Spaces:
Running
Running
Update app.py
Browse files
app.py
CHANGED
|
@@ -7,68 +7,51 @@ import gradio as gr
|
|
| 7 |
from ultralytics import YOLO
|
| 8 |
from paddleocr import PaddleOCR
|
| 9 |
|
| 10 |
-
#
|
| 11 |
yolo = YOLO("models/best.pt")
|
| 12 |
ocr = PaddleOCR(
|
| 13 |
-
|
| 14 |
-
rec=True,
|
| 15 |
text_recognition_model_dir="models/ocr_model",
|
| 16 |
use_textline_orientation=False,
|
| 17 |
lang="en"
|
| 18 |
)
|
| 19 |
|
| 20 |
-
#
|
| 21 |
def format_turkish_plate(s: str) -> str:
|
| 22 |
s = re.sub(r'[^A-Z0-9]', '', s.upper())
|
| 23 |
m = re.match(r'^(\d{2})([A-Z]{1,3})(\d{2,4})$', s)
|
| 24 |
return f"{m.group(1)} {m.group(2)} {m.group(3)}" if m else "Unknown"
|
| 25 |
|
| 26 |
-
#
|
| 27 |
-
def extract_text_score(recs):
|
| 28 |
-
if not recs:
|
| 29 |
-
return "", 0.0
|
| 30 |
-
first = recs[0]
|
| 31 |
-
# format: [text, score]
|
| 32 |
-
if isinstance(first, (list,tuple)) and len(first)==2 and isinstance(first[0], str):
|
| 33 |
-
text, score = first[0], first[1]
|
| 34 |
-
# format: [[...], (text,score)] veya [[...], [text,score]]
|
| 35 |
-
elif isinstance(first, (list,tuple)) and len(first)==2 and isinstance(first[1], (list,tuple)):
|
| 36 |
-
text, score = first[1][0], first[1][1]
|
| 37 |
-
else:
|
| 38 |
-
return "", 0.0
|
| 39 |
-
try:
|
| 40 |
-
score = float(score)
|
| 41 |
-
except:
|
| 42 |
-
score = 0.0
|
| 43 |
-
return text, score
|
| 44 |
-
|
| 45 |
-
# βββ 4) Singleβimage inference βββββββββββββββββββββββββββββββββββββ
|
| 46 |
def run_image(img, conf=0.25):
|
| 47 |
bgr = cv2.cvtColor(img, cv2.COLOR_RGB2BGR)
|
| 48 |
res = yolo(bgr, conf=conf)[0]
|
| 49 |
out = bgr.copy()
|
| 50 |
|
| 51 |
-
for box in res.boxes.xyxy.cpu().numpy()
|
| 52 |
-
|
|
|
|
| 53 |
crop = out[y1:y2, x1:x2]
|
| 54 |
-
if crop.size
|
| 55 |
-
continue
|
| 56 |
|
| 57 |
plate_img = cv2.resize(crop, (128,32))
|
| 58 |
-
|
| 59 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
| 60 |
|
| 61 |
-
raw, score = extract_text_score(recs)
|
| 62 |
plate = format_turkish_plate(raw)
|
| 63 |
-
label = f"{plate} ({
|
| 64 |
|
| 65 |
cv2.rectangle(out, (x1,y1),(x2,y2), (0,255,0), 2)
|
| 66 |
-
cv2.putText(out, label, (x1,
|
| 67 |
cv2.FONT_HERSHEY_SIMPLEX, 0.6, (0,255,0), 2)
|
| 68 |
|
| 69 |
return cv2.cvtColor(out, cv2.COLOR_BGR2RGB), f"{len(res.boxes)} plate(s) detected"
|
| 70 |
|
| 71 |
-
#
|
| 72 |
def run_video(video_file, conf=0.25):
|
| 73 |
cap = cv2.VideoCapture(video_file)
|
| 74 |
fps = cap.get(cv2.CAP_PROP_FPS)
|
|
@@ -86,21 +69,21 @@ def run_video(video_file, conf=0.25):
|
|
| 86 |
res = yolo(frame, conf=conf)[0]
|
| 87 |
for (x1,y1,x2,y2) in res.boxes.xyxy.cpu().numpy().astype(int):
|
