def test(generator):
filenames = ["1.jpg", "2.jpg", "3.jpg"]
# orignal image for predict without mask
for filename in filenames:
img = process_image('test/' + filename)
## image for dreawing
temp_img = process_image('test/' + filename)
print("Testing ...")
mask = erase_img(temp_img)
img = np.expand_dims(img, 0)
mask = np.expand_dims(mask, 0)
completion_image = generator.predict([img, mask])
# # Delete Batch dimension
completion_image = np.squeeze(completion_image, 0)
img = np.squeeze(img, 0)
#cv2 show
#completion_image = cv2.cvtColor(completion_image, cv2.COLOR_BGR2RGB)
#img = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)
plt.figure(figsize=(6, 3))
plot_image(temp_img, 'Input', 1)
plot_image(completion_image, 'Output', 2)
plot_image(img, 'Ground Truth', 3)
plt.savefig("result/" + filename.split('.')[0] + "_test")
plt.show()
# cv2.imshow("result",completion_image)
# cv2.waitKey()
print("Done.....")
def predict(self, image_path, topk, cat_to_name):
""" Predict the class (or classes) of an image using a trained deep learning model. """
# --- Load image to get prediction ---------
img_np = util.process_image(image_path)
print("Getting prediction ... ", end="")
# --- Convert image to tensor for prediction ---------
img_tensor = torch.from_numpy(img_np).type(torch.FloatTensor)
img_tensor.unsqueeze_(0)
# --- Get probabilities ---------
self.model.eval()
with torch.no_grad():
img_variable = Variable(img_tensor)
log_ps = self.model(img_variable)
ps = torch.exp(log_ps)
top_ps, top_class = ps.topk(topk)
top_ps = top_ps.detach().numpy().tolist()[0]
top_class = top_class.detach().numpy().tolist()[0]
# --- Convert indices to classes and invert ---------
idx_to_class = {val: key for key, val in self.class_to_idx.items()}
top_labels = [idx_to_class[label] for label in top_class]
top_flowers = [cat_to_name[idx_to_class[label]] for label in top_class]
print("done!")
return top_ps, top_labels, top_flowers
def predict(image_path, model, topk=5, device='cpu'):
'''Predict the class (or classes) of an image using a trained deep learning model.
'''
print(f"Inferring using '{device}'...")
model.to(device)
# Retrieve file and covert to tensor
input_im = torch.tensor(process_image(image_path)).type(torch.FloatTensor)
# Retrieve label from input image path
label_class = Path(image_path).parts[2]
# Switch class mapping for convenience
idx_to_class = {d:k for k,d in model.class_to_idx.items()}
model.eval()
with active_session():
with torch.no_grad():
input_im = input_im.to(device)
logps = model(input_im.unsqueeze_(0))
ps = torch.exp(logps)
# Get top probabilities, indices
top_p, top_idx = ps.topk(topk, dim=1)
# Convert top indices to classes
top_class = [idx_to_class[k] for k in top_idx.tolist()[0]]
model.train()
input_im = input_im.squeeze().cpu()
# Return top probabilities, corresponding classes, input image tensor, and correct label class
return top_p.tolist()[0], top_class, input_im, label_class
def predict(self, image_path, topk=5, gpu=False):
"""
Predict the class (or classes) of an image using a trained deep learning model.
:param image_path:
:param topk: the number of classes to return
:param gpu: flag to use a GPU when predicting
:return: tuple containing [0]the list of probabilities and [1]the list of classes
"""
# Setup Cuda
device = torch.device(
"cuda:0" if gpu and torch.cuda.is_available() else "cpu")
self.model.to(device)
# Make sure model is in eval mode
self.model.eval()
# Process image into numpy image, then convert to torch tensor
np_image = process_image(image_path)
torch_image = torch.from_numpy(np_image)
torch_image = torch_image.to(device)
with torch.no_grad():
output = self.model(torch_image.unsqueeze_(0))
probabilities = torch.exp(output)
kprobs, kindex = probabilities.topk(topk)
kprobs_list = kprobs[0].cpu().numpy().tolist()
kindex_list = kindex[0].cpu().numpy().tolist()
# For every kindex value, look up the class and return it instead of the index
idx_to_class = {v: k for k, v in self.class_to_idx.items()}
class_list = [idx_to_class[idx] for idx in kindex_list]
return kprobs_list, class_list
def predict(self, image_path, top_k, cat_to_name_path):
''' Predict the class (or classes) of an image using a trained deep learning model.
