def readVideo(vid, clipDuration=64):
video = []
clip = []
i = 0
cap = cv2.VideoCapture(vid)
if (cap.isOpened() == False):
print("Error opening video stream or file")
while (cap.isOpened()):
ret, frame = cap.read()
if ret == True:
frame = preprocess_input(frame)
clip.append(frame)
i += 1
if (i == clipDuration):
video.append(clip)
clip = []
i = 0
else:
break
return np.asarray(video, dtype=np.float32)
Y_pred = model.predict(X)
Y_pred = np.argmax(Y_pred, axis=1)
Y_true = np.argmax(Y, axis=1)
cm = confusion_matrix(Y_true, Y_pred)
plt.figure(figsize=figsize)
ax = sns.heatmap(cm, cmap=cmap, annot=True, square=True)
ax.set_ylabel('Actual', fontsize=30)
ax.set_xlabel('Predicted', fontsize=30)
plt.show()
if __name__ == '__main__':
train_data = load_train()
X_train, Y_train = separate_train(train_data)
X_train, Y_train = preprocess_input(X_train, Y_train)
X_train, X_val, Y_train, Y_val = train_test_split(X_train, Y_train,
test_size=0.1,
random_state=SEED)
# train_model(X_train, X_val, Y_train, Y_val)
model = load_model('model.h5')
# To test new data, load the data, separate the features from the labels
# and preprocess the data. Then change X_val and Y_val to the desired data
final_loss, final_accuracy = model.evaluate(X_val, Y_val, verbose=0)
print('Final Loss: {:.4f}, Final Accuracy: {:.4f}'.format(
final_loss, final_accuracy))
plot_confusion_matrix(X_val, Y_val)
<<<<<<< HEAD
if not os.path.isdir(patch_dir):
=======
if not os.path.exists(patch_dir):
>>>>>>> c5fe2d2caa1e79289dfd72d25fa79a5f5915a4d5
os.makedirs(patch_dir)
if args.make_list:
f = open(os.path.join(patch_dir,'list.txt'), 'w')
# get all name of data
image_dir = os.path.join(image_dir, ' ')[0:-1]
for sample_file in tqdm(glob.glob(image_dir+'*-00128spp.exr')):
num = sample_file[len(image_dir):sample_file.index('-')]
gt_file = image_dir+'{}-08192spp.exr'.format(num)
if args.check_time:
prev_time = time.time()
data = preprocess_input(sample_file, gt_file)
patches = importanceSampling(data, patch_size, n_patches, figure_num)
cropped = list(crop(data, tuple(pos), patch_size) for pos in patches)
for i in range(len(cropped)):
file_name = '{}_{}.pt'.format(num, i)
torch.save(cropped[i], os.path.join(patch_dir, file_name))
if args.make_list:
f.write(file_name+'\n')
if args.check_time:
print('Time to get patches from one image', time.time() - prev_time)
if args.make_list:
f.close()
nn = test.shape[0]
if o > 0:
sub = 'result/%s_%s_' % (net, size) + time.strftime(
'%Y_%m_%d_%H_%M_%S', time.localtime(time.time()))
if os.path.exists(sub) == False:
os.makedirs(sub)
cost_time = []
for i in range(0, nn):
path = test.filepath[i]
fp = test.filename[i].split('.')[0]
t0 = time.time()
image = cv2.imread(path)
w, h = image.shape[:2]
image = cv2.resize(image, (size, size), cv2.INTER_CUBIC)
image = image[:, :, ::-1]
image = preprocess_input(image).reshape(1, size, size, 3)
pred = infer_model(image)
pred = pred.numpy().reshape(size, size)
print('==> ', i, fp, pred.mean())
pred[pred >= alpha] = 1.
pred[pred < alpha] = 0
pred = pred * 255.
pred = pred.astype(np.uint8)
#mask=cv2.erode(pred,(7,7))
mask = cv2.resize(pred, (h, w), cv2.INTER_CUBIC)
mask[mask > 0] = 255
cost_time.append(time.time() - t0)
cv2.imwrite(sub + '/' + fp + '_mask.jpg', mask)
del (mask, image, pred)
gc.collect()
print(' avg cost time :%.6f' % np.array(cost_time).mean())