| 88 |
crop = frame[y1:y2, x1:x2]
|
| 89 |
-
if crop.size
|
| 90 |
-
continue
|
| 91 |
-
|
| 92 |
plate_img = cv2.resize(crop, (128,32))
|
| 93 |
-
recs = ocr.ocr(plate_img,
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 94 |
|
| 95 |
-
raw, score = extract_text_score(recs)
|
| 96 |
plate = format_turkish_plate(raw)
|
| 97 |
if plate != "Unknown":
|
| 98 |
-
records.append({"time_s":round(t,2),
|
| 99 |
-
"plate":plate,
|
| 100 |
-
"conf":round(score,3)})
|
| 101 |
|
| 102 |
cv2.rectangle(frame, (x1,y1),(x2,y2), (0,255,0), 2)
|
| 103 |
-
cv2.putText(frame, plate, (x1,
|
| 104 |
cv2.FONT_HERSHEY_SIMPLEX, 0.6, (0,255,0), 2)
|
| 105 |
|
| 106 |
writer.write(frame)
|
|
@@ -108,25 +91,25 @@ def run_video(video_file, conf=0.25):
|
|
| 108 |
cap.release(); writer.release()
|
| 109 |
with open("output.json","w") as f:
|
| 110 |
json.dump(records, f, indent=2)
|
| 111 |
-
return outfp
|
| 112 |
|
| 113 |
-
#
|
| 114 |
with gr.Blocks() as demo:
|
| 115 |
gr.Markdown("## π License Plate Detection + Recognition")
|
| 116 |
with gr.Row():
|
| 117 |
with gr.Column():
|
| 118 |
-
img_in = gr.Image(type="numpy"
|
| 119 |
-
vid_in = gr.File(label="
|
| 120 |
-
conf = gr.Slider(0,1,0.25,0.01
|
| 121 |
b1 = gr.Button("Run Image")
|
| 122 |
b2 = gr.Button("Run Video")
|
| 123 |
with gr.Column():
|
| 124 |
-
img_out = gr.Image(type="numpy"
|
| 125 |
-
vid_out = gr.Video(
|
| 126 |
-
status = gr.Textbox(
|
| 127 |
|
| 128 |
b1.click(run_image, [img_in,conf], [img_out,status])
|
| 129 |
b2.click(run_video, [vid_in,conf], [vid_out,status])
|
| 130 |
|
| 131 |
if __name__=="__main__":
|
| 132 |
-
demo.launch()
|
|
|
|
| 7 |
from ultralytics import YOLO
|
| 8 |
from paddleocr import PaddleOCR
|
| 9 |
|
| 10 |
+
# 1) Load models
|
| 11 |
yolo = YOLO("models/best.pt")
|
| 12 |
ocr = PaddleOCR(
|
| 13 |
+
text_detection_model_dir=None,
|
|
|
|
| 14 |
text_recognition_model_dir="models/ocr_model",
|
| 15 |
use_textline_orientation=False,
|
| 16 |
lang="en"
|
| 17 |
)
|
| 18 |
|
| 19 |
+
# 2) Turkish plate formatter
|
| 20 |
def format_turkish_plate(s: str) -> str:
|
| 21 |
s = re.sub(r'[^A-Z0-9]', '', s.upper())
|
| 22 |
m = re.match(r'^(\d{2})([A-Z]{1,3})(\d{2,4})$', s)
|
| 23 |
return f"{m.group(1)} {m.group(2)} {m.group(3)}" if m else "Unknown"
|
| 24 |
|
| 25 |
+
# 3) Singleβimage inference
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 26 |
def run_image(img, conf=0.25):
|
| 27 |
bgr = cv2.cvtColor(img, cv2.COLOR_RGB2BGR)
|
| 28 |
res = yolo(bgr, conf=conf)[0]
|
| 29 |
out = bgr.copy()
|
| 30 |
|
| 31 |
+
for box, yolo_score in zip(res.boxes.xyxy.cpu().numpy(),
|
| 32 |
+
res.boxes.conf.cpu().numpy()):
|
| 33 |
+