'''
self.model.to(self.device)
im = Image.open(image_path)
im_tensor = util.process_image(im)
# print(im_tensor.shape)
im_tensor.unsqueeze_(0)
# print(im_tensor.shape)
# log_ps = self.model.forward(im_tensor.float())
im_tensor = im_tensor.to(self.device)
# get probabilities without gradients
with torch.no_grad():
log_ps = self.model.forward(im_tensor.float())
ps = torch.exp(log_ps)
# get top probabilities
top_probs, top_indexes = ps.topk(top_k)
# print(f'top_probs: {top_probs}')
# print(f'top_indexes: {top_indexes}')
# flatten
top_probs = top_probs.cpu().numpy().reshape(top_k,)
# print(f'top_probs: {top_probs}')
top_indexes = top_indexes.cpu().numpy().reshape(top_k,)
# print(f'top_indexes: {top_indexes}')
#print(f'model.class_to_idx: {model.class_to_idx}')
idx_to_class = {v: str(k) for k, v in self.class_to_idx.items()}
#print(f'idx_to_class: {idx_to_class}')
top_classes = []
for tc in top_indexes: top_classes.append(idx_to_class[tc])
#[idx_to_class[tc] for tc in top_indexes.cpu().numpy()]
# get class names
class_to_name = util.read_json_file(cat_to_name_path)
return top_probs, [class_to_name[tc] for tc in top_classes]
def predict(image_path, model, k=5):
image = Image.open(image_path)
image = np.asarray(image)
processed_image = process_image(image)
processed_image = processed_image[np.newaxis]
predictions_array = model.predict(processed_image)
predictions_array = np.squeeze(predictions_array)
top_k = np.argpartition(-predictions_array, k)[:k]
classes = []
for i in top_k:
classes.append(i)
return predictions_array[top_k], classes
def predict():
args = cli()
device = torch.device("cuda" if args.gpu else "cpu")
print(f'Device: {device}')
image = process_image(args.image_path)
model = load_model()
with open(args.category_names, 'r') as f:
cat_to_name = json.load(f)
top_ps, top_class = _predict(model, image, args.top_k, device)
for i, c in enumerate(top_class):
print(f"Prediction {i + 1}: "
f"{cat_to_name[c]} .. "
f"({100.0 * top_ps[i]:.3f}%)")
def predict(image_path, model, topk=5, device='cuda'):
''' Predict the class (or classes) of an image using a trained deep learning model.
'''
# : Implement the code to predict the class from an image file
im = Image.open(image_path)
processed_im = process_image(im).unsqueeze(0)
model.to(device)
model.eval()
with torch.no_grad():
processed_im = processed_im.to(device).float()
output = model(processed_im)
ps = torch.exp(output)
pred = ps.topk(topk)
flower_ids = pred[1][0].to('cpu')
flower_ids = torch.Tensor.numpy(flower_ids)
probs = pred[0][0].to('cpu')
idx_to_class = {k: v for v, k in checkpoint['class_to_idx'].items()}
flower_names = np.array([cat_to_name[idx_to_class[x]] for x in flower_ids])
return probs, flower_names
# Use GPU for inference:
gpu_enabled = results.gpu
with open('correction.json', 'r') as c:
correct_index = json.load(c)
with open(category_names, 'r') as f:
cat_to_name = json.load(f)
# Prediction
net = util.load_model(checkpoint)
device = torch.device(
"cuda" if torch.cuda.is_available() and gpu_enabled else "cpu")
im = Image.open(image)
im = util.process_image(im)
net.model = net.model.double()
im = im.to(device)
logps = net.model(im)
ps = torch.exp(logps)
top_p, top_class = ps.topk(top_k)
index = []
probabilities = []
for i in range(len(top_class[0])):
ind = str(top_class[0][i].item())
ind = correct_index[ind]
index.append(str(ind))
for j in range(len(top_p[0])):
parser.add_argument('--gpu', action='store_true', help='use gpu to infer classes')
parser.add_argument('--topk', action = 'store', dest = 'topk', type=int, default = 5, required = False, help = 'Return top K most likely classes')
parser.add_argument('--category_names', action='store', help='Label mapping file')
arguments = parser.parse_args()
try:
# Use GPU if it's available
#device = util.choose_device(arguments.gpu)
#loads a checkpoint and rebuilds the model
model = util.load_checkpoint(arguments.checkpoint_file)
model.eval()
#Image Preprocessing
img_file = random.choice(os.listdir(arguments.img_path))
image_path = arguments.img_path+img_file
img = util.process_image(image_path)
# Class Prediction
probs, classes = util.predict(image_path, model, arguments.gpu, arguments.topk)
# Sanity Checking
cat_to_name = util.cat_to_name(classes, model, arguments.category_names)
for i in range(len(cat_to_name)):
print(f"class = {cat_to_name[i]} prob = {probs.data[0][i]:.3f}")
#util.view_classify(image_path, probs, classes, cat_to_name)
except Exception as e:
logging.exception("Exception occurred")
from PIL import ImageGrab, Image
import deep_q_agent
agent = deep_q_agent.Agent()
HOST = 'localhost'
PORT = 8080
s = socket.socket(socket.AF_INET, socket.SOCK_STREAM)
s.bind((HOST, PORT))
s.listen()
conn, addr = s.accept()
print('Connected by', addr)
frame_count = 0
start_image = util.process_image(Image.open("images/start_image.png"))
start = ([start_image for _ in range(4)], 0
) # feed a list of blank images of a dummy checkpoint to start with
max_reward = 0
curr_reward = 0
eps_since_max = 0
ep = 0
images = []
images, agent.curr_checkpoint = start
last_act = np.array([[0, 0, 0, 0, 0]])
while True:
frame_count = (frame_count + 1) % 1000000
# Receive relevant data
data = conn.recv(1024)
'--cat_to_name_json',
type=str,
help='Json file to load for class values to name conversion',
default='cat_to_name.json')
args = parser.parse_args()
image_path = args.image_path
with open(args.cat_to_name_json, 'r') as f:
cat_to_name = json.load(f)
# : Load Checkpoint
model, checkpoint = load_checkpoint(args.checkpoint)
# : Process a PIL image for use in a PyTorch model
im = Image.open(image_path)
processed_im = process_image(im)
def predict(image_path, model, topk=5, device='cuda'):
''' Predict the class (or classes) of an image using a trained deep learning model.
