def evaluate(args):
test_data, test_label = load_data(
'experiment/data/modelnet40_ply_hdf5_2048/', train=False)
testDataset = ModelNetDataLoader(test_data, test_label)
testDataLoader = torch.utils.data.DataLoader(testDataset,
batch_size=args.batch_size,
shuffle=False)
log.debug('Building Model', args.model_name)
if args.model_name == 'pointnet':
num_class = 40
model = PointNetCls(num_class, args.feature_transform)
else:
model = PointNet2ClsMsg()
torch.backends.cudnn.benchmark = True
model = torch.nn.DataParallel(model).cuda()
log.debug('Using gpu:', args.gpu)
if args.pretrain is None:
log.err('No pretrain model')
return
log.debug('Loading pretrain model...')
state_dict = torch.load(args.pretrain)
model.load_state_dict(state_dict)
acc = test_clf(model.eval(), testDataLoader)
log.msg(Test_Accuracy='%.5f' % (acc))
def vis(args):
test_data, test_label = load_data(root, train=False)
log.info(test_data=test_data.shape, test_label=test_label.shape)
log.debug('Building Model', args.model_name)
if args.model_name == 'pointnet':
num_class = 40
model = PointNetCls(num_class, args.feature_transform).cuda()
else:
model = PointNet2ClsMsg().cuda()
torch.backends.cudnn.benchmark = True
model = torch.nn.DataParallel(model)
model.cuda()
log.info('Using multi GPU:', args.gpu)
if args.pretrain is None:
log.err('No pretrain model')
return
log.debug('Loading pretrain model...')
checkpoint = torch.load(args.pretrain)
model.load_state_dict(checkpoint)
model.eval()
log.info('Press space to exit, press Q for next frame')
for idx in range(test_data.shape[0]):
point_np = test_data[idx:idx + 1]
gt = test_label[idx][0]
points = torch.from_numpy(point_np)
points = points.transpose(2, 1).cuda()
pred, trans_feat = model(points)
pred_choice = pred.data.max(1)[1]
log.info(gt=class_names[gt],
pred_choice=class_names[pred_choice.cpu().numpy().item()])
point_cloud = open3d.geometry.PointCloud()
point_cloud.points = open3d.utility.Vector3dVector(point_np[0])
vis = open3d.visualization.VisualizerWithKeyCallback()
vis.create_window()
vis.get_render_option().background_color = np.asarray([0, 0, 0])
vis.add_geometry(point_cloud)
vis.register_key_callback(32, lambda vis: exit())
vis.run()
vis.destroy_window()
def main(args):
''' --- SELECT DEVICES --- '''
# Select either gpu or cpu
device = torch.device("cuda" if args.cuda else "cpu")
# Select among available GPUs
if args.cuda:
os.environ["CUDA_VISIBLE_DEVICES"] = ','.join(
str(x) for x in args.gpudevice)
''' --- CREATE EXPERIMENTS DIRECTORY AND LOGGERS IN TENSORBOARD --- '''
projdir = sys.path[0]
# Path for saving and loading the network.
saveloadpath = os.path.join(projdir, 'experiment\\checkpoints',
args.exp_name + '.pth')
Path(os.path.dirname(saveloadpath)).mkdir(exist_ok=True, parents=True)
# timestamp = str(datetime.datetime.now().strftime('%Y-%m-%d-%H-%M'))
tblogdir = os.path.join(projdir, 'experiment\\tensorboardX',
args.exp_name) # + '_' + timestamp )
Path(tblogdir).mkdir(exist_ok=True, parents=True)
# Create tb_writer(the writer will be used to write the information on tb) by using SummaryWriter,
# flush_secs defines how much seconds need to wait for writing information.
tb_writer = SummaryWriter(logdir=tblogdir,
flush_secs=3,
write_to_disk=True)
''' --- INIT DATASETS AND DATALOADER (FOR SINGLE EPOCH) --- '''
# Read data from file, and create training data and testing data which are both in multiple frames. Beware Ts is
# recording for every frame, i.e. every 82ms the automotive radar records once to form single frame(We need this information for LSTM).
train_dataset, test_dataset, class_names = read_dataset(
args.datapath, Ts=0.082, train_test_split=0.8)
# Prepare the traing and testing dataset. both trainDataset and testDataset are dataset have multiple frames data,
# for each frame it contains the "unified number of detection points"(NMAX detection points per frame).
# Init test dataset(Beware we should NOT use data augmentation for test dataset)
test_dataTransformations = transforms.Compose(
[NormalizeTime(), Resampling(maxPointsPerFrame=10)])
testDataset = RadarClassDataset(dataset=test_dataset,
transforms=test_dataTransformations,
sequence_length=1)
# Init train datasets
train_dataTransformations = transforms.Compose([
NormalizeTime(),
DataAugmentation(),
Resampling(maxPointsPerFrame=10)
])
trainDataset = RadarClassDataset(dataset=train_dataset,
transforms=train_dataTransformations,
sequence_length=1)
# Create dataloader for training by using batch_size frames' data in each batch
trainDataLoader = DataLoader(trainDataset,
batch_size=args.batchsize,
shuffle=True,
num_workers=args.num_workers)
''' --- INIT NETWORK MODEL --- '''
# Load selected network model and put it to right device
if args.model_name == 'pointnet':
classifier = PointNetCls(dim=args.pointCoordDim,
num_class=args.numclasses,
feature_transform=args.feature_transform)
elif args.model_name == 'pointnet2':
classifier = PointNet2ClsMsg(
dim=args.pointCoordDim,
num_class=args.numclasses,
)
else:
raise Exception(
'Argument "model_name" does not match existent networks')
classifier = classifier.to(device)
''' --- LOAD NETWORK IF EXISTS --- '''
if os.path.exists(saveloadpath):
print('Using pretrained model found...')
checkpoint = torch.load(saveloadpath)
start_epoch = checkpoint[
'epoch'] + 1 # Just becase make sure counting starts from 1, 2, ..., rather than 0, 1, ..., when print the information of start_epoch
iteration = checkpoint['iteration']
best_test_acc = checkpoint['test_accuracy']
classifier.load_state_dict(checkpoint['model_state_dict'])
else:
print('No existing model, starting training from scratch...')
