small fixes

annotation_tab/annotation_logic.py

@@ -7,7 +7,7 @@ from config import OUTPUT_DIR
 # ==== CONFIG ====
 IMAGE_FOLDER = os.path.join(OUTPUT_DIR,"blobs")
 os.makedirs(IMAGE_FOLDER, exist_ok=True)
-CSV_FILE = os.path.join(OUTPUT_DIR,"annotations")
+CSV_FILE = os.path.join(OUTPUT_DIR,"annotations.csv")
 
 # ==== STATE ====
 if os.path.exists(CSV_FILE):
@@ -56,7 +56,7 @@ def save_annotation(user_text):
         if filename in annotated_ids:
             df_annotations.loc[df_annotations["blob_id"] == filename, "human_ocr"] = text_value
         else:
-            new_row = pd.DataFrame([{"blob_id": filename, "human_ocr": text_value}])
+            new_row = pd.DataFrame([{"blob_id": os.path.splitext(filename)[0], "human_ocr": text_value}])
             df_annotations = pd.concat([df_annotations, new_row], ignore_index=True)
             annotated_ids.add(filename)
 
@@ -117,3 +117,8 @@ def save_and_exit(user_text):
         save_annotation(user_text)
     threading.Timer(1, shutdown).start()
     return None, "", gr.update(visible=True, value="Session closed."), ""
+
+
+def get_current_annotations_path():
+    import os
+    return os.path.join(OUTPUT_DIR, "annotations.csv")

annotation_tab/annotation_setup.py

@@ -1,7 +1,7 @@
 import gradio as gr
 from .annotation_logic import (
     save_and_next, previous_image, delete_and_next, save_and_exit,
-    get_current_image_path, get_annotation_for_image
+    get_current_image_path, get_annotation_for_image, get_current_annotations_path
 )
 
 def get_annotation_widgets():
@@ -16,10 +16,16 @@ def get_annotation_widgets():
         next_btn = gr.Button("Save & Next")
         del_btn = gr.Button("Delete & Next", variant="stop")
         exit_btn = gr.Button("Save & Exit", variant="secondary")
+        download_btn = gr.Button("Save Annotations")
+    with gr.Row():
+        download_file = gr.File(label="Download CSV", interactive=False)  
+
 
     next_btn.click(save_and_next, inputs=txt, outputs=[img, txt, message, image_path_display])
     prev_btn.click(previous_image, outputs=[img, txt, message, image_path_display])
     del_btn.click(delete_and_next, outputs=[img, txt, message, image_path_display])
     exit_btn.click(save_and_exit, inputs=txt, outputs=[img, txt, message, image_path_display])
+    download_btn.click(lambda: get_current_annotations_path(),outputs=download_file)
+
 
     return [message, image_path_display, img, txt, hint, prev_btn, next_btn, del_btn, exit_btn]

app.py

@@ -1,3 +1,6 @@
+# [DEBUG]
+#from osgeo import gdal
+
 import gradio as gr
 import logging
 from inference_tab import get_inference_widgets, run_inference
@@ -6,14 +9,17 @@ from annotation_tab import get_annotation_widgets
 # setup logging
 logging.basicConfig(level=logging.DEBUG)
 
+
 with gr.Blocks() as demo:
     with gr.Tab("Inference"):
         get_inference_widgets(run_inference)
     with gr.Tab("Annotation"):
         get_annotation_widgets()
 
+# [DEBUG]
+#demo.launch(inbrowser=True)
 
-
+# [PROD]
 demo.launch(server_name="0.0.0.0", server_port=7860, inbrowser=False)
 
 

inference_tab/inference_logic.py

@@ -46,15 +46,13 @@ def run_inference(image_path, gcp_path, city_name, score_th):
         yield msg, None
     
     # === POST OCR ===
-    for msg in fuzzyMatch():
+    for msg in fuzzyMatch(score_th):
         if msg.endswith(".csv"):
-            yield f"Finished! CSV saved at {msg}", msg
+            yield f"Finished! CSV saved at {msg}. Street labels are ready for manual input.", msg
         else:
             yield msg, None
 
-    return f"Street labels are ready for manual input.\nImage: {image_path}", None
-
-
+    
 
 def getBBoxes(image_path, tile_size=256, overlap=0.3, confidence_threshold=0.25):
     yield f"DEBUG: Received image_path: {image_path}"
@@ -542,7 +540,7 @@ def extractSegments(image_path, min_size=500, margin=10):
 
 
 def blobsOCR(image_path):
-    
+    yield "Load OCR model.."
     # Load model + processor
     processor = TrOCRProcessor.from_pretrained("microsoft/trocr-base-str")
     model = VisionEncoderDecoderModel.from_pretrained("muk42/trocr_streets")
@@ -558,7 +556,7 @@ def blobsOCR(image_path):
 
