added tiling num to annotations
Browse files- annotation_tab/annotation_logic.py +57 -43
- annotation_tab/annotation_setup.py +9 -5
- app.py +3 -2
- inference_tab/inference_setup.py +6 -5
annotation_tab/annotation_logic.py
CHANGED
|
@@ -3,25 +3,16 @@ import pandas as pd
|
|
| 3 |
import threading
|
| 4 |
import gradio as gr
|
| 5 |
from config import OUTPUT_DIR
|
|
|
|
| 6 |
|
| 7 |
# ==== CONFIG ====
|
| 8 |
IMAGE_FOLDER = os.path.join(OUTPUT_DIR, "blobs")
|
| 9 |
os.makedirs(IMAGE_FOLDER, exist_ok=True)
|
| 10 |
-
CSV_FILE = os.path.join(OUTPUT_DIR, "annotations.csv")
|
| 11 |
|
| 12 |
-
|
| 13 |
-
|
| 14 |
-
if not all_images_paths:
|
| 15 |
-
return "No images loaded"
|
| 16 |
-
return f"Image {current_index + 1} of {len(all_images_paths)}"
|
| 17 |
|
| 18 |
# ==== STATE ====
|
| 19 |
-
if os.path.exists(CSV_FILE):
|
| 20 |
-
df_annotations = pd.read_csv(CSV_FILE, dtype={"blob_id": str, "human_ocr": str})
|
| 21 |
-
else:
|
| 22 |
-
df_annotations = pd.DataFrame(columns=["blob_id", "human_ocr"])
|
| 23 |
-
df_annotations.to_csv(CSV_FILE, index=False)
|
| 24 |
-
|
| 25 |
all_images = [
|
| 26 |
f for f in os.listdir(IMAGE_FOLDER)
|
| 27 |
if f.lower().endswith(('.png', '.jpg', '.jpeg')) and '_margin' in f
|
|
@@ -29,34 +20,66 @@ all_images = [
|
|
| 29 |
all_images_paths = [os.path.join(IMAGE_FOLDER, f) for f in all_images]
|
| 30 |
current_index = 0
|
| 31 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 32 |
# ==== HELPERS ====
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 33 |
def get_current_image_path():
|
| 34 |
if 0 <= current_index < len(all_images_paths):
|
| 35 |
return all_images_paths[current_index]
|
| 36 |
return None
|
| 37 |
|
| 38 |
-
def
|
| 39 |
-
""
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 40 |
blob_id = os.path.basename(image_path).replace("_margin", "")
|
| 41 |
row = df_annotations[df_annotations["blob_id"] == blob_id]
|
| 42 |
if not row.empty:
|
| 43 |
val = str(row["human_ocr"].values[-1]).strip()
|
| 44 |
-
return val != "" #
|
| 45 |
return False
|
| 46 |
|
| 47 |
-
def
|
| 48 |
blob_id = os.path.basename(image_path).replace("_margin", "")
|
| 49 |
row = df_annotations[df_annotations["blob_id"] == blob_id]
|
| 50 |
if not row.empty:
|
| 51 |
-
return str(row["human_ocr"].values[-1])
|
| 52 |
-
return
|
|
|
|
|
|
|
|
|
|
|
|
|
| 53 |
|
| 54 |
-
def
|
|
|
|
| 55 |
n = len(all_images_paths)
|
| 56 |
idx = start
|
| 57 |
for _ in range(n):
|
| 58 |
idx = (idx + 1) % n
|
| 59 |
-
if not
|
| 60 |
return idx
|
| 61 |
return None
|
| 62 |
|
|
@@ -64,7 +87,6 @@ def find_next_unannotated_index(start):
|
|
| 64 |
def save_annotation(user_text):
|
| 65 |
"""Save the current annotation for the active image."""