'''
# : Implement the code to predict the class from an image file
im = Image.open(image_path)
processed_im = process_image(im).unsqueeze(0)
model.to(device)
model.eval()
with torch.no_grad():
processed_im = processed_im.to(device).float()
output = model(processed_im)
ps = torch.exp(output)
pred = ps.topk(topk)
#Load the checkpoint and rebuild the model
checkpoint = torch.load(args.checkpoint)
model = classifier_util.construct_model(checkpoint['arch'], checkpoint['hidden_units'])
#Load the trained weights
model.load_state_dict(checkpoint['model_state'])
model.class_to_idx = checkpoint['classtoidx']
#Select the device used for inference (either CPU or GPU/cuda)
device = 'cuda' if args.gpu else 'cpu'
model.to(device)
#Convert a PIL image into an object that can be used as input to a trained model
image = Image.open(args.input)
image_processed = torch.from_numpy(util.process_image(image)).float().to(device)
#Feed the input image to the model and calculate the output probabilities
model.eval()
logprob = model.forward(image_processed.resize_(1, 3, 224, 224))
prob = torch.exp(logprob)
#Categories with top k probabilities
top_p, top_index = prob.topk(args.top_k)
#Translate indices to categories
mapping = model.class_to_idx
inverse_mapping = {ind: cls for cls, ind in mapping.items()}
top_prob = top_p.detach().cpu().numpy().squeeze()
top_index_list = top_index.cpu().numpy().squeeze().tolist()
def build_model():
optimizer = Adadelta()
# build Completion Network model
org_img = Input(shape=input_shape, dtype='float32')
mask = Input(shape=(input_shape[0], input_shape[1], 1))
generator, completion_out = model_generator(org_img, mask)
completion_model = generator.compile(loss='mse', optimizer=optimizer)
# build Discriminator model
in_pts = Input(shape=(4, ), dtype='int32') # [y1,x1,y2,x2]
discriminator = model_discriminator(input_shape, local_shape)
d_container = Network(inputs=[org_img, in_pts],
outputs=discriminator([org_img, in_pts]))
d_out = d_container([org_img, in_pts])
d_model = Model([org_img, in_pts], d_out)
d_model.compile(loss='binary_crossentropy', optimizer=optimizer)
d_container.trainable = False
# build Discriminator & Completion Network models
all_model = Model([org_img, mask, in_pts], [completion_out, d_out])
all_model.compile(loss=['mse', 'binary_crossentropy'],
loss_weights=[1.0, alpha],
optimizer=optimizer)
X_train = filenames[:5000]
valid = np.ones((batch_size, 1)) ## label
fake = np.zeros((batch_size, 1)) ## label
for n in range(n_epoch):
progbar = generic_utils.Progbar(len(X_train))
for i in range(int(len(X_train) // batch_size)):
X_batch = X_train[i * batch_size:(i + 1) * batch_size]
inputs = np.array([
process_image(filename, input_shape[:2])
for filename in X_batch
])
points, masks = get_points(batch_size)
completion_image = generator.predict([inputs, masks])
g_loss = 0.0
d_loss = 0.0
if n < tc:
g_loss = generator.train_on_batch([inputs, masks], inputs)
else:
d_loss_real = d_model.train_on_batch([inputs, points], valid)
d_loss_fake = d_model.train_on_batch(
[completion_image, points], fake)
d_loss = 0.5 * np.add(d_loss_real, d_loss_fake)
if n >= tc + td:
g_loss = all_model.train_on_batch([inputs, masks, points],
[inputs, valid])
g_loss = g_loss[0] + alpha * g_loss[1]
progbar.add(inputs.shape[0],
values=[("Epoch", int(n + 1)), ("D loss", d_loss),
("G mse", g_loss)])
# show_image
show_image(batch_size, n, inputs, masks, completion_image)
# save model
generator.save("model/generator.h5")
discriminator.save("model/discriminator.h5")
