Update src/streamlit_app.py

src/streamlit_app.py

@@ -262,9 +262,10 @@ col1, col2 = st.columns([3, 2])
 display_field_name = SINGLE_FIELDS[0]
 storage_field_name = SINGLE_FIELDS[0]
 base_field_for_color = SINGLE_FIELDS[0]
+field_color = FIELD_COLORS[base_field_for_color]
 
+# ====== RHS TOP: FIELD SELECTION + ZOOM ======
 with col2:
-    # ========== FIELD SELECTION (RHS) ==========
     st.markdown("#### 🎯 Field Selection")
     
     def add_line_item():
@@ -293,7 +294,6 @@ with col2:
     else:
         num_items = st.session_state.num_line_items[selected_name]
         
-        # Single level of columns inside col2 (no deeper nesting)
         line_col1, add_col, rem_col = st.columns([2, 1, 1])
         with line_col1:
             line_item_options = [f"Line {i+1}" for i in range(num_items)]
@@ -311,14 +311,15 @@ with col2:
         storage_field_name = f"Line {line_item_num}: {base_field}"
         base_field_for_color = base_field
 
-    # Guard in case something weird happens
     if not storage_field_name:
         storage_field_name = display_field_name
 
     field_color = FIELD_COLORS.get(base_field_for_color or display_field_name, "#FF0000")
-    st.markdown(f"**Current:** <span style='color:{field_color}'>●</span> {display_field_name}", unsafe_allow_html=True)
+    st.markdown(
+        f"**Current:** <span style='color:{field_color}'>●</span> {display_field_name}",
+        unsafe_allow_html=True,
+    )
 
-    # ========== ZOOM CONTROLS (RHS) ==========
     st.markdown("#### 🔍 Zoom")
     
     current_zoom = st.session_state.zoom_values[selected_name]
@@ -369,115 +370,13 @@ with col2:
 
     st.caption(f"Original: {pil_image.width}×{pil_image.height}")
 
-    # ========== OCR & VALUE SECTION ==========
-    st.markdown("#### ✏️ OCR & Value")
-
-    current_rect_orig = st.session_state.field_rects_orig[selected_name].get(storage_field_name)
-    current_val = st.session_state.field_values[selected_name].get(storage_field_name, "")
-    
-    new_val = st.text_area(
-        "Value (auto-filled by OCR)",
-        value=current_val,
-        height=80,
-        label_visibility="collapsed",
-        placeholder="Value (auto-filled by OCR)",
-    )
-    
-    col_btn1, col_btn2, col_btn3 = st.columns(3)
-    
-    with col_btn1:
-        if st.button("💾 Save"):
-            st.session_state.field_values[selected_name][storage_field_name] = new_val
-            st.success("Saved!")
-    
-    with col_btn2:
-        if current_rect_orig and st.button("🔄 Re-OCR"):
-            x1 = max(0, int(current_rect_orig["left"]))
-            y1 = max(0, int(current_rect_orig["top"]))
-            x2 = min(pil_image.width, int(current_rect_orig["left"] + current_rect_orig["width"]))
-            y2 = min(pil_image.height, int(current_rect_orig["top"] + current_rect_orig["height"]))
-            if x2 > x1 and y2 > y1:
-                crop = pil_image.crop((x1, y1, x2, y2))
-                try:
-                    text = pytesseract.image_to_string(crop, config="--psm 6").strip()
-                    if text:
-                        st.session_state.field_values[selected_name][storage_field_name] = text
-                        st.success(f"OCR: {text}")
-                    else:
-                        st.warning("Empty result")
-                except Exception as e:
-                    st.error(f"OCR failed: {e}")
-            # force rerun so text area picks up new value
-            st.experimental_rerun()
-    
-    with col_btn3:
-        def delete_rect():
-            st.session_state.pending_delete = (selected_name, storage_field_name)
-        
-        if current_rect_orig:
-            st.button("🗑️ Delete", on_click=delete_rect)
-
-    # ========== ALL VALUES SECTION ==========
-    with st.expander("📋 All Values"):
-        for f in SINGLE_FIELDS:
-            v = st.session_state.field_values[selected_name].get(f, "")
-            if v.strip():
-                st.write(f"**{f}:** {v}")
-        num_items = st.session_state.num_line_items[selected_name]
-        for i in range(1, num_items + 1):
-            vals = [(lif, st.session_state.field_values[selected_name].get(f"Line {i}: {lif}", "")) 
-                    for lif in LINE_ITEM_FIELDS]
-            vals = [(lif, v) for lif, v in vals if v.strip()]
-            if vals:
-                st.write(f"**Line {i}:**")
-                for lif, v in vals:
-                    st.write(f"  {lif}: {v}")
-
-    # ========== EXPORT SECTION ==========
-    st.markdown("#### 📤 JSONL Export")
-
-    # Export ALL labeled remittances
-    records_all = [
-        build_gt_record_for_file(img["name"])
-        for img in images
-        if has_any_label(img["name"])
-    ]
-
-    if records_all:
-        all_jsonl_str = "\n".join(
-            json.dumps(rec, ensure_ascii=False) for rec in records_all
-        )
-        st.download_button(
-            "⬇️ Export ALL labeled (JSONL)",
-            data=all_jsonl_str.encode("utf-8"),
-            file_name="remittances_ground_truth.jsonl",
-            mime="application/json",
-        )
-    else:
-        st.caption("No labeled remittances yet.")
-
-    # Export CURRENT remittance
-    current_record = build_gt_record_for_file(selected_name)
-    with st.expander("Preview CURRENT JSON"):
-        st.json(current_record)
-
-    current_jsonl_str = json.dumps(current_record, ensure_ascii=False) + "\n"
-    st.download_button(
-        "⬇️ Export CURRENT (JSONL)",
-        data=current_jsonl_str.encode("utf-8"),
-        file_name=f"{os.path.splitext(selected_name)[0]}_remittance.jsonl",
-        mime="application/json",
-    )
-
+# ====== LHS: CANVAS / IMAGE ======
 with col1:
-    # ========== CANVAS / IMAGE (LEFT SIDE) ==========
     zoom = st.session_state.zoom_values[selected_name]
-    
     scale = zoom / 100.0
     disp_w = int(pil_image.width * scale)
     disp_h = int(pil_image.height * scale)
     
