'anchors': [[10, 14], [23, 27], [37, 58], [81, 82], [135, 169],
[344, 319]],
'anchors_mask': [[1, 2, 3], [3, 4, 5]],
'image_shape': (416, 416),
'num_classes': 80,
'score_threshold': 0.6,
'iou_threshold': 0.3,
'batch_size': 2,
'drop_rate': 0.2,
'block_size': 3
}
class_name = get_classes('./coco_classes.txt')
model = creat_CSPYOLO(**default)
load_weights(model, './yolov4-tiny.weights')
img = cv2.imread('./kite.jpg')
img = resiresize_img(img)
img = padding_img(img)
pred_img = np.expand_dims(img, axis=0).astype(np.float32) / 255
result = model.predict(pred_img)
boxes, class_, score = output_result(result, **default)
draw_img(img, boxes, class_, score, class_name, **default)
cv2.imwrite('./predict.jpg', img)
cv2.imshow('img', img)
cv2.waitKey(0)
cv2.destroyAllWindows()
boxes = preprocess_true_boxes(loc, **self.model_dict)
image_data.append(img)
y_true.append(boxes)
image_data = (np.array(image_data) / 255).astype(np.float32)
y_true = np.array(y_true)
return image_data, y_true
if __name__ == "__main__":
model = creat_CSPYOLO(**train_default)
if train_default['use_pretrain']:
custom = False if train_default['num_classes'] == 80 else True
load_weights(model, './yolov4.weights', custom_cls=custom)
model.summary()
annotation_path = 'path'
log_dir = 'path'
classes_path = 'path'
class_name = get_classes(classes_path)
xmls = os.listdir(annotation_path)
total_train = len(xmls)
callbacks = [tf.keras.callbacks.ModelCheckpoint(filepath= log_dir + 'best_loss.h5',
save_best_only=True,
save_weights_only=True,
monitor='loss',
[36, 75], [76, 55], [72, 146],
[142, 110], [192, 243], [459, 401]],
'anchors_mask': [[0, 1, 2], [3, 4, 5], [6, 7, 8]],
'image_shape': (608, 608),
'num_classes': 80,
'score_threshold': 0.6,
'iou_threshold': 0.3,
'batch_size': 2,
'drop_rate' : 0.2,
'block_size' : 3
}
class_name = get_classes('./coco_classes.txt')
model = creat_CSPYOLO(**default)
load_weights(model, './yolov4.weights')
img = cv2.imread('./kite.jpg')
img = resiresize_img(img)
img = padding_img(img)
pred_img = np.expand_dims(img, axis=0).astype(np.float32) / 255
result = model.predict(pred_img)
boxes, class_, score = output_result(result, **default)
draw_img(img, boxes, class_, score, class_name, **default)
cv2.imwrite('./predict.jpg', img)
cv2.imshow('img', img)
cv2.waitKey(0)
cv2.destroyAllWindows()