 
     # Open output file for writing
-    OCR_PATH = os.path.join(OUTPUT_DIR,"ocr")
+    OCR_PATH = os.path.join(OUTPUT_DIR,"ocr.csv")
     with open(OCR_PATH, "w", encoding="utf-8") as f_out:
         # Process each image
         image_folder = os.path.join(OUTPUT_DIR,"blobs")
@@ -684,15 +682,21 @@ def georefImg(image_path, gcp_path):
 
     
 
-    yield f"Done."
+    yield "Done."
 
 
 def extractStreetNet(city_name):
     yield f"Extract OSM street network for {city_name}"
-    G = ox.graph_from_place(city_name, network_type='drive')
+    G = ox.graph_from_place(city_name, network_type='all')
     G_proj = ox.project_graph(G)
-    nodes, edges = ox.graph_to_gdfs(G_proj)
+    edges = ox.graph_to_gdfs(G_proj, nodes=False, edges=True, fill_edge_geometry=True)
     edges_3857 = edges.to_crs(epsg=3857)
+    edges_3857 = edges_3857[['osmid','name', 'geometry']]
+    edges_3857 = edges_3857[edges_3857['name'].notnull()]
+
+    edges_3857['name'] = edges_3857['name'].apply(
+                            lambda x: x[0] if isinstance(x, list) and len(x) > 0 else x)
+
     OSM_PATH=os.path.join(OUTPUT_DIR,"osm_extract.geojson")
     edges_3857.to_file(OSM_PATH, driver="GeoJSON")
     yield "Done."
@@ -709,11 +713,13 @@ def best_street_match(point, query_name, edges_gdf, max_distance=100):
         best_match = process.extractOne(query_name, candidate_names, scorer=fuzz.ratio)
         return best_match  # (name, score, index)
 
-def fuzzyMatch():
+def fuzzyMatch(score_th):
     COORD_PATH=os.path.join(OUTPUT_DIR,"centroids.csv")
     OCR_PATH=os.path.join(OUTPUT_DIR,"ocr.csv")
     coords_df = pd.read_csv(COORD_PATH)
-    names_df = pd.read_csv(OCR_PATH,sep="\t",columns=[['blob_id','pred_text']]) 
+    names_df = pd.read_csv(OCR_PATH,
+                           names=['blob_id','pred_text'],
+                           dtype={"blob_id": "int64", "pred_text": "string"}) 
     merged_df = coords_df.merge(names_df, on="blob_id")
 
     gdf = gpd.GeoDataFrame(
@@ -723,13 +729,12 @@ def fuzzyMatch():
     )
 
     OSM_PATH=os.path.join(OUTPUT_DIR,"osm_extract.geojson")
-    osm_gdf = gpd.read_file(OSM_PATH)
-    osm_gdf = osm_gdf[osm_gdf['name'].notnull()]
+    osm_gdf = gpd.read_file(OSM_PATH,dtype={"name": "str"})
 
     yield "Process OSM candidates..."
     results = []
     for _, row in gdf.iterrows():
-        match = best_street_match(row.geometry, row['name'], osm_gdf, max_distance=100)
+        match = best_street_match(row.geometry, row['pred_text'], osm_gdf, max_distance=100)
         if match:
             results.append({
                 "blob_id": row.blob_id,
@@ -752,4 +757,13 @@ def fuzzyMatch():
     results_df = pd.DataFrame(results)
     RES_PATH=os.path.join(OUTPUT_DIR,"street_matches.csv")
     results_df.to_csv(RES_PATH, index=False)
-    yield f"{RES_PATH}/street_matches.csv"
+
+    # remove street labels from blobs folder that are more than or equal to score threshold
+    manual_df = results_df[results_df['osm_match_score'] >= int(score_th)]
+    for blob_id in manual_df['blob_id']:
+        file_path = os.path.join(OUTPUT_DIR,"blobs",f"{blob_id}.png") 
+        
+        if os.path.exists(file_path):
+            os.remove(file_path)
+
+    yield f"{RES_PATH}"

requirements.txt

@@ -13,4 +13,5 @@ Shapely==2.1.1
 torch==2.7.1
 transformers==4.53.2
 ultralytics==8.3.94
+huggingface_hub[hf_xet]
 GDAL==3.6.2