-    # Get display image - fresh PIL object each time from stable bytes
     display_image = get_display_image_from_bytes(image_bytes, disp_w, disp_h)
 
     def orig_to_display(rect_orig, s):
@@ -518,12 +417,10 @@ with col1:
     initial_drawing = {"version": "4.4.0", "objects": all_display_objects}
     expected_count = len(all_display_objects)
 
-    # Canvas key: includes rect count to force refresh when rectangles change
     rect_ver = st.session_state.rect_version[selected_name]
     num_rects = len(st.session_state.field_rects_orig[selected_name])
     canvas_key = f"canvas_{selected_name}_z{zoom}_rv{rect_ver}_n{num_rects}"
 
-    # Render canvas
     canvas_result = st_canvas(
         background_image=display_image,
         height=disp_h,
@@ -537,29 +434,32 @@ with col1:
         key=canvas_key,
     )
 
-    # Detect new rectangle
+    # Detect new rectangle and auto-OCR
     if canvas_result.json_data is not None:
         objs = canvas_result.json_data.get("objects", [])
         if len(objs) > expected_count:
             new_rect_display = objs[-1]
             new_rect_orig = display_to_orig(new_rect_display, scale)
             new_rect_orig["stroke"] = field_color
+
+            # overwrite previous rect for this field
             st.session_state.field_rects_orig[selected_name][storage_field_name] = new_rect_orig
-            # bump rect version so canvas resets to only newest rect for this field
             st.session_state.rect_version[selected_name] += 1
-            
-            # Auto-run OCR
+
             x1 = max(0, int(new_rect_orig["left"]))
             y1 = max(0, int(new_rect_orig["top"]))
             x2 = min(pil_image.width, int(new_rect_orig["left"] + new_rect_orig["width"]))
             y2 = min(pil_image.height, int(new_rect_orig["top"] + new_rect_orig["height"]))
-            
+
             if x2 > x1 and y2 > y1:
                 crop = pil_image.crop((x1, y1, x2, y2))
                 try:
                     text = pytesseract.image_to_string(crop, config="--psm 6").strip()
                     if text:
+                        # update field values and the text-area state key
                         st.session_state.field_values[selected_name][storage_field_name] = text
+                        value_state_key = f"value_{selected_name}_{storage_field_name}"
+                        st.session_state[value_state_key] = text
                         st.toast(f"✅ OCR: {text[:50]}{'...' if len(text) > 50 else ''}")
                     else:
                         st.toast("✅ Rectangle saved (no text detected)")
@@ -567,5 +467,111 @@ with col1:
                     st.toast("✅ Rectangle saved")
             else:
                 st.toast("✅ Rectangle saved")
-            # force rerun so RHS text box shows new OCR value
-            st.experimental_rerun()
+
+# ====== RHS BOTTOM: OCR VALUE + EXPORT ======
+with col2:
+    st.markdown("#### ✏️ OCR & Value")
+
+    current_rect_orig = st.session_state.field_rects_orig[selected_name].get(storage_field_name)
+    value_state_key = f"value_{selected_name}_{storage_field_name}"
+
+    # initialise text state from saved field value on first use
+    if value_state_key not in st.session_state:
+        st.session_state[value_state_key] = st.session_state.field_values[selected_name].get(
+            storage_field_name, ""
+        )
+
+    col_btn1, col_btn2, col_btn3 = st.columns(3)
+
+    with col_btn1:
+        if st.button("💾 Save"):
+            st.session_state.field_values[selected_name][storage_field_name] = st.session_state[value_state_key]
