helm / app.py
ethanrom's picture
Update app.py
2b7f277
import cv2
import torch
import torch.backends.cudnn as cudnn
from models.experimental import attempt_load
from utils.general import non_max_suppression
from torchvision import models
from torchvision import transforms
from PIL import Image
import time
import streamlit as st
import IPython
import numpy as np
yolov5_weight_file = 'model100e.pt'
device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
yolov5_model = attempt_load(yolov5_weight_file, device=device, inplace=True, fuse=True)
cudnn.benchmark = True
names = yolov5_model.module.names if hasattr(yolov5_model, 'module') else yolov5_model.names
conf_set = 0.1
frame_size = (800, 480)
colors = {
'helmet': (255, 0, 0),
'rider': (0, 255, 0),
'number': (0, 0, 255),
'no_helmet': (0, 100, 255),
}
def detect_objects(frame):
img = torch.from_numpy(frame)
img = img.permute(2, 0, 1).float().to(device)
img /= 255.0
if img.ndimension() == 3:
img = img.unsqueeze(0)
with torch.no_grad():
pred = yolov5_model(img, augment=False)[0]
pred = non_max_suppression(pred, conf_set, 0.30)
detections = []
for det in pred:
if len(det):
for d in det: # d = (x1, y1, x2, y2, conf, cls)
x1 = int(d[0].item())
y1 = int(d[1].item())
x2 = int(d[2].item())
y2 = int(d[3].item())
conf = round(d[4].item(), 2)
c = int(d[5].item())
detected_name = names[c]
detections.append((x1, y1, x2, y2, conf, detected_name))
color = colors.get(detected_name, (255, 255, 255))
cv2.rectangle(frame, (x1, y1), (x2, y2), color, 2)
cv2.putText(frame, detected_name, (x1, y1), cv2.FONT_HERSHEY_DUPLEX, 1, color, 2)
return detections
def display_detections(input_image, output_image, detections):
for det in detections:
x1, y1, x2, y2, conf, detected_name = det
color = colors.get(detected_name, (255, 255, 255))
cv2.rectangle(output_image, (x1, y1), (x2, y2), color, 2)
cv2.putText(output_image, f"{detected_name} ({conf:.2f})", (x1, y1 - 10), cv2.FONT_HERSHEY_SIMPLEX, 0.5, color, 2)
return output_image
def app():
st.title("Helmet Detection App")
st.write("This app uses YOLOv5 to detect helmets and riders in images and videos.")
# Select input type
input_type = st.radio("Select input type:", options=["Image", "Video"])
# Upload file or use webcam
if input_type == "Image":
uploaded_file = st.file_uploader("Upload image", type=["jpg", "jpeg", "png"])
if uploaded_file is not None:
image = Image.open(uploaded_file)
st.image(image, caption="Uploaded Image", use_column_width=True)
detections = detect_objects(np.array(image))
output_image = display_detections(np.array(image), np.array(image), detections)
st.image(output_image, caption="Output Image", use_column_width=True)
elif input_type == "Video":
st.write("Select an option to get the input video:")
video_option = st.radio("", options=["Webcam", "Upload video"])
if video_option == "Webcam":
cap = cv2.VideoCapture(0)
elif video_option == "Upload video":
uploaded_file = st.file_uploader("Upload video", type=["mp4"])
if uploaded_file is not None:
temp_file = NamedTemporaryFile(delete=False)
temp_file.write(uploaded_file.read())
st.write("Video uploaded successfully!")
cap = cv2.VideoCapture(temp_file.name)
if 'cap' in locals():
frame_size = (800, 480)
show_video = st.checkbox("Show video", value=True)
save_video = st.checkbox("Save video", value=False)
font = cv2.FONT_HERSHEY_DUPLEX
while True:
ret, frame = cap.read()
if ret:
frame = cv2.resize(frame, frame_size)
detections = detect_objects(frame)
display_frame = display_detections(frame, detections)
fps = 1 / (time.time() - start_time)
start_time = time.time()
cv2.putText(display_frame, f'FPS: {fps:.2f}', (10, 30), font, 1, (0, 255, 0), 2, cv2.LINE_AA)
if show_video:
stframe.image(display_frame, channels="BGR")
if save_video:
out.write(display_frame)
if cv2.waitKey(1) & 0xFF == ord('q'):
break
cap.release()
if save_video:
out.release()
if __name__ == "__main__":
app()