start_epoch = 1 # Just becase make sure counting starts from 1, 2, ..., rather than 0, 1, ..., when print the information of start_epoch
iteration = 1 # Just becase make sure counting starts from 1, 2, ..., rather than 0, 1, ..., when print the information of iteration
best_test_acc = 0
''' --- CREATE OPTIMIZER ---'''
if args.optimizer == 'SGD':
optimizer = torch.optim.SGD(classifier.parameters(),
lr=args.lr,
momentum=0.9)
elif args.optimizer == 'ADAM':
optimizer = torch.optim.Adam(classifier.parameters(),
lr=args.lr,
betas=(0.9, 0.999),
eps=1e-08,
weight_decay=args.decay_rate)
scheduler = torch.optim.lr_scheduler.StepLR(
optimizer, step_size=args.lr_epoch_half,
gamma=0.5) # half(0.5) the learning rate every 'step_size' epochs
# log info
printparams = 'Model parameters:' + json.dumps(
vars(args), indent=4, sort_keys=True)
print(printparams)
tb_writer.add_text('hyper-parameters', printparams,
iteration) # tb_writer.add_hparam(args)
tb_writer.add_text(
'dataset', 'dataset sample size: training: {}, test: {}'.format(
train_dataset.shape[0], test_dataset.shape[0]), iteration)
''' --- START TRANING ---'''
for epoch in range(start_epoch, args.epoch + 1):
print('Epoch %d/%s:' % (epoch, args.epoch))
# Add the "learning rate" into tensorboard scalar which will be shown in tensorboard
tb_writer.add_scalar('learning_rate', optimizer.param_groups[0]['lr'],
iteration)
for batch_id, data in tqdm(enumerate(trainDataLoader, 0),
total=len(trainDataLoader),
smoothing=0.9):
points, target = data # (B:batch x S:seq x C:features x N:points) , (B x S:seq)
# Squeeze to drop Sequence dimension, which is equal to 1, convert all the data to float(otherwise there will be data type problems when running the model) and move to device
points, target = points.squeeze(
dim=1).float().to(device), target.float().to(
device) # (B:batch x C:features x N:points) , (B)
# points, target = points.float().to(device), target.float().to(device)
# Reset gradients
optimizer.zero_grad()
# Sets the module in training mode
classifier = classifier.train()
# Forward propagation
pred = classifier(points)
# Calculate cross entropy loss (In the pointnet/pointnet2 network model, it outputs log_softmax result. Since
# "log_softmax -> nll_loss" == CrossEntropyLoss, so that we just need to call F.nll_loss)
loss = F.nll_loss(pred, target.long())
if args.model_name == 'pointnet':
loss += feature_transform_regularizer(classifier.trans) * 0.001
if args.feature_transform:
loss += feature_transform_regularizer(
classifier.trans_feat) * 0.001
# Back propagate
loss.backward()
# Update weights
optimizer.step()
# Log once for every 5 batches, add the "train_loss/cross_entropy" into tensorboard scalar which will be shown in tensorboard
if not batch_id % 5:
tb_writer.add_scalar('train_loss/cross_entropy', loss.item(),
iteration)
iteration += 1
# if batch_id> 2: break
scheduler.step()
''' --- TEST AND SAVE NETWORK --- '''
if not epoch % 10: # Doing the following things every epoch.
# Perform predictions on the training data.
train_targ, train_pred = test(classifier,
trainDataset,
device,
num_workers=args.num_workers,
batch_size=1800)
# Perform predictions on the testing data.
test_targ, test_pred = test(classifier,
testDataset,
device,
num_workers=args.num_workers,
batch_size=1800)
# Calculate the accuracy rate for training data.
train_acc = metrics_accuracy(train_targ, train_pred)
# Calculate the accuracy rate for testing data.
test_acc = metrics_accuracy(test_targ, test_pred)
print('\r Training loss: {}'.format(loss.item()))
print('Train Accuracy: {}\nTest Accuracy: {}'.format(
train_acc, test_acc))
# Add the "train_acc" "test_acc" into tensorboard scalars which will be shown in tensorboard.
tb_writer.add_scalars('metrics/accuracy', {
'train': train_acc,
'test': test_acc
}, iteration)
# Calculate confusion matrix.
confmatrix_test = metrics_confusion_matrix(test_targ, test_pred)
print('Test confusion matrix: \n', confmatrix_test)
# Log confusion matrix.
fig, ax = plot_confusion_matrix(confmatrix_test,
class_names,
normalize=False,
title='Test Confusion Matrix')
# Log normalized confusion matrix.
fig_n, ax_n = plot_confusion_matrix(
confmatrix_test,
class_names,
normalize=True,
title='Test Confusion Matrix - Normalized')
# Add the "confusion matrix" "normalized confusion matrix" into tensorboard figure which will be shown in tensorboard.
tb_writer.add_figure('test_confusion_matrix/abs',
fig,
global_step=iteration,
close=True)
tb_writer.add_figure('test_confusion_matrix/norm',
fig_n,
global_step=iteration,
close=True)
# Log precision recall curves.
for idx, clsname in enumerate(class_names):
# Convert log_softmax to softmax(which is actual probability) and select the desired class.
test_pred_binary = torch.exp(test_pred[:, idx])
test_targ_binary = test_targ.eq(idx)
# Add the "precision recall curves" which will be shown in tensorboard.
tb_writer.add_pr_curve(tag='pr_curves/' + clsname,
labels=test_targ_binary,
predictions=test_pred_binary,
global_step=iteration)
''' --- SAVE NETWORK --- '''
# if (test_acc >= best_test_acc): # For now lets save every time, since we are only testing in a subset of the test dataset
best_test_acc = test_acc # if test_acc > best_test_acc else best_test_acc
state = {
'epoch': epoch,
'iteration': iteration,
'train_accuracy': train_acc if args.train_metric else 0.0,
'test_accuracy': best_test_acc,
'model_state_dict': classifier.state_dict(),
'optimizer_state_dict': optimizer.state_dict(),
}
torch.save(state, saveloadpath)
print('Model saved!!!')