|
| 66 |
global df_annotations
|
| 67 |
-
|
| 68 |
img_path = get_current_image_path()
|
| 69 |
if not img_path:
|
| 70 |
return
|
|
@@ -90,15 +112,14 @@ def save_and_next(user_text):
|
|
| 90 |
|
| 91 |
save_annotation(user_text)
|
| 92 |
|
| 93 |
-
|
| 94 |
-
if all(is_annotated(p) for p in all_images_paths):
|
| 95 |
current_index = 0
|
| 96 |
img_path = get_current_image_path()
|
| 97 |
annotation = get_annotation_for_image(img_path)
|
| 98 |
return img_path, annotation, gr.update(visible=True, value="All images annotated."), img_path, get_progress_text()
|
| 99 |
|
| 100 |
-
next_idx =
|
| 101 |
-
current_index = next_idx
|
| 102 |
img_path = get_current_image_path()
|
| 103 |
annotation = get_annotation_for_image(img_path)
|
| 104 |
return img_path, annotation, gr.update(visible=False), img_path, get_progress_text()
|
|
@@ -116,7 +137,6 @@ def previous_image():
|
|
| 116 |
def delete_and_next():
|
| 117 |
"""Mark current image as DELETED and move to next image."""
|
| 118 |
global current_index, df_annotations
|
| 119 |
-
|
| 120 |
img_path = get_current_image_path()
|
| 121 |
if not img_path:
|
| 122 |
return None, "", gr.update(visible=True, value="No images available."), "No image loaded", "No images loaded"
|
|
@@ -133,41 +153,29 @@ def delete_and_next():
|
|
| 133 |
|
| 134 |
df_annotations.to_csv(CSV_FILE, index=False)
|
| 135 |
|
| 136 |
-
|
| 137 |
-
if all(is_annotated(p) for p in all_images_paths):
|
| 138 |
current_index = 0
|
| 139 |
img_path = get_current_image_path()
|
| 140 |
annotation = get_annotation_for_image(img_path)
|
| 141 |
return img_path, annotation, gr.update(visible=True, value="All images annotated."), img_path, get_progress_text()
|
| 142 |
|
| 143 |
-
|
| 144 |
-
|
| 145 |
-
if next_idx is not None:
|
| 146 |
-
current_index = next_idx
|
| 147 |
-
else:
|
| 148 |
-
current_index = 0
|
| 149 |
-
|
| 150 |
img_path = get_current_image_path()
|
| 151 |
annotation = get_annotation_for_image(img_path)
|
| 152 |
return img_path, annotation, gr.update(visible=False), img_path, get_progress_text()
|
| 153 |
|
| 154 |
-
|
| 155 |
def save_and_exit(user_text):
|
| 156 |
if get_current_image_path() is not None:
|
| 157 |
save_annotation(user_text)
|
| 158 |
threading.Timer(1, lambda: os._exit(0)).start()
|
| 159 |
return None, "", gr.update(visible=True, value="Session closed."), "", get_progress_text()
|
| 160 |
|
| 161 |
-
def get_current_annotations_path():
|
| 162 |
-
return CSV_FILE
|
| 163 |
-
|
| 164 |
def refresh_image_list():
|
| 165 |
-
"""Reload images
|
| 166 |
global all_images_paths, current_index, df_annotations
|
| 167 |
|
| 168 |
-
|
| 169 |
-
df_annotations.to_csv(CSV_FILE, index=False)
|
| 170 |
-
|
| 171 |
all_images = [
|
| 172 |
f for f in os.listdir(IMAGE_FOLDER)
|
| 173 |
if f.lower().endswith(('.png', '.jpg', '.jpeg')) and '_margin' in f
|
|
@@ -175,9 +183,15 @@ def refresh_image_list():
|
|
| 175 |
all_images_paths = [os.path.join(IMAGE_FOLDER, f) for f in all_images]
|
| 176 |
current_index = 0
|
| 177 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 178 |
if not all_images_paths:
|
| 179 |
return None, "", gr.update(visible=True, value="No images available."), "No image loaded", "No images loaded"
