerate(
+            zip(loader.images, predictions, loader.names)
+        ):
+            polygons = [p.polygon for p in pred.bboxes]
+            bbox_image = draw_polys_on_image(polygons, copy.deepcopy(image))
+            bbox_image.save(os.path.join(loader.result_path, f"{name}_{idx}_bbox.png"))
+
+            if loader.debug:
+                heatmap = pred.heatmap
+                heatmap.save(os.path.join(loader.result_path, f"{name}_{idx}_heat.png"))
+
+    predictions_by_page = defaultdict(list)
+    for idx, (pred, name, image) in enumerate(
+        zip(predictions, loader.names, loader.images)
+    ):
+        out_pred = pred.model_dump(exclude=["heatmap", "affinity_map"])
+        out_pred["page"] = len(predictions_by_page[name]) + 1
+        predictions_by_page[name].append(out_pred)
+
+    with open(
+        os.path.join(loader.result_path, "results.json"), "w+", encoding="utf-8"
+    ) as f:
+        json.dump(predictions_by_page, f, ensure_ascii=False)
+
+    logger.info(f"Wrote results to {loader.result_path}")
+
+
+if __name__ == "__main__":
+    detect_text_cli()