print('Best Accuracy: %f' % best_test_acc)
tb_writer.close()
worker_init_fn=worker_init_fn,
collate_fn=my_collate,
)
is_resume = 0
if args.load_model and args.load_epoch != -1:
is_resume = 1
if is_resume or args.mode == 'test':
model = torch.load(args.load_model,
map_location='cuda:{}'.format(args.gpu))
model.device_ids = [args.gpu]
print('load model {}'.format(args.load_model))
else:
model = PointNetCls(num_points=grasp_points_num,
input_chann=point_channel,
k=2)
if args.cuda:
if args.gpu != -1:
torch.cuda.set_device(args.gpu)
model = model.cuda()
else:
device_id = [0, 1, 2, 3]
torch.cuda.set_device(device_id[0])
model = nn.DataParallel(model, device_ids=device_id).cuda()
optimizer = optim.Adam(model.parameters(), lr=args.lr)
scheduler = StepLR(optimizer, step_size=30, gamma=0.5)
def train(model, loader, epoch):
scheduler.step()
import torch
import torch.utils.data
# import torch.nn as nn
# import torch.nn.functional as F
import numpy as np
import sys
from os import path
from model.pointnet import PointNetCls
import torch.multiprocessing as mp
sys.path.append(path.dirname(path.dirname(path.abspath("__file__"))))
# torch.cuda.manual_seed(1) # don't delete
grasp_points_num = 1024
model = PointNetCls(num_points=grasp_points_num, input_chann=3, k=2)
model.cuda()
model.eval()
torch.set_grad_enabled(False)
def load_weight(weights_path):
print('[Python] Load weight: {}'.format(weights_path))
model.load_state_dict(torch.load(weights_path))
def classify_pcs(local_pcs, output_cls=0):
""" Classify point clouds FPS:650 """
print("[Python] Classify point clouds...")
def main():
''' --- SELECT DEVICES --- '''
# Select either gpu or cpu
device = torch.device("cuda" if args.cuda else "cpu")
# Select among available GPUs
if args.cuda:
os.environ["CUDA_VISIBLE_DEVICES"] = ','.join(
str(x) for x in args.gpudevice)
''' --- INIT NETWORK MODEL --- '''
projdir = sys.path[0]
# Path for saving and loading the network.
saveloadpath = os.path.join(projdir, 'experiment\\checkpoints',
args.exp_name + '.pth')
Path(os.path.dirname(saveloadpath)).mkdir(exist_ok=True, parents=True)
# Load selected network model and put it to right device
if args.model_name == 'pointnet':
classifier = PointNetCls(dim=args.pointCoordDim,
num_class=len(args.categories),
feature_transform=args.feature_transform)
elif args.model_name == 'pointnet2':
classifier = PointNet2ClsMsg(dim=args.pointCoordDim,
num_class=len(args.categories))
else:
raise Exception(
'Argument "model_name" does not match existent networks')
classifier = classifier.to(device)
''' --- LOAD NETWORK IF EXISTS --- '''
if os.path.exists(saveloadpath):
print('Using pretrained model found...')
checkpoint = torch.load(saveloadpath)
start_epoch = checkpoint[
'epoch'] + 1 # Just becase make sure counting starts from 1, 2, ..., rather than 0, 1, ..., when print the information of start_epoch
iteration = checkpoint['iteration']
best_test_acc = checkpoint['test_accuracy']
classifier.load_state_dict(checkpoint['model_state_dict'])
else:
raise Exception('Model in: {} does not exists'.format(saveloadpath))
# put classifier in evaluation mode
classifier = classifier.eval()
''' --- INIT DATASETS AND DATALOADER (FOR SINGLE EPOCH) --- '''
# Ideal for PointNet and pointLSTM - dataloader will return (B:batch, S:seq, C:features, N:points)
dataTransformations = transforms.Compose([
ToSeries(),
DataAugmentation(),
Resampling(maxPointsPerFrame=10),
ToTensor()
])
# init dataset
nusc = NuScenes(version=args.nuscenes_eval_dir,
dataroot=args.nuscenes_dir,
verbose=True)
''' Iterate over samples '''
idx = list(range(len(nusc.sample)))
random.shuffle(idx) # shuffle samples
# for sample_rec in nusc.sample:
for i in idx:
sample_rec = nusc.sample[i]
sample_token = sample_rec['token']
# read all sensors data and merge them to a common reference frame
sensorlist = []
for sensor in args.sensors:
sensorlist.append(
MyRadarPointCloud.from_sample_token(nusc, sample_token,
sensor))
global radar_pc
radar_pc = MyRadarPointCloud.merge(sensorlist, 0)
# get annotations for actual sample
ann_tokens = nusc.get('sample', sample_token)['anns']
# filter annotations that are objects of interest for our classifier
def get_ann_properties(nusc, ann_token: str):
ann_rec = nusc.get('sample_annotation', ann_token)
category = ann_rec['category_name']
attr_tokens = ann_rec['attribute_tokens']
attr = [
nusc.get('attribute', attr_token)['name']
for attr_token in attr_tokens
]
if len(attr) == 0: attr = ['']
print('object category:{:30} attr:{:40} center:{:20}'.format(
category, str(attr),
str(ann_rec['translation'] - radar_pc.A_cs_2_gl[:3, 3])))
return category, attr[
0] # for now lets return only the first attribute (we might have more per object)
def isInterestingObject(ann_token):
cat, att = get_ann_properties(nusc, ann_token)
# category-attribute pair must match at least one of the desired category-attr pair
for desired_cat, desired_att in zip(args.categories,
args.attributes):
if cat in desired_cat and att in desired_att:
return True
return False
print('\n\nGround-truth objects on this frame:')
ann_tokens = list(filter(isInterestingObject, ann_tokens))
def getObjLabel(ann_token):
cat, att = get_ann_properties(nusc, ann_token)
for idx, (desired_cat, desired_att) in enumerate(
zip(args.categories, args.attributes)):
if cat in desired_cat and att in desired_att:
return idx
return np.NaN
labels = list(map(getObjLabel, ann_tokens))
assert np.all(
~np.isnan(labels)
), 'Something strange happened... object was selected as interesting but we can not find its label'
# create bounding boxes from annotations
ann_boxes = []
for ann_token, label in zip(ann_tokens, labels):
box = radar_pc.box(ann_token)
box.label = label
