Added easyocr for text extraction

handler.py

@@ -18,6 +18,7 @@ from huggingface_hub import logging
 from concurrent.futures import ThreadPoolExecutor, as_completed
 
 import timeit
+import easyocr
 
 device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
 
@@ -47,6 +48,7 @@ class EndpointHandler:
 
         logging.set_verbosity_debug()
         self.logger = logging.get_logger(__name__)
+        self.reader = easyocr.Reader(["de", "en"])
 
     def download_image(self, url: str) -> bytes:
         """
@@ -183,6 +185,12 @@ class EndpointHandler:
                                 "source_type": "images",
                                 **image_metadata,
                             }
+                            # Extract text from image using easyocr
+                            extracted_text = self.reader.readtext(
+                                np.array(image), detail=0
+                            )
+                            complete_image_metadata["extracted_text"] = extracted_text
+
                             processed_metadata.append(complete_image_metadata)
 
                         except Exception as e:

requirements.txt

@@ -21,4 +21,6 @@ tokenizers==0.13.3
 tqdm==4.66.1
 transformers==4.27.2
 typing_extensions==4.8.0
-urllib3==2.0.7
+urllib3==2.0.7
+easyocr==1.7.1
+opencv_python_headless==4.8.1