def __init__(self, v1_model_path, v2_model_path, v3_model_path):
model = FrontNet(PreActBlock, [1, 1, 1])
ModelManager.Read(v1_model_path, model)
#state_dict = torch.load(v1_model_path, map_location='cpu')
rospy.loginfo(v1_model_path)
#model.load_state_dict(state_dict['model'])
#model.load_state_dict(torch.load(v1_model_path, map_location='cpu'))
self.trainer = ModelTrainer(model)
model2 = FrontNet(PreActBlock, [1, 1, 1])
#smodel2.load_state_dict(torch.load(v2_model_path, map_location='cpu'))
ModelManager.Read(v2_model_path, model2)
self.trainer2 = ModelTrainer(model2)
self.pose_pub = rospy.Publisher("predicted_pose", PoseStamped, queue_size=1)
def main():
logging.basicConfig(level=logging.INFO,
format="%(asctime)s - %(levelname)s - %(message)s",
datefmt="%Y-%m-%d %H:%M:%S",
filename="log.txt",
filemode='w')
console = logging.StreamHandler()
console.setLevel(logging.INFO)
formatter = logging.Formatter('%(message)s')
console.setFormatter(formatter)
logging.getLogger('').addHandler(console)
model = FrontNet(PreActBlock, [1, 1, 1])
ModelManager.Read('Models/FrontNetNicky.pt', model)
DATA_PATH = "/Users/usi/PycharmProjects/data/"
[x_test, y_test,
z_test] = DataProcessor.ProcessTestData(DATA_PATH + "TestNicky.pickle",
60, 108)
#x_test = x_test
#y_test = y_test
test_set = Dataset(x_test, y_test)
params = {'batch_size': 1, 'shuffle': False, 'num_workers': 0}
test_generator = data.DataLoader(test_set, **params)
trainer = ModelTrainer(model)
valid_loss_x, valid_loss_y, valid_loss_z, valid_loss_phi, outputs, gt_labels = trainer.ValidateSingleEpoch(
test_generator)
VizDroneBEV(x_test, y_test, z_test, outputs)
def main():
# Training settings
parser = argparse.ArgumentParser(description='PyTorch FrontNet')
args = Parse(parser)
torch.manual_seed(args.seed)
# [NeMO] Setup of console logging.
logging.basicConfig(level=logging.INFO,
format="%(asctime)s - %(levelname)s - %(message)s",
datefmt="%Y-%m-%d %H:%M:%S",
filename="log.txt",
filemode='w')
console = logging.StreamHandler()
console.setLevel(logging.INFO)
formatter = logging.Formatter('%(message)s')
console.setFormatter(formatter)
logging.getLogger('').addHandler(console)
train_loader, validation_loader, test_loader = LoadData(args)
# [NeMO] Loading of the JSON regime file.
regime = {}
if args.regime is None:
print("ERROR!!! Missing regime JSON.")
raise Exception
else:
with open(args.regime, "r") as f:
rr = json.load(f)
for k in rr.keys():
try:
regime[int(k)] = rr[k]
except ValueError:
regime[k] = rr[k]
if args.gray is not None:
model = Dronet(PreActBlock, [1, 1, 1], True)
else:
model = Dronet(PreActBlock, [1, 1, 1], False)