box.name = args.objectNames[label]
ann_boxes.append(box)
''' APPLY DBSCAN ON EACH SCENE'''
from sklearn.cluster import DBSCAN
points_scene = radar_pc.points # points_gl = <5,N>
# filter out objects, whose speed is less than 0.3 m/s
idx_moving = np.linalg.norm(points_scene[2:4, :].T, axis=1) > 0.25
points_scene = points_scene[:, idx_moving]
# apply DBSCAN
clustering = DBSCAN(eps=3, min_samples=1).fit(points_scene[:2, :].T)
''' For each cluster, run network and predict class '''
pred_results = []
for cluster_idx in range(max(clustering.labels_)):
# select points from the current cluster
points_idx = clustering.labels_ == cluster_idx
points_obj = points_scene[:, points_idx]
# apply necessary transformations
features = dataTransformations(points_obj)
# convert to torch.tensor
features = torch.tensor(features).type(
torch.FloatTensor).unsqueeze(0).to(device)
# calculate network prediction
pred = classifier(features).argmax().item()
# store result
pred_results.append((points_obj, pred))
''' RENDER '''
OBJCOLORS = ['red', 'green', 'blue', 'grey', 'orange', 'white']
fig = plt.figure(constrained_layout=True)
gs = GridSpec(2, 6, figure=fig)
gs.update(wspace=0, hspace=0)
axs = [
fig.add_subplot(gs[1, :3]),
fig.add_subplot(gs[1, 3:]),
fig.add_subplot(gs[0, 0]),
fig.add_subplot(gs[0, 1]),
fig.add_subplot(gs[0, 2]),
fig.add_subplot(gs[0, 3]),
fig.add_subplot(gs[0, 4]),
fig.add_subplot(gs[0, 5])
]
# render annotation boxes
for box in ann_boxes:
color = OBJCOLORS[box.label]
box.render(axs[0], colors=(color, ) * 3)
box.render(axs[1], colors=(color, ) * 3)
# render cars
radar_pc.car.render(axs[0], colors=('orange', 'k', 'k'))
radar_pc.car.render(axs[1], colors=('orange', 'k', 'k'))
# render pointcloud
radar_pc._render_pc(axs[0], radar_pc.points, color_channel='k')
for cluster_points, pred in pred_results:
# radar_pc._render_pc( axs[1], cluster_points, color_channel=OBJCOLORS[pred])
if pred != 5:
axs[1].scatter(cluster_points[0, :],
cluster_points[1, :],
s=30,
c=OBJCOLORS[pred],
edgecolors='k',
linewidths=1,
zorder=100)
# plot ellipses
if cluster_points.shape[1] > 1:
confidence_ellipse(axs[1],
cluster_points[0, :],
cluster_points[1, :],
edgecolor=OBJCOLORS[pred],
facecolor=OBJCOLORS[pred],
alpha=0.5)
# render camera images
cameras = [
'CAM_BACK_LEFT', 'CAM_FRONT_LEFT', 'CAM_FRONT', 'CAM_FRONT_RIGHT',
'CAM_BACK_RIGHT', 'CAM_BACK'
]
for ax_idx, cam_name in enumerate(cameras):
cam_token = sample_rec['data'][cam_name]
cam_rec = nusc.get('sample_data', cam_token)
img_filename = os.path.join(nusc.dataroot, cam_rec['filename'])
img = mpimg.imread(img_filename)
# axs[ax_idx].set_axis_off()
axs[ax_idx + 2].get_xaxis().set_visible(False)
axs[ax_idx + 2].get_yaxis().set_visible(False)
axs[ax_idx + 2].imshow(img)
axs[ax_idx + 2].set_title(cam_name)
# format
legend_elements = [
Line2D([0], [0], color=color, lw=4, label=name)
for color, name in zip(OBJCOLORS, args.objectNames)
]
axs[0].legend(handles=legend_elements,
loc='upper right',
prop={'size': 8})
axs[1].legend(handles=legend_elements,
loc='upper right',
prop={'size': 8})
axs[0].axis('equal')
axs[1].axis('equal')
mng = plt.get_current_fig_manager()
mng.window.showMaximized()
plt.tight_layout()
lims = (-60, 60)
axs[0].set_xlim(lims)
axs[0].set_ylim(lims)
axs[0].set_title(
'Radar detection points and ground truth bounding boxes of MOVING objects'
)
axs[1].set_xlim(lims)
axs[1].set_ylim(lims)
axs[1].set_title(
'Segmented segmentation of detection points (DBSCAN + PointNet classifier)\n3-sigma ellipses for each cluster and ground-truth bbs'
)
plt.tight_layout()
plt.show()
# plot scene with the standard Nuscenes method
# nusc.render_sample(sample_token)
# plt.show()
# sample_token = nusc.get('sample', sample_token)['next']
tb_writer.close()
dataloader = torch.utils.data.DataLoader(
dataset,
batch_size=args.batch_size,
num_workers=16,
pin_memory=True,
shuffle=True,
worker_init_fn=worker_init_fn,
collate_fn=my_collate,
)
is_resume = 0
if args.load_model and args.load_epoch != -1:
is_resume = 1
model = PointNetCls(num_points=grasp_points_num,
input_chann=point_channel,
k=2)
if is_resume or args.mode == 'test':
model.load_state_dict(torch.load(args.load_model))
print('load weights {}'.format(args.load_model))
if args.cuda:
base_model = model.cuda()
model = nn.DataParallel(base_model)
optimizer = optim.Adam(model.parameters(), lr=args.lr)
scheduler = StepLR(optimizer, step_size=30, gamma=0.5)
def train(model, dataset, loader, epoch):
scheduler.step()
dataset.train()
model.train()
def main(args):
'''HYPER PARAMETER'''
os.environ["CUDA_VISIBLE_DEVICES"] = args.gpu
# datapath = './data/ModelNet/'
datapath = './data/objecnn20_data_hdf5_2048/'
if args.rotation is not None:
ROTATION = (int(args.rotation[0:2]),int(args.rotation[3:5]))
else:
ROTATION = None
'''CREATE DIR'''
experiment_dir = Path('./experiment/')
experiment_dir.mkdir(exist_ok=True)
file_dir = Path(str(experiment_dir) +'/%sObjectNNClf-'%args.model_name+ str(datetime.datetime.now().strftime('%Y-%m-%d_%H-%M')))
file_dir.mkdir(exist_ok=True)
checkpoints_dir = file_dir.joinpath('checkpoints/')
checkpoints_dir.mkdir(exist_ok=True)
log_dir = file_dir.joinpath('logs/')
log_dir.mkdir(exist_ok=True)
'''LOG'''
args = parse_args()
logger = logging.getLogger(args.model_name)
logger.setLevel(logging.INFO)
formatter = logging.Formatter('%(asctime)s - %(name)s - %(levelname)s - %(message)s')
file_handler = logging.FileHandler(str(log_dir) + '/train_%s_ObjectNNClf.txt'%args.model_name)
file_handler.setLevel(logging.INFO)
file_handler.setFormatter(formatter)
logger.addHandler(file_handler)
logger.info('---------------------------------------------------TRANING---------------------------------------------------')
logger.info('PARAMETER ...')