x1,y1,x2,y2 = box.astype(int)
|
| 34 |
crop = out[y1:y2, x1:x2]
|
| 35 |
+
if crop.size==0: continue
|
|
|
|
| 36 |
|
| 37 |
plate_img = cv2.resize(crop, (128,32))
|
| 38 |
+
recs = ocr.ocr(plate_img, cls=False)
|
| 39 |
+
|
| 40 |
+
if recs and recs[0]:
|
| 41 |
+
_, (raw, ocr_score) = recs[0]
|
| 42 |
+
else:
|
| 43 |
+
raw, ocr_score = "", 0.0
|
| 44 |
|
|
|
|
| 45 |
plate = format_turkish_plate(raw)
|
| 46 |
+
label = f"{plate} ({ocr_score:.2f})"
|
| 47 |
|
| 48 |
cv2.rectangle(out, (x1,y1),(x2,y2), (0,255,0), 2)
|
| 49 |
+
cv2.putText(out, label, (x1,y1-5),
|
| 50 |
cv2.FONT_HERSHEY_SIMPLEX, 0.6, (0,255,0), 2)
|
| 51 |
|
| 52 |
return cv2.cvtColor(out, cv2.COLOR_BGR2RGB), f"{len(res.boxes)} plate(s) detected"
|
| 53 |
|
| 54 |
+
# 4) Video inference (same pattern)
|
| 55 |
def run_video(video_file, conf=0.25):
|
| 56 |
cap = cv2.VideoCapture(video_file)
|
| 57 |
fps = cap.get(cv2.CAP_PROP_FPS)
|
|
|
|
| 69 |
res = yolo(frame, conf=conf)[0]
|
| 70 |
for (x1,y1,x2,y2) in res.boxes.xyxy.cpu().numpy().astype(int):
|
| 71 |
crop = frame[y1:y2, x1:x2]
|
| 72 |
+
if crop.size==0: continue
|
|
|
|
|
|
|
| 73 |
plate_img = cv2.resize(crop, (128,32))
|
| 74 |
+
recs = ocr.ocr(plate_img, cls=False)
|
| 75 |
+
|
| 76 |
+
if recs and recs[0]:
|
| 77 |
+
_, (raw, ocr_score) = recs[0]
|
| 78 |
+
else:
|
| 79 |
+
raw, ocr_score = "", 0.0
|
| 80 |
|
|
|
|
| 81 |
plate = format_turkish_plate(raw)
|
| 82 |
if plate != "Unknown":
|
| 83 |
+
records.append({"time_s":round(t,2), "plate":plate, "conf":round(ocr_score,3)})
|
|
|
|
|
|
|
| 84 |
|
| 85 |
cv2.rectangle(frame, (x1,y1),(x2,y2), (0,255,0), 2)
|
| 86 |
+
cv2.putText(frame, plate, (x1,y1-5),
|
| 87 |
cv2.FONT_HERSHEY_SIMPLEX, 0.6, (0,255,0), 2)
|
| 88 |
|
| 89 |
writer.write(frame)
|
|
|
|
| 91 |
cap.release(); writer.release()
|
| 92 |
with open("output.json","w") as f:
|
| 93 |
json.dump(records, f, indent=2)
|
| 94 |
+
return outfp
|
| 95 |
|
| 96 |
+
# 5) Gradio UI
|
| 97 |
with gr.Blocks() as demo:
|
| 98 |
gr.Markdown("## π License Plate Detection + Recognition")
|
| 99 |
with gr.Row():
|
| 100 |
with gr.Column():
|
| 101 |
+
img_in = gr.Image(type="numpy")
|
| 102 |
+
vid_in = gr.File(label="Video (.mp4)")
|
| 103 |
+
conf = gr.Slider(0,1,0.25,0.01)
|
| 104 |
b1 = gr.Button("Run Image")
|
| 105 |
b2 = gr.Button("Run Video")
|
| 106 |
with gr.Column():
|
| 107 |
+
img_out = gr.Image(type="numpy")
|
| 108 |
+
vid_out = gr.Video()
|
| 109 |
+
status = gr.Textbox()
|
| 110 |
|
| 111 |
b1.click(run_image, [img_in,conf], [img_out,status])
|
| 112 |
b2.click(run_video, [vid_in,conf], [vid_out,status])
|
| 113 |
|
| 114 |
if __name__=="__main__":
|
| 115 |
+
demo.launch()
|