# [NeMO] This used to preload the model with pretrained weights.
if args.load_model is not None:
ModelManager.Read(args.load_model, model)
trainer = ModelTrainer(model, args, regime)
if args.quantize:
#logging.disable(logging.INFO)
trainer.Quantize(validation_loader)
#trainer.Train(train_loader, validation_loader)
#trainer.Predict(test_loader)
if args.save_model is not None:
ModelManager.Write(trainer.GetModel(), 100, args.save_model)
def TestInference():
#logging.disable(logging.INFO)
frame = cv2.imread("../Deployment/dataset/87.pgm", 0)
frame = np.reshape(frame, (60, 108, 1))
model = Dronet(PreActBlock, [1, 1, 1], True)
ModelManager.Read("Models/DronetHimax108x60.pt", model)
trainer = ModelTrainer(model)
v1_pred = trainer.InferSingleSample(frame)
print("output")
print(v1_pred)
def Test():
model = Dronet(PreActBlock, [1, 1, 1], True)
ModelManager.Read("Models/DronetHimax160x90.pt", model)
trainer = ModelTrainer(model)
#ModelManager.Read("Models/FrontNetGray.pt", model)
[x_test, y_test] = DataProcessor.ProcessTestData(
"/Users/usi/PycharmProjects/data/160x90HimaxStatic_12_03_20.pickle")
test_set = Dataset(x_test, y_test)
params = {'batch_size': 64, 'shuffle': False, 'num_workers': 1}
test_loader = data.DataLoader(test_set, **params)
trainer.Predict(test_loader)
def TestInference():
frame = cv2.imread("test13.pgm", 0)
frame = frame[92:152, 108:216]
frame = np.reshape(frame, (60, 108, 1))
#print(frame.flatten()[:10])
#cv2.imshow("", frame)
#cv2.waitKey()
model = Dronet(PreActBlock, [1, 1, 1], True)
ModelManager.Read("Models/DronetGray.pt", model)
# weight = model.conv.weight.data
# weight = np.reshape(weight, (-1))
# weight = weight[:10]
# print(weight)
trainer = ModelTrainer(model)
v1_pred = trainer.InferSingleSample(frame)
print("output")
print(v1_pred)
def main():
logging.basicConfig(level=logging.INFO,
format="%(asctime)s - %(levelname)s - %(message)s",
datefmt="%Y-%m-%d %H:%M:%S",
filename="log.txt",
filemode='w')
console = logging.StreamHandler()
console.setLevel(logging.INFO)
formatter = logging.Formatter('%(message)s')
console.setFormatter(formatter)
logging.getLogger('').addHandler(console)
model = Dronet(PreActBlock, [1, 1, 1], True)
ModelManager.Read('../PyTorch/Models/DronetGray.pt', model)
trainer = ModelTrainer(model)
InferenceData(trainer)
def main():
logging.basicConfig(level=logging.INFO,
format="%(asctime)s - %(levelname)s - %(message)s",
datefmt="%Y-%m-%d %H:%M:%S",
filename="log.txt",
filemode='w')
console = logging.StreamHandler()
console.setLevel(logging.INFO)
formatter = logging.Formatter('%(message)s')
console.setFormatter(formatter)
logging.getLogger('').addHandler(console)
DATA_PATH = "/Users/usi/PycharmProjects/data/160x90/"
# Get baseline results
picklename = "160x90HimaxMixedTest_12_03_20.pickle"
[x_test, y_test] = DataProcessor.ProcessTestData(DATA_PATH + picklename)
test_set = Dataset(x_test, y_test)
params = {'batch_size': 1, 'shuffle': False, 'num_workers': 1}
test_generator = data.DataLoader(test_set, **params)
model = Dronet(PreActBlock, [1, 1, 1], True)
ModelManager.Read('../PyTorch/Models/DronetHimax160x90AugCrop.pt', model)
trainer = ModelTrainer(model)
MSE2, MAE2, r2_score2, outputs2, gt_labels2 = trainer.Test(test_generator)
# Get pitch values
picklename = "160x90HimaxMixedTest_12_03_20Rot.pickle"
r_test = DataProcessor.GetRollFromTestData(DATA_PATH + picklename)
print(r_test)
if picklename.find(".pickle"):
picklename = picklename.replace(".pickle", '')
Plot2Models(r_test, picklename, r2_score2)
def Filter():
[x_test, y_test] = DataProcessor.ProcessTestData(
"/Users/usi/PycharmProjects/data/test_vignette4.pickle")
x_test2 = []
y_test2 = []
for i in range(len(x_test)):
gt = y_test[i]
if ((gt[0] > 1.0) and (gt[0] < 2.0)):
x_test2.append(x_test[i])
y_test2.append(y_test[i])