logger.info(args)
'''DATA LOADING'''
logger.info('Load dataset ...')
train_data, train_label, test_data, test_label = load_data(datapath, classification=True)
logger.info("The number of training data is: %d",train_data.shape[0])
logger.info("The number of test data is: %d", test_data.shape[0])
trainDataset = ObjectNNDataLoader(train_data, train_label, rotation=ROTATION)
if ROTATION is not None:
print('The range of training rotation is',ROTATION)
testDataset = ObjectNNDataLoader(test_data, test_label, rotation=ROTATION)
trainDataLoader = torch.utils.data.DataLoader(trainDataset, batch_size=args.batchsize, shuffle=True)
testDataLoader = torch.utils.data.DataLoader(testDataset, batch_size=args.batchsize, shuffle=False)
'''MODEL LOADING'''
num_class = 20
classifier = PointNetCls(num_class,args.feature_transform).cuda() if args.model_name == 'pointnet' else PointNet2ClsMsg().cuda()
if args.pretrain is not None:
print('Use pretrain model...')
logger.info('Use pretrain model')
checkpoint = torch.load(args.pretrain)
start_epoch = checkpoint['epoch']
# classifier.load_state_dict(checkpoint['model_state_dict'])
# print(checkpoint['model_state_dict'])
model_dict = classifier.state_dict()
pretrained_dict = {k: v for k, v in checkpoint['model_state_dict'].items() if k in model_dict}
model_dict.update(pretrained_dict)
classifier.load_state_dict(model_dict)
else:
print('No existing model, starting training from scratch...')
start_epoch = 0
if args.optimizer == 'SGD':
optimizer = torch.optim.SGD(classifier.parameters(), lr=0.01, momentum=0.9)
elif args.optimizer == 'Adam':
optimizer = torch.optim.Adam(
classifier.parameters(),
lr=args.learning_rate,
betas=(0.9, 0.999),
eps=1e-08,
weight_decay=args.decay_rate
)
scheduler = torch.optim.lr_scheduler.StepLR(optimizer, step_size=20, gamma=0.5) # 调整学习率
global_epoch = 0
global_step = 0
best_tst_accuracy = 0.0
blue = lambda x: '\033[94m' + x + '\033[0m'
'''TRANING'''
logger.info('Start training...')
for epoch in range(start_epoch,args.epoch):
print('Epoch %d (%d/%s):' % (global_epoch + 1, epoch + 1, args.epoch))
logger.info('Epoch %d (%d/%s):' ,global_epoch + 1, epoch + 1, args.epoch)
scheduler.step()
for batch_id, data in tqdm(enumerate(trainDataLoader, 0), total=len(trainDataLoader), smoothing=0.9):
points, target = data
target = target[:, 0]
points = points.transpose(2, 1)
points, target = points.cuda(), target.cuda()
optimizer.zero_grad()
classifier = classifier.train()
pred, trans_feat, global_feature = classifier(points)
loss = F.nll_loss(pred, target.long())
if args.feature_transform and args.model_name == 'pointnet':
loss += feature_transform_reguliarzer(trans_feat) * 0.001
loss.backward()
optimizer.step()
global_step += 1
train_acc = test(classifier.eval(), trainDataLoader) if args.train_metric else None
acc = test(classifier, testDataLoader)
print('\r Loss: %f' % loss.data)
logger.info('Loss: %.2f', loss.data)
if args.train_metric:
print('Train Accuracy: %f' % train_acc)
logger.info('Train Accuracy: %f', (train_acc))
print('\r Test %s: %f' % (blue('Accuracy'),acc))
logger.info('Test Accuracy: %f', acc)
if (acc >= best_tst_accuracy) and epoch > 5:
best_tst_accuracy = acc
logger.info('Save model...')
save_checkpoint(
global_epoch + 1,
train_acc if args.train_metric else 0.0,
acc,
classifier,
optimizer,
str(checkpoints_dir),
args.model_name)
print('Saving model....')
global_epoch += 1
print('Best Accuracy: %f'%best_tst_accuracy)
logger.info('End of training...')
def main(testDataset):
args = parse_args()
'''HYPER PARAMETER'''
# os.environ["CUDA_VISIBLE_DEVICES"] = args.gpu
os.environ["CUDA_VISIBLE_DEVICES"] = args.gpu if args.multi_gpu is None else '0,1,2,3'
datapath = './data/ModelNet/'
if args.rotation is not None:
ROTATION = (int(args.rotation[0:2]),int(args.rotation[3:5]))
else:
ROTATION = None
'''CREATE DIR'''
experiment_dir = Path('./experiment/')
experiment_dir.mkdir(exist_ok=True)
checkpoints_dir = Path('./experiment/checkpoints/')
checkpoints_dir.mkdir(exist_ok=True)
log_dir = Path('./experiment/logs/')
log_dir.mkdir(exist_ok=True)
'''LOG'''
args = parse_args()
logger = logging.getLogger("PointNet2")
logger.setLevel(logging.INFO)
formatter = logging.Formatter('%(asctime)s - %(name)s - %(levelname)s - %(message)s')
file_handler = logging.FileHandler('./experiment/logs/train_%s_'%args.model_name+ str(datetime.datetime.now().strftime('%Y-%m-%d %H-%M'))+'.txt')
file_handler.setLevel(logging.INFO)
file_handler.setFormatter(formatter)
logger.addHandler(file_handler)
logger.info('---------------------------------------------------TRANING---------------------------------------------------')
logger.info('PARAMETER ...')
logger.info(args)
'''DATA LOADING'''
logger.info('Load dataset ...')