+            st.success("Saved!")
+
+    with col_btn2:
+        if current_rect_orig and st.button("🔄 Re-OCR"):
+            x1 = max(0, int(current_rect_orig["left"]))
+            y1 = max(0, int(current_rect_orig["top"]))
+            x2 = min(pil_image.width, int(current_rect_orig["left"] + current_rect_orig["width"]))
+            y2 = min(pil_image.height, int(current_rect_orig["top"] + current_rect_orig["height"]))
+            if x2 > x1 and y2 > y1:
+                crop = pil_image.crop((x1, y1, x2, y2))
+                try:
+                    text = pytesseract.image_to_string(crop, config="--psm 6").strip()
+                    if text:
+                        st.session_state.field_values[selected_name][storage_field_name] = text
+                        st.session_state[value_state_key] = text
+                        st.success(f"OCR: {text}")
+                    else:
+                        st.warning("Empty result")
+                except Exception as e:
+                    st.error(f"OCR failed: {e}")
+
+    with col_btn3:
+        def delete_rect():
+            st.session_state.pending_delete = (selected_name, storage_field_name)
+        if current_rect_orig:
+            st.button("🗑️ Delete", on_click=delete_rect)
+
+    # Text area bound to state key (so it reflects auto-OCR & Re-OCR without reruns)
+    st.text_area(
+        "Value (auto-filled by OCR)",
+        key=value_state_key,
+        height=80,
+        label_visibility="collapsed",
+        placeholder="Value (auto-filled by OCR)",
+    )
+
+    # ========== ALL VALUES SECTION ==========
+    with st.expander("📋 All Values"):
+        for f in SINGLE_FIELDS:
+            v = st.session_state.field_values[selected_name].get(f, "")
+            if v.strip():
+                st.write(f"**{f}:** {v}")
+        num_items = st.session_state.num_line_items[selected_name]
+        for i in range(1, num_items + 1):
+            vals = [
+                (lif, st.session_state.field_values[selected_name].get(f"Line {i}: {lif}", ""))
+                for lif in LINE_ITEM_FIELDS
+            ]
+            vals = [(lif, v) for lif, v in vals if v.strip()]
+            if vals:
+                st.write(f"**Line {i}:**")
+                for lif, v in vals:
+                    st.write(f"  {lif}: {v}")
+
+    # ========== EXPORT SECTION ==========
+    st.markdown("#### 📤 JSONL Export")
+
+    # Export ALL labeled remittances
+    records_all = [
+        build_gt_record_for_file(img["name"])
+        for img in images
+        if has_any_label(img["name"])
+    ]
+
+    if records_all:
+        all_jsonl_str = "\n".join(
+            json.dumps(rec, ensure_ascii=False) for rec in records_all
+        )
+        st.download_button(
+            "⬇️ Export ALL labeled (JSONL)",
+            data=all_jsonl_str.encode("utf-8"),
+            file_name="remittances_ground_truth.jsonl",
+            mime="application/json",
+        )
+    else:
+        st.caption("No labeled remittances yet.")
+
+    # Export CURRENT remittance
+    current_record = build_gt_record_for_file(selected_name)
+    with st.expander("Preview CURRENT JSON"):
+        st.json(current_record)
+
+    current_jsonl_str = json.dumps(current_record, ensure_ascii=False) + "\n"
+    st.download_button(
+        "⬇️ Export CURRENT (JSONL)",
+        data=current_jsonl_str.encode("utf-8"),
+        file_name=f"{os.path.splitext(selected_name)[0]}_remittance.jsonl",
+        mime="application/json",
+    )