x_test2 = np.asarray(x_test2)
y_test2 = np.asarray(y_test2)
test_set = Dataset(x_test2, y_test2)
params = {'batch_size': 64, 'shuffle': False, 'num_workers': 0}
test_loader = data.DataLoader(test_set, **params)
model = Dronet(PreActBlock, [1, 1, 1], True)
ModelManager.Read("Models/DronetGray.pt", model)
trainer = ModelTrainer(model)
trainer.Predict(test_loader)
def main():
logging.basicConfig(level=logging.INFO,
format="%(asctime)s - %(levelname)s - %(message)s",
datefmt="%Y-%m-%d %H:%M:%S",
filename="log.txt",
filemode='w')
console = logging.StreamHandler()
console.setLevel(logging.INFO)
formatter = logging.Formatter('%(message)s')
console.setFormatter(formatter)
logging.getLogger('').addHandler(console)
DATA_PATH = "/Users/usi/PycharmProjects/data/160x90/"
[x_test, y_test, z_test] = DataProcessor.ProcessTestData(
DATA_PATH + "160x90HimaxMixedTest_12_03_20.pickle", True)
model = Dronet(PreActBlock, [1, 1, 1], True)
ModelManager.Read('../PyTorch/Models/DronetHimax160x90Augmented.pt', model)
trainer = ModelTrainer(model)
VizHeatMapsByAngle(z_test, x_test, y_test, trainer)
def main():
logging.basicConfig(level=logging.INFO,
format="%(asctime)s - %(levelname)s - %(message)s",
datefmt="%Y-%m-%d %H:%M:%S",
filename="log.txt",
filemode='w')
console = logging.StreamHandler()
console.setLevel(logging.INFO)
formatter = logging.Formatter('%(message)s')
console.setFormatter(formatter)
logging.getLogger('').addHandler(console)
model = Dronet(PreActBlock, [1, 1, 1], True)
ModelManager.Read('../PyTorch/Models/DronetHimax160x90.pt', model)
DATA_PATH = "/Users/usi/PycharmProjects/data/"
picklename = "160x90HimaxDynamic_12_03_20Shift.pickle"
[x_test, y_test, z_test] = DataProcessor.ProcessTestData(DATA_PATH + picklename, True)
test_set = Dataset(x_test, y_test)
params = {'batch_size': 1,
'shuffle': False,
'num_workers': 1}
test_generator = data.DataLoader(test_set, **params)
trainer = ModelTrainer(model)
MSE, MAE, r2_score, outputs, gt_labels = trainer.Test(test_generator)
h = x_test.shape[2]
w = x_test.shape[3]
x_test = np.reshape(x_test, (-1, h, w))
if picklename.find(".pickle"):
picklename = picklename.replace(".pickle", '')
VizWorldTopView(x_test, y_test, z_test, outputs, True, picklename)
def main():
# Training settings
parser = argparse.ArgumentParser(description='PyTorch FrontNet')
args = Parse(parser)
torch.manual_seed(args.seed)
# [NeMO] Setup of console logging.
logging.basicConfig(level=logging.INFO,
format="%(asctime)s - %(levelname)s - %(message)s",
datefmt="%Y-%m-%d %H:%M:%S",
filename="log.txt",
filemode='w')
console = logging.StreamHandler()
console.setLevel(logging.INFO)
formatter = logging.Formatter('%(message)s')
console.setFormatter(formatter)
logging.getLogger('').addHandler(console)
train_loader, validation_loader = LoadData(args)
# [NeMO] Loading of the JSON regime file.
regime = {}
if args.regime is None:
print("ERROR!!! Missing regime JSON.")
raise Exception
else:
with open(args.regime, "r") as f:
rr = json.load(f)
for k in rr.keys():
try:
regime[int(k)] = rr[k]
except ValueError:
regime[k] = rr[k]
model = HannaNet(ConvBlock, [1, 1, 1], True)
h = 96
w = 160
if args.trainq:
epoch = ModelManager.Read(args.load_model, model)
trainer = ModelTrainer(model, args, regime, teacher=True)
trainer.TrainQuantized(train_loader, validation_loader, h, w,
args.epochs)
if args.quantize and not args.trainq:
model = nemo.transform.quantize_pact(model,
dummy_input=torch.ones(
(1, 1, h, w)))
logging.info("[ETHQ2] Model: %s", model)
epoch, prec_dict = ModelManager.ReadQ(args.load_model, model)
trainer = ModelTrainer(model, args, regime)
trainer.Deploy(validation_loader, h, w, prec_dict)
if args.save_model is not None:
# torch.save(trainer.model.state_dict(), args.save_model)
ModelManager.Write(trainer.GetModel(), 100, args.save_model)