# train_data, train_label, test_data, test_label = load_data(datapath, classification=True)
# logger.info("The number of training data is: %d",train_data.shape[0])
# logger.info("The number of test data is: %d", test_data.shape[0])
# trainDataset = ModelNetDataLoader(train_data, train_label, rotation=ROTATION)
#if ROTATION is not None:
# print('The range of training rotation is',ROTATION)
# testDataset = ModelNetDataLoader(test_data, test_label, rotation=ROTATION)
# trainDataLoader = torch.utils.data.DataLoader(trainDataset, batch_size=args.batchsize, shuffle=True)
testDataLoader = torch.utils.data.DataLoader(testDataset, batch_size=args.batchsize, shuffle=False)
'''MODEL LOADING'''
num_class = 40
classifier = PointNetCls(num_class,args.feature_transform).cuda() if args.model_name == 'pointnet' else PointNet2ClsMsg().cuda()
'''GPU selection and multi-GPU'''
if args.multi_gpu is not None:
device_ids = [int(x) for x in args.multi_gpu.split(',')]
torch.backends.cudnn.benchmark = True
classifier.cuda(device_ids[0])
classifier = torch.nn.DataParallel(classifier, device_ids=device_ids)
else:
classifier.cuda()
''' 使用预训练模型 '''
if args.pretrain is not None:
print('Use pretrain model...')
logger.info('Use pretrain model')
checkpoint = torch.load(args.pretrain)
start_epoch = checkpoint['epoch']
classifier.load_state_dict(checkpoint['model_state_dict'])
else:
print('No existing model, starting training from scratch...')
start_epoch = 0
if args.optimizer == 'SGD':
optimizer = torch.optim.SGD(classifier.parameters(), lr=0.01, momentum=0.9)
elif args.optimizer == 'Adam':
optimizer = torch.optim.Adam(
classifier.parameters(),
lr=args.learning_rate,
betas=(0.9, 0.999),
eps=1e-08,
weight_decay=args.decay_rate
)
scheduler = torch.optim.lr_scheduler.StepLR(optimizer, step_size=20, gamma=0.5)
global_epoch = 0
global_step = 0
best_tst_accuracy = 0.0
blue = lambda x: '\033[94m' + x + '\033[0m'
start_epoch = 0
args.epoch = 1
'''TRANING'''
logger.info('Start training...')
for epoch in range(start_epoch,args.epoch):
print('Epoch %d (%d/%s):' % (global_epoch + 1, epoch + 1, args.epoch))
logger.info('Epoch %d (%d/%s):' ,global_epoch + 1, epoch + 1, args.epoch)
# scheduler.step()
# for batch_id, data in tqdm(enumerate(trainDataLoader, 0), total=len(trainDataLoader), smoothing=0.9):
# points, target = data
# target = target[:, 0]
# points = points.transpose(2, 1)
# points, target = points.cuda(), target.cuda()
# optimizer.zero_grad()
# classifier = classifier.train()
# pred, trans_feat = classifier(points)
# loss = F.nll_loss(pred, target.long())
# if args.feature_transform and args.model_name == 'pointnet':
# loss += feature_transform_reguliarzer(trans_feat) * 0.001
# loss.backward()
# optimizer.step()
# global_step += 1
train_acc = test(classifier.eval(), trainDataLoader) if args.train_metric else None
acc, fts = test(classifier, testDataLoader)
# return fts
# print('\r Loss: %f' % loss.data)
# logger.info('Loss: %.2f', loss.data)
if args.train_metric:
print('Train Accuracy: %f' % train_acc)
logger.info('Train Accuracy: %f', (train_acc))
print('\r Test %s: %f' % (blue('Accuracy'),acc))
logger.info('Test Accuracy: %f', acc)
return fts
if (acc >= best_tst_accuracy) and epoch > 5:
best_tst_accuracy = acc
logger.info('Save model...')
save_checkpoint(
global_epoch + 1,
train_acc if args.train_metric else 0.0,
acc,
classifier,
optimizer,
str(checkpoints_dir),
args.model_name)
print('Saving model....')
global_epoch += 1
print('Best Accuracy: %f'%best_tst_accuracy)
logger.info('End of training...')
def main():
train_loader = torch.utils.data.DataLoader(
PointGraspOneViewDataset(
grasp_points_num=grasp_points_num,
path=args.data_path,
tag='train',
grasp_amount_per_file=6500,
thresh_good=thresh_good,
thresh_bad=thresh_bad,
),
batch_size=args.batch_size,
num_workers=32,
pin_memory=True,
shuffle=True,
worker_init_fn=worker_init_fn,
collate_fn=my_collate,
)
test_loader = torch.utils.data.DataLoader(
PointGraspOneViewDataset(
grasp_points_num=grasp_points_num,
path=args.data_path,
tag='test',
grasp_amount_per_file=500,
thresh_good=thresh_good,
thresh_bad=thresh_bad,
with_obj=True,
),
batch_size=args.batch_size,
num_workers=32,
pin_memory=True,
shuffle=True,
worker_init_fn=worker_init_fn,
collate_fn=my_collate,
)
is_resume = 0
if args.load_model and args.load_epoch != -1:
is_resume = 1
if is_resume or args.mode == 'test':
model = torch.load(args.load_model,
map_location='cuda:{}'.format(args.gpu))
model.device_ids = [args.gpu]
print('load model {}'.format(args.load_model))
else:
model = PointNetCls(num_points=grasp_points_num,
input_chann=point_channel,
k=2)
if args.cuda:
if args.gpu != -1:
torch.cuda.set_device(args.gpu)
model = model.cuda()
else:
device_id = [0, 1, 2, 3]
torch.cuda.set_device(device_id[0])
model = nn.DataParallel(model, device_ids=device_id).cuda()
if args.mode == 'train':
for epoch in range(is_resume * args.load_epoch, args.epoch):
acc_train = train(model, train_loader, epoch)
print('Train done, acc={}'.format(acc_train))
acc, loss = test(model, test_loader)
print('Test done, acc={}, loss={}'.format(acc, loss))
logger.add_scalar('train_acc', acc_train, epoch)
logger.add_scalar('test_acc', acc, epoch)
logger.add_scalar('test_loss', loss, epoch)
if epoch % args.save_interval == 0:
path = os.path.join(args.model_path,
args.tag + '_{}.model'.format(epoch))
torch.save(model, path)
print('Save model @ {}'.format(path))
else:
print('testing...')