|
| 180 |
|
|
|
|
| 181 |
img_path = get_current_image_path()
|
| 182 |
annotation = get_annotation_for_image(img_path)
|
| 183 |
return img_path, annotation, gr.update(visible=False), img_path, get_progress_text()
|
|
|
|
| 3 |
import threading
|
| 4 |
import gradio as gr
|
| 5 |
from config import OUTPUT_DIR
|
| 6 |
+
import re
|
| 7 |
|
| 8 |
# ==== CONFIG ====
|
| 9 |
IMAGE_FOLDER = os.path.join(OUTPUT_DIR, "blobs")
|
| 10 |
os.makedirs(IMAGE_FOLDER, exist_ok=True)
|
|
|
|
| 11 |
|
| 12 |
+
CSV_FILE = None
|
| 13 |
+
df_annotations = pd.DataFrame(columns=["blob_id", "human_ocr"])
|
|
|
|
|
|
|
|
|
|
| 14 |
|
| 15 |
# ==== STATE ====
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 16 |
all_images = [
|
| 17 |
f for f in os.listdir(IMAGE_FOLDER)
|
| 18 |
if f.lower().endswith(('.png', '.jpg', '.jpeg')) and '_margin' in f
|
|
|
|
| 20 |
all_images_paths = [os.path.join(IMAGE_FOLDER, f) for f in all_images]
|
| 21 |
current_index = 0
|
| 22 |
|
| 23 |
+
# ==== TILE CSV ====
|
| 24 |
+
def switch_tile_csv(selected_tile):
|
| 25 |
+
global CSV_FILE, df_annotations
|
| 26 |
+
tile_filename = os.path.basename(selected_tile["tile_path"])
|
| 27 |
+
tile_name, _ = os.path.splitext(tile_filename)
|
| 28 |
+
CSV_FILE = os.path.join(OUTPUT_DIR, f"annotations_{tile_name}.csv")
|
| 29 |
+
|
| 30 |
+
if os.path.exists(CSV_FILE):
|
| 31 |
+
df_annotations = pd.read_csv(CSV_FILE, dtype={"blob_id": str, "human_ocr": str})
|
| 32 |
+
else:
|
| 33 |
+
df_annotations = pd.DataFrame(columns=["blob_id", "human_ocr"])
|
| 34 |
+
df_annotations.to_csv(CSV_FILE, index=False)
|
| 35 |
+
|
| 36 |
+
return CSV_FILE
|
| 37 |
+
|
| 38 |
# ==== HELPERS ====
|
| 39 |
+
def get_progress_text():
|
| 40 |
+
if not all_images_paths:
|
| 41 |
+
return "No images loaded"
|
| 42 |
+
return f"Image {current_index + 1} of {len(all_images_paths)}"
|
| 43 |
+
|
| 44 |
def get_current_image_path():
|
| 45 |
if 0 <= current_index < len(all_images_paths):
|
| 46 |
return all_images_paths[current_index]
|
| 47 |
return None
|
| 48 |
|
| 49 |
+
def get_annotation_for_image(image_path):
|
| 50 |
+
blob_id = os.path.basename(image_path).replace("_margin", "")
|
| 51 |
+
row = df_annotations[df_annotations["blob_id"] == blob_id]
|
| 52 |
+
if not row.empty:
|
| 53 |
+
return str(row["human_ocr"].values[-1])
|
| 54 |
+
return ""
|
| 55 |
+
|
| 56 |
+
def is_annotated_or_deleted(image_path):
|
| 57 |
+
"""Return True if image has an annotation or is deleted."""
|
| 58 |
blob_id = os.path.basename(image_path).replace("_margin", "")
|
| 59 |
row = df_annotations[df_annotations["blob_id"] == blob_id]
|
| 60 |
if not row.empty:
|
| 61 |
val = str(row["human_ocr"].values[-1]).strip()
|
| 62 |
+
return val != "" # includes 'DELETED' as counted
|
| 63 |
return False
|
| 64 |
|
| 65 |
+
def is_deleted(image_path):
|
| 66 |
blob_id = os.path.basename(image_path).replace("_margin", "")
|
| 67 |
row = df_annotations[df_annotations["blob_id"] == blob_id]
|
| 68 |
if not row.empty:
|
| 69 |
+
return str(row["human_ocr"].values[-1]).strip() == "DELETED"
|
| 70 |
+
return False
|
| 71 |
+
|
| 72 |
+
def all_processed():
|
| 73 |
+
"""Return True if all images are either annotated or deleted."""
