def evaluate(label_indices = {'1': 0, '2': 1, '3': 2 ,'4': 3,'5': 4},
channel_means = np.array([147.12697, 160.21092, 167.70029]),
data_path = '../data_trial',
minibatch_size = 32,
num_batches_to_test = 10,
checkpoint_dir = 'tf_data/sample_model'):
print("1. Loading data")
data = data_loader(label_indices = label_indices,
channel_means = channel_means,
train_test_split = 0.5,
data_path = data_path)
print("2. Instantiating the model")
M = Model(mode = 'test')
#Evaluate on test images:
GT = Generator(data.test.X, data.test.y, minibatch_size = minibatch_size)
num_correct = 0
num_total = 0
print("3. Evaluating on test images")
for i in range(num_batches_to_test):
GT.generate()
yhat = M.predict(X = GT.X, checkpoint_dir = checkpoint_dir)
correct_predictions = (np.argmax(yhat, axis = 1) == np.argmax(GT.y, axis = 1))
num_correct += np.sum(correct_predictions)
num_total += len(correct_predictions)
accuracy = round(num_correct/num_total,4)
return accuracy
def do_run(self):
for i in range(0, self.num_simulations):
model = Model(self.model_config)
model.initialize_agents()
result = model.do_update()
self.results.append(result)
self.write_results(model)
def evaluate_with_path(image_path):
im = cv2.imread(image_path)
input_image_size = (227, 227)
im = cv2.resize(im, (input_image_size[0], input_image_size[1]))
channel_means = np.array([147.12697, 160.21092, 167.70029])
X = np.zeros((1, input_image_size[0], input_image_size[1], 3),
dtype='float32')
X[0, :, :, :] = im - channel_means
M = Model(mode='test')
#Evaluate on test images:
# print(f'Predicted {}')
return M.predict(X)
def evaluate(label_indices={
'mountain_bikes': 1,
'road_bikes': 0
},
channel_means=np.array([147.12697, 160.21092, 167.70029]),
data_path='test_data',
minibatch_size=32,
num_batches_to_test=10,
checkpoint_dir='tf_data/sample_model'):
print("1. Loading data")
data = data_loader(label_indices=label_indices,
channel_means=channel_means,
train_test_split=0,
data_path=data_path)
print("2. Instantiating the model")
M = Model(mode='test')
#Evaluate on test images:
accuracy = M.test(data)
return accuracy
## ## Simple Training Script ## from sample_model import Model M = Model(mode='train') M.train()
import numpy as np
import time
from sample_model import Model
from data_loader import data_loader
from generator import Generator
checkpoint_dir='tf_data/sample_model'
X='C:/Users/Karthick/Desktop/cvproject/data/5/00000_00000.ppmspeed_2_.ppm'
M = Model(mode = 'test')
yhat = M.predict(X = X, checkpoint_dir = checkpoint_dir)
# save_dir="C:/Users/Karthick/Desktop/cvproject/speedlimitckp/"
# #saver = tf.train.Saver()
# sess = tf.Session()
# saver = tf.train.import_meta_graph('C:/Users/Karthick/Desktop/cvproject/src/tf_data/sample_model/model_epoch70.ckpt.meta')
# saver.restore(sess,tf.train.latest_checkpoint('C:/Users/Karthick/Desktop/cvproject/src/tf_data/sample_model/'))
# #checkpoint_name = tf.train.latest_checkpoint(save_dir)
# #saver.restore(sess, checkpoint_name)
# yhat_numpy = sess.run(yhat, feed_dict = {X : X, keep_prob: 1.0})
# print(yhat_numpy)
# #C:/Users/Karthick/Desktop/cvproject/src/tf_data/sample_model