acc, loss = test(model, test_loader)
print('Test done, acc={}, loss={}'.format(acc, loss))
def train(args):
experiment_dir = mkdir('./experiment/')
checkpoints_dir = mkdir('./experiment/clf/%s/' % (args.model_name))
train_data, train_label, test_data, test_label = load_data(
'experiment/data/modelnet40_ply_hdf5_2048/')
trainDataset = ModelNetDataLoader(train_data,
train_label,
data_augmentation=args.augment)
trainDataLoader = DataLoader(trainDataset,
batch_size=args.batch_size,
shuffle=True)
testDataset = ModelNetDataLoader(test_data, test_label)
testDataLoader = torch.utils.data.DataLoader(testDataset,
batch_size=args.batch_size,
shuffle=False)
log.info('Building Model', args.model_name)
if args.model_name == 'pointnet':
num_class = 40
model = PointNetCls(num_class, args.feature_transform).cuda()
else:
model = PointNet2ClsMsg().cuda()
torch.backends.cudnn.benchmark = True
model = torch.nn.DataParallel(model).cuda()
log.debug('Using gpu:', args.gpu)
if args.pretrain is not None:
log.info('Use pretrain model...')
state_dict = torch.load(args.pretrain)
model.load_state_dict(state_dict)
init_epoch = int(args.pretrain[:-4].split('-')[-1])
log.info('start epoch from', init_epoch)
else:
log.info('Training from scratch')
init_epoch = 0
if args.optimizer == 'SGD':
optimizer = torch.optim.SGD(model.parameters(), lr=0.01, momentum=0.9)
elif args.optimizer == 'Adam':
optimizer = torch.optim.Adam(model.parameters(),
lr=args.learning_rate,
betas=(0.9, 0.999),
eps=1e-08,
weight_decay=args.decay_rate)
scheduler = torch.optim.lr_scheduler.StepLR(optimizer,
step_size=20,
gamma=0.5)
LEARNING_RATE_CLIP = 1e-5
global_epoch = 0
global_step = 0
best_tst_accuracy = 0.0
log.info('Start training...')
for epoch in range(init_epoch, args.epoch):
scheduler.step()
lr = max(optimizer.param_groups[0]['lr'], LEARNING_RATE_CLIP)
log.debug(job='clf',
model=args.model_name,
gpu=args.gpu,
epoch='%d/%s' % (epoch, args.epoch),
lr=lr)
for param_group in optimizer.param_groups:
param_group['lr'] = lr
for batch_id, data in tqdm(enumerate(trainDataLoader, 0),
total=len(trainDataLoader),
smoothing=0.9):
points, target = data
target = target[:, 0]
points = points.transpose(2, 1)
points, target = points.cuda(), target.cuda()
optimizer.zero_grad()
model = model.train()
pred, trans_feat = model(points)
loss = F.nll_loss(pred, target.long())
if args.feature_transform and args.model_name == 'pointnet':
loss += feature_transform_reguliarzer(trans_feat) * 0.001
loss.backward()
optimizer.step()
global_step += 1
log.debug('clear cuda cache')
torch.cuda.empty_cache()
acc = test_clf(model, testDataLoader)
log.info(loss='%.5f' % (loss.data))
log.info(Test_Accuracy='%.5f' % acc)
if acc >= best_tst_accuracy:
best_tst_accuracy = acc
fn_pth = 'clf-%s-%.5f-%04d.pth' % (args.model_name, acc, epoch)
log.debug('Saving model....', fn_pth)
torch.save(model.state_dict(),
os.path.join(checkpoints_dir, fn_pth))
global_epoch += 1
log.info(Best_Accuracy=best_tst_accuracy)
log.info('End of training...')
def main():
''' --- SELECT DEVICES --- '''
# Select either gpu or cpu
device = torch.device("cuda" if args.cuda else "cpu")
# Select among available GPUs
if args.cuda: os.environ["CUDA_VISIBLE_DEVICES"] = ','.join(str(x) for x in args.gpudevice)
''' --- CREATE EXPERIMENTS DIRECTORY AND LOGGERS IN TENSORBOARD --- '''
projdir = sys.path[0]
# Path for saving and loading the network.
saveloadpath = os.path.join( projdir, 'experiment\\checkpoints', args.exp_name+'.pth')
Path(os.path.dirname(saveloadpath)).mkdir(exist_ok=True, parents=True)
# timestamp = str(datetime.datetime.now().strftime('%Y-%m-%d-%H-%M'))
tblogdir = os.path.join( projdir, 'experiment\\tensorboardX', args.exp_name ) # + '_' + timestamp )
Path(tblogdir).mkdir(exist_ok=True, parents=True)