|
| 74 |
+
return all(is_annotated_or_deleted(p) for p in all_images_paths)
|
| 75 |
|
| 76 |
+
def find_next_unprocessed_index(start):
|
| 77 |
+
"""Return the next image index that is neither annotated nor deleted."""
|
| 78 |
n = len(all_images_paths)
|
| 79 |
idx = start
|
| 80 |
for _ in range(n):
|
| 81 |
idx = (idx + 1) % n
|
| 82 |
+
if not is_annotated_or_deleted(all_images_paths[idx]):
|
| 83 |
return idx
|
| 84 |
return None
|
| 85 |
|
|
|
|
| 87 |
def save_annotation(user_text):
|
| 88 |
"""Save the current annotation for the active image."""
|
| 89 |
global df_annotations
|
|
|
|
| 90 |
img_path = get_current_image_path()
|
| 91 |
if not img_path:
|
| 92 |
return
|
|
|
|
| 112 |
|
| 113 |
save_annotation(user_text)
|
| 114 |
|
| 115 |
+
if all_processed():
|
|
|
|
| 116 |
current_index = 0
|
| 117 |
img_path = get_current_image_path()
|
| 118 |
annotation = get_annotation_for_image(img_path)
|
| 119 |
return img_path, annotation, gr.update(visible=True, value="All images annotated."), img_path, get_progress_text()
|
| 120 |
|
| 121 |
+
next_idx = find_next_unprocessed_index(current_index)
|
| 122 |
+
current_index = next_idx if next_idx is not None else 0
|
| 123 |
img_path = get_current_image_path()
|
| 124 |
annotation = get_annotation_for_image(img_path)
|
| 125 |
return img_path, annotation, gr.update(visible=False), img_path, get_progress_text()
|
|
|
|
| 137 |
def delete_and_next():
|
| 138 |
"""Mark current image as DELETED and move to next image."""
|
| 139 |
global current_index, df_annotations
|
|
|
|
| 140 |
img_path = get_current_image_path()
|
| 141 |
if not img_path:
|
| 142 |
return None, "", gr.update(visible=True, value="No images available."), "No image loaded", "No images loaded"
|
|
|
|
| 153 |
|
| 154 |
df_annotations.to_csv(CSV_FILE, index=False)
|
| 155 |
|
| 156 |
+
if all_processed():
|
|
|
|
| 157 |
current_index = 0
|
| 158 |
img_path = get_current_image_path()
|
| 159 |
annotation = get_annotation_for_image(img_path)
|
| 160 |
return img_path, annotation, gr.update(visible=True, value="All images annotated."), img_path, get_progress_text()
|
| 161 |
|
| 162 |
+
next_idx = find_next_unprocessed_index(current_index)
|
| 163 |
+
current_index = next_idx if next_idx is not None else 0
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 164 |
img_path = get_current_image_path()
|
| 165 |
annotation = get_annotation_for_image(img_path)
|
| 166 |
return img_path, annotation, gr.update(visible=False), img_path, get_progress_text()
|
| 167 |
|
|
|
|
| 168 |
def save_and_exit(user_text):
|
| 169 |
if get_current_image_path() is not None:
|
| 170 |
save_annotation(user_text)
|
| 171 |
threading.Timer(1, lambda: os._exit(0)).start()
|
| 172 |
return None, "", gr.update(visible=True, value="Session closed."), "", get_progress_text()
|
| 173 |
|
|
|
|
|
|
|
|
|
|
| 174 |
def refresh_image_list():
|
| 175 |
+
"""Reload images for the current tile and clear the CSV."""