print(model)
def main():
logging.basicConfig(level=logging.INFO,
format="%(asctime)s - %(levelname)s - %(message)s",
datefmt="%Y-%m-%d %H:%M:%S",
filename="log.txt",
filemode='w')
console = logging.StreamHandler()
console.setLevel(logging.INFO)
formatter = logging.Formatter('%(message)s')
console.setFormatter(formatter)
logging.getLogger('').addHandler(console)
DATA_PATH = "/Users/usi/PycharmProjects/data/160x90/"
[x_test, y_test, z_test] = DataProcessor.ProcessTestData(DATA_PATH + "160x90HimaxMixedTrain_12_03_20.pickle", True)
model = Dronet(PreActBlock, [1, 1, 1], True)
ModelManager.Read('../PyTorch/Models/DronetHimax160x90Mixed.pt', model)
trainer = ModelTrainer(model)
df = pd.DataFrame({'x': z_test[:, 0], 'y': z_test[:, 1], 'z': z_test[:,2], 'phi': z_test[:, 3]})
walls = SortByAngle(df)
fig, a = plt.subplots(2, 2, figsize=(9,9))
for w in range(len(walls)):
zw_set = z_test[walls[w]]
xw_set = x_test[walls[w]]
yw_set = y_test[walls[w]]
df_w = pd.DataFrame({'x': zw_set[:, 0], 'y': zw_set[:, 1], 'z': zw_set[:, 2], 'phi': zw_set[:, 3]})
min = np.amin(zw_set, axis=0)
max = np.amax(zw_set, axis=0)
xcells = SortByCoordinates(df_w, 'x')
ycells = SortByCoordinates(df_w, 'y')
valid_loss = []
samples = []
for i in range(len(xcells)):
for j in range(len(ycells)):
cell = intersection(xcells[i], ycells[j])
if len(cell) > 0:
x = xw_set[cell]
y = yw_set[cell]
test_set = Dataset(x, y)
params = {'batch_size': 1, 'shuffle': False, 'num_workers': 0}
test_generator = data.DataLoader(test_set, **params)
MSE, MAE, r2_score, y_pred, gt_labels = trainer.Test(test_generator)
loss = MAE[1]
else:
loss = 0
valid_loss.append(loss)
samples.append(len(cell))
VizHeatMap(valid_loss, a, samples, len(xcells), len(ycells), w, min, max)
fig.tight_layout()
fig.suptitle('Y axis Error vs Pose')
plt.savefig("yheatmap.png")
plt.show()
def main():
logging.basicConfig(level=logging.INFO,
format="%(asctime)s - %(levelname)s - %(message)s",
datefmt="%Y-%m-%d %H:%M:%S",
filename="log.txt",
filemode='w')
console = logging.StreamHandler()
console.setLevel(logging.INFO)
formatter = logging.Formatter('%(message)s')
console.setFormatter(formatter)
logging.getLogger('').addHandler(console)
model = Dronet(PreActBlock, [1, 1, 1], True)
ModelManager.Read('../PyTorch/Models/DronetGray.pt', model)
DATA_PATH = "/Users/usi/PycharmProjects/data/"
picklename = "HimaxDynamic_12_03_20.pickle"
[x_test, y_test,
z_test] = DataProcessor.ProcessTestData(DATA_PATH + picklename, True)
t_test = DataProcessor.GetTimeStampsFromTestData(DATA_PATH + picklename)
if picklename.find(".pickle"):
picklename = picklename.replace(".pickle", '')
x_test2 = []
y_test2 = []
z_test2 = []
# for i in range(len(x_test)):
# gt = y_test[i]
# if ((gt[0] > 1.0) and (gt[0] < 2.0)):
# x_test2.append(x_test[i])
# y_test2.append(y_test[i])
# z_test2.append(z_test[i])
#
# x_test = np.asarray(x_test2)
# y_test = np.asarray(y_test2)
# z_test = np.asarray(z_test2)
test_set = Dataset(x_test, y_test)
params = {'batch_size': 1, 'shuffle': False, 'num_workers': 1}
test_generator = data.DataLoader(test_set, **params)
trainer = ModelTrainer(model)
MSE, MAE, r2_score, outputs, gt_labels = trainer.Test(test_generator)
# utils.SaveResultsToCSV(gt_labels, outputs, t_test, "wow.csv")
model2 = Dronet(PreActBlock, [1, 1, 1], True)
ModelManager.Read('../PyTorch/Models/DronetGrayAug120.pt', model2)
trainer2 = ModelTrainer(model2)
MSE, MAE, r2_score, outputs2, gt_labels2 = trainer2.Test(test_generator)
utils.SaveResultsToCSVWithCamPoses(gt_labels, outputs2, t_test, z_test,
picklename + ".csv")
outputs2 = np.reshape(outputs2, (-1, 4))
h = x_test.shape[2]
w = x_test.shape[3]
x_test = np.reshape(x_test, (-1, h, w))
VizWorldTopView(x_test, y_test, z_test, outputs, outputs2, True,
picklename)
