debug set and locking

src/doctr_service/dataset_manager.py

@@ -42,4 +42,33 @@ class ImageTextDataset:
         with open(txt_path, "r", encoding="utf-8") as f:
             text = f.read()
 
-        return image, text
+        return image, text
+
+    def get_output_path(self, idx, output_subdir, debugset_root="/app/debugset"):
+        """Get output path for saving OCR result to debugset folder.
+
+        Args:
+            idx: Sample index
+            output_subdir: Subdirectory name (e.g., 'paddle_text', 'doctr_text')
+            debugset_root: Root folder for debug output (default: /app/debugset)
+
+        Returns:
+            Path like /app/debugset/doc1/{output_subdir}/page_001.txt
+        """
+        img_path, _ = self.samples[idx]
+        # img_path: /app/dataset/doc1/img/page_001.png
+        # Extract relative path: doc1/img/page_001.png
+        parts = img_path.split("/dataset/", 1)
+        if len(parts) == 2:
+            rel_path = parts[1]  # doc1/img/page_001.png
+        else:
+            rel_path = os.path.basename(img_path)
+
+        # Replace /img/ with /{output_subdir}/
+        rel_parts = rel_path.rsplit("/img/", 1)
+        doc_folder = rel_parts[0]  # doc1
+        fname = os.path.splitext(rel_parts[1])[0] + ".txt"  # page_001.txt
+
+        out_dir = os.path.join(debugset_root, doc_folder, output_subdir)
+        os.makedirs(out_dir, exist_ok=True)
+        return os.path.join(out_dir, fname)

src/doctr_service/docker-compose.yml

@@ -14,6 +14,7 @@ services:
       - "8003:8000"
     volumes:
       - ../dataset:/app/dataset:ro
+      - ../debugset:/app/debugset:rw
       - doctr-cache:/root/.cache/doctr
     environment:
       - PYTHONUNBUFFERED=1
@@ -35,6 +36,7 @@ services:
       - "8003:8000"
     volumes:
       - ../dataset:/app/dataset:ro
+      - ../debugset:/app/debugset:rw
       - doctr-cache:/root/.cache/doctr
     environment:
       - PYTHONUNBUFFERED=1

src/doctr_service/doctr_tuning_rest.py

@@ -169,6 +169,7 @@ class EvaluateRequest(BaseModel):
     # Page range
     start_page: int = Field(5, ge=0, description="Start page index (inclusive)")
     end_page: int = Field(10, ge=1, description="End page index (exclusive)")
+    save_output: bool = Field(False, description="Save OCR predictions to debugset folder")
 
 
 class EvaluateResponse(BaseModel):
@@ -302,6 +303,12 @@ def evaluate(request: EvaluateRequest):
             )
             time_per_page_list.append(float(time.time() - tp0))
 
+            # Save prediction to debugset if requested
+            if request.save_output:
+                out_path = state.dataset.get_output_path(idx, "doctr_text")
+                with open(out_path, "w", encoding="utf-8") as f:
+                    f.write(pred)
+
             m = evaluate_text(ref, pred)
             cer_list.append(m["CER"])
             wer_list.append(m["WER"])