|
| 176 |
global all_images_paths, current_index, df_annotations
|
| 177 |
|
| 178 |
+
# Reload images
|
|
|
|
|
|
|
| 179 |
all_images = [
|
| 180 |
f for f in os.listdir(IMAGE_FOLDER)
|
| 181 |
if f.lower().endswith(('.png', '.jpg', '.jpeg')) and '_margin' in f
|
|
|
|
| 183 |
all_images_paths = [os.path.join(IMAGE_FOLDER, f) for f in all_images]
|
| 184 |
current_index = 0
|
| 185 |
|
| 186 |
+
# Reset CSV for current tile
|
| 187 |
+
df_annotations = pd.DataFrame(columns=["blob_id", "human_ocr"])
|
| 188 |
+
if CSV_FILE:
|
| 189 |
+
df_annotations.to_csv(CSV_FILE, index=False)
|
| 190 |
+
|
| 191 |
if not all_images_paths:
|
| 192 |
return None, "", gr.update(visible=True, value="No images available."), "No image loaded", "No images loaded"
|
| 193 |
|
| 194 |
+
# Return first image
|
| 195 |
img_path = get_current_image_path()
|
| 196 |
annotation = get_annotation_for_image(img_path)
|
| 197 |
return img_path, annotation, gr.update(visible=False), img_path, get_progress_text()
|
annotation_tab/annotation_setup.py
CHANGED
|
@@ -1,12 +1,14 @@
|
|
|
|
|
| 1 |
import gradio as gr
|
|
|
|
|
|
|
| 2 |
from .annotation_logic import (
|
| 3 |
save_and_next, previous_image, delete_and_next, save_and_exit,
|
| 4 |
get_current_image_path, get_annotation_for_image,
|
| 5 |
-
|
| 6 |
)
|
| 7 |
|
| 8 |
-
|
| 9 |
-
def get_annotation_widgets():
|
| 10 |
message = gr.Markdown("", visible=False)
|
| 11 |
image_path_display = gr.Markdown(
|
| 12 |
value=get_current_image_path() or "No image loaded",
|
|
@@ -53,8 +55,10 @@ def get_annotation_widgets():
|
|
| 53 |
#exit_btn.click(save_and_exit, inputs=txt, outputs=[img, txt, message, image_path_display, progress_display])
|
| 54 |
|
| 55 |
download_btn.click(
|
| 56 |
-
lambda: get_current_annotations_path(),
|
| 57 |
-
|
|
|
|
|
|
|
| 58 |
)
|
| 59 |
|
| 60 |
return [
|
|
|
|
| 1 |
+
import os
|
| 2 |
import gradio as gr
|
| 3 |
+
import pandas as pd
|
| 4 |
+
from config import OUTPUT_DIR
|
| 5 |
from .annotation_logic import (
|
| 6 |
save_and_next, previous_image, delete_and_next, save_and_exit,
|
| 7 |
get_current_image_path, get_annotation_for_image,
|
| 8 |
+
refresh_image_list, switch_tile_csv
|
| 9 |
)
|
| 10 |
|
| 11 |
+
def get_annotation_widgets(selected_tile_state):
|
|
|
|
| 12 |
message = gr.Markdown("", visible=False)
|
| 13 |
image_path_display = gr.Markdown(
|
| 14 |
value=get_current_image_path() or "No image loaded",
|
|
|
|
| 55 |
#exit_btn.click(save_and_exit, inputs=txt, outputs=[img, txt, message, image_path_display, progress_display])
|
| 56 |
|
| 57 |
download_btn.click(
|
| 58 |
+
#lambda: get_current_annotations_path(),
|
| 59 |
+
fn=lambda selected_tile: switch_tile_csv(selected_tile),
|
| 60 |
+
inputs=[selected_tile_state],
|
| 61 |
+
outputs=[download_file]
|
| 62 |
)
|
| 63 |
|
| 64 |
return [
|
app.py
