CTX-5257: Changed document-ocr-fn's date_of_birth output to return ob…

…ject with numerical fields repesenting day, month and year

tasks/document-ocr-fn/resources/function/ocr.py

@@ -1,14 +1,9 @@
+from dateutil import parser
+
 from PIL import Image
 from transformers import TrOCRProcessor, VisionEncoderDecoderModel
 
-import pytesseract
-import easyocr
-import numpy as np
-
 
-# EasyOCR
-reader = easyocr.Reader(["en"])
-# TrOCR
 modelVersion = "microsoft/trocr-base-printed"
 processor = TrOCRProcessor.from_pretrained(modelVersion)
 model = VisionEncoderDecoderModel.from_pretrained(modelVersion)
@@ -20,18 +15,30 @@ def trOCR(image: Image.Image) -> str:
     return processor.batch_decode(generatedIds, skip_special_tokens = True)[0]  # type: ignore
 
 
-def performOCR(images: list[Image.Image], classes: list[str]) -> list[dict[str, str]]:
-    detections: list[dict[str, str]] = []
+def performOCR(images: list[Image.Image], classes: list[str]) -> dict[str, str]:
+    detections: dict[str, str] = {}
     for i, image in enumerate(images):
-        # tesseractOutput = pytesseract.image_to_string(image).replace("\n", " ")
-        # easyOcrOutput = " ".join([e[1] for e in reader.readtext(np.array(image))])
         trOcrOutput = trOCR(image)
 
-        detections.append({
-            "class": classes[i],
-            # "tesseract": tesseractOutput,
-            # "easyOCR": easyOcrOutput,
-            "trOCR": trOcrOutput
-        })
+        if classes[i] == "date_of_birth":
+            detections["date_of_birth_raw"] = trOcrOutput
+
+            try:
+                trOcrDateTime = parser.parse(trOcrOutput)
+                trOcrOutputJson = {
+                    "year": trOcrDateTime.year,
+                    "month": trOcrDateTime.month,
+                    "day": trOcrDateTime.day
+                }
+            except ValueError as e:
+                trOcrOutputJson = {
+                    "year": None,
+                    "month": None,
+                    "day": None
+                }
+
+            trOcrOutput = trOcrOutputJson
+
+        detections[classes[i]] = trOcrOutput
 
     return detections

tasks/document-ocr-fn/resources/function/requirements.txt

@@ -13,9 +13,7 @@ seaborn>=0.11.0
 coremltools==6.3.0
 tensorflow==2.8
 tensorflowjs>=3.9.0
+transformers==4.36.2
+ultralytics==8.0.226
 thop
 coretex
-pytesseract
-easyocr
-transformers
-ultralytics