CHANGED
|
@@ -15,10 +15,11 @@ logging.basicConfig(level=logging.DEBUG)
|
|
| 15 |
|
| 16 |
|
| 17 |
with gr.Blocks() as demo:
|
|
|
|
| 18 |
with gr.Tab("Inference"):
|
| 19 |
-
image_input, gcp_input, city_name,user_crs, score_th, hist_th, hist_dic, run_button, output, download_file = get_inference_widgets(run_inference,georefImg)
|
| 20 |
with gr.Tab("Annotation"):
|
| 21 |
-
get_annotation_widgets()
|
| 22 |
with gr.Tab("Map"):
|
| 23 |
get_map_widgets(city_name)
|
| 24 |
|
|
|
|
| 15 |
|
| 16 |
|
| 17 |
with gr.Blocks() as demo:
|
| 18 |
+
selected_tile_state=gr.State(value=None)
|
| 19 |
with gr.Tab("Inference"):
|
| 20 |
+
image_input, gcp_input, city_name,user_crs, score_th, hist_th, hist_dic, run_button, output, download_file = get_inference_widgets(run_inference,georefImg, selected_tile_state)
|
| 21 |
with gr.Tab("Annotation"):
|
| 22 |
+
get_annotation_widgets(selected_tile_state)
|
| 23 |
with gr.Tab("Map"):
|
| 24 |
get_map_widgets(city_name)
|
| 25 |
|
inference_tab/inference_setup.py
CHANGED
|
@@ -77,7 +77,7 @@ def enable_textbox(file):
|
|
| 77 |
return gr.update(interactive=bool(file))
|
| 78 |
|
| 79 |
|
| 80 |
-
def get_inference_widgets(run_inference,georefImg):
|
| 81 |
with gr.Row():
|
| 82 |
# Left column
|
| 83 |
with gr.Column(scale=1,min_width=500):
|
|
@@ -120,7 +120,8 @@ def get_inference_widgets(run_inference,georefImg):
|
|
| 120 |
file_types=[".csv"],
|
| 121 |
type="filepath")
|
| 122 |
|
| 123 |
-
|
|
|
|
| 124 |
|
| 125 |
|
| 126 |
# Wire events
|
|
@@ -129,12 +130,12 @@ def get_inference_widgets(run_inference,georefImg):
|
|
| 129 |
outputs=[annotated_out, run_button]
|
| 130 |
)
|
| 131 |
annotated_out.select(
|
| 132 |
-
fn=select_tile, inputs=[
|
| 133 |
-
outputs=[selected_tile, run_button,
|
| 134 |
)
|
| 135 |
run_button.click(
|
| 136 |
fn=run_inference,
|
| 137 |
-
inputs=[
|
| 138 |
outputs=[output, download_file]
|
| 139 |
)
|
| 140 |
|
|
|
|
| 77 |
return gr.update(interactive=bool(file))
|
| 78 |
|
| 79 |
|
| 80 |
+
def get_inference_widgets(run_inference,georefImg,selected_tile_state):
|
| 81 |
with gr.Row():
|
| 82 |
# Left column
|
| 83 |
with gr.Column(scale=1,min_width=500):
|
|
|
|
| 120 |
file_types=[".csv"],
|
| 121 |
type="filepath")
|
| 122 |
|
| 123 |
+
# pass globally instead
|
| 124 |
+
#selected_tile_state = gr.State()
|
| 125 |
|
| 126 |
|
| 127 |
# Wire events
|
|
|
|
| 130 |
outputs=[annotated_out, run_button]
|
| 131 |
)
|
| 132 |
annotated_out.select(
|
| 133 |
+
fn=select_tile, inputs=[selected_tile_state],
|
| 134 |
+
outputs=[selected_tile, run_button, selected_tile_state]
|
| 135 |
)
|
| 136 |
run_button.click(
|
| 137 |
fn=run_inference,
|
| 138 |
+
inputs=[selected_tile_state, gcp_input,user_crs, city_name, score_th, hist_th,hist_dic],
|
| 139 |
outputs=[output, download_file]
|
| 140 |
)
|
| 141 |
|