# Create tb_writer(the writer will be used to write the information on tb) by using SummaryWriter,
# flush_secs defines how much seconds need to wait for writing information.
tb_writer = SummaryWriter( logdir=tblogdir, flush_secs=3, write_to_disk=True)
''' --- INIT DATASETS AND DATALOADER (FOR SINGLE EPOCH) --- '''
# Ideal for PointNet and pointLSTM - dataloader will return (B:batch, S:seq, C:features, N:points)
dataTransformations = transforms.Compose([
ToSeries(),
DataAugmentation(),
Resampling(maxPointsPerFrame=10),
ToTensor()
])
# Init nuScenes datasets
nusc_train = NuScenes(version=args.nuscenes_train_dir, dataroot=args.nuscenes_dir, verbose=True)
train_dataset = RadarClassDataset(nusc_train, categories=args.categories, sensors=args.sensors, transforms=dataTransformations, sequence_length=1)
nusc_test = NuScenes(version=args.nuscenes_test_dir, dataroot=args.nuscenes_dir, verbose=True)
test_dataset = RadarClassDataset(nusc_test, categories=args.categories, sensors=args.sensors, transforms=dataTransformations, sequence_length=1)
# Init training data loader
trainDataLoader = DataLoader(train_dataset, batch_size=args.batchsize, shuffle=True, num_workers=args.num_workers)
''' --- INIT NETWORK MODEL --- '''
# Load selected network model and put it to right device
if args.model_name == 'pointnet':
classifier = PointNetCls(dim=args.pointCoordDim, num_class=len(args.categories), feature_transform=args.feature_transform)
elif args.model_name == 'pointnet2':
classifier = PointNet2ClsMsg(dim=args.pointCoordDim, num_class=len(args.categories) )
else:
raise Exception('Argument "model_name" does not match existent networks')
classifier = classifier.to(device)
''' --- INIT LOSS FUNCTION --- '''
loss_fun = FocalLoss(gamma=args.focalLoss_gamma, num_classes=len(args.categories), alpha=args.weight_cat).to(device)
''' --- LOAD NETWORK IF EXISTS --- '''
if os.path.exists(saveloadpath):
print('Using pretrained model found...')
checkpoint = torch.load(saveloadpath)
start_epoch = checkpoint['epoch'] +1 # Just becase make sure counting starts from 1, 2, ..., rather than 0, 1, ..., when print the information of start_epoch
iteration = checkpoint['iteration']
best_test_acc = checkpoint['test_accuracy']
classifier.load_state_dict(checkpoint['model_state_dict'])
else:
print('No existing model, starting training from scratch...')
start_epoch = 1 # Just becase make sure counting starts from 1, 2, ..., rather than 0, 1, ..., when print the information of start_epoch
iteration = 1 # Just becase make sure counting starts from 1, 2, ..., rather than 0, 1, ..., when print the information of iteration
best_test_acc = 0
''' --- CREATE OPTIMIZER ---'''
if args.optimizer == 'SGD':
optimizer = torch.optim.SGD(
classifier.parameters(),
lr=args.lr,
momentum=0.9)
elif args.optimizer == 'ADAM':
optimizer = torch.optim.Adam(
classifier.parameters(),
lr=args.lr,
betas=(0.9, 0.999),
eps=1e-08,
weight_decay=args.decay_rate)
scheduler = torch.optim.lr_scheduler.StepLR(optimizer, step_size=args.lr_epoch_half, gamma=0.5) # half(0.5) the learning rate every 'step_size' epochs
# Log info
printparams = 'Model parameters:' + json.dumps(vars(args), indent=4, sort_keys=True)
print(printparams)
tb_writer.add_text('hyper-parameters',printparams,iteration) # tb_writer.add_hparam(args)
tb_writer.add_text('dataset','dataset sample size: training: {}, test: {}'.format(len(train_dataset),len(test_dataset)),iteration)
''' --- START TRANING ---'''
for epoch in range(start_epoch, args.epoch+1):
# epoch = start_epoch
print('Epoch %d/%s:' % (epoch, args.epoch))
# Add the "learning rate" into tensorboard scalar which will be shown in tensorboard
tb_writer.add_scalar('learning_rate', optimizer.param_groups[0]['lr'], iteration)
# Beware epochs_left = args.epoch - epoch
for batch_id, data in tqdm(enumerate(trainDataLoader, 0), total=len(trainDataLoader), smoothing=0.9):
points, target = data # (B:batch x S:seq x C:features x N:points) , (B x S:seq)
# Squeeze to drop Sequence dimension, which is equal to 1, convert all the data to float(otherwise there will be data type problems when running the model) and move to device
points, target = points.squeeze(dim=1).float().to(device), target.float().to(device) # (B:batch x C:features x N:points) , (B)
# points, target = points.float().to(device), target.float().to(device)
# Reset gradients
optimizer.zero_grad()
# Sets the module in training mode
classifier = classifier.train()
# Forward propagation
pred = classifier(points)
# MLE estimator = min (- log (softmax(x)) ) = min nll_loss(log_softmax(x))
# loss = F.nll_loss(pred, target.long())
loss = loss_fun(pred, target.long())
if args.model_name == 'pointnet':
loss += feature_transform_regularizer(classifier.trans) * 0.001
if args.feature_transform:
loss += feature_transform_regularizer(classifier.trans_feat) * 0.001
# Back propagate
loss.backward()
# Update weights
optimizer.step()
# Log once for every 5 batches, add the "train_loss/cross_entropy" into tensorboard scalar which will be shown in tensorboard
if not batch_id % 5: tb_writer.add_scalar('train_loss/cross_entropy', loss.item(), iteration)
iteration += 1
# Plot train confusion matrix every X steps
if not iteration % 20:
confmatrix_train = metrics_confusion_matrix(target, pred)
print('\nTrain confusion matrix: \n',confmatrix_train)
# We just finished one epoch
# if not batch_id+1 % int(train_dataset.len__()/args.batchsize):
''' --- TEST NETWORK --- '''
if not epoch % int(args.test_every_X_epochs): # Doing the following things every epoch.
# Perform predictions on the training data.
train_targ, train_pred = test(classifier, train_dataset, device, num_workers=0, batch_size=512)
# Perform predictions on the testing data.
test_targ, test_pred = test(classifier, test_dataset, device, num_workers=0, batch_size=512)
# Calculate the accuracy rate for training data.
train_acc = metrics_accuracy(train_targ, train_pred)
# Calculate the accuracy rate for testing data.
test_acc = metrics_accuracy(test_targ, test_pred)
print('\r Training loss: {}'.format(loss.item()))
print('Train Accuracy: {}\nTest Accuracy: {}'.format(train_acc, test_acc) )
# Add the "train_acc" "test_acc" into tensorboard scalars which will be shown in tensorboard.
tb_writer.add_scalars('metrics/accuracy', {'train':train_acc, 'test':test_acc}, iteration)
# Calculate confusion matrix.
confmatrix_test = metrics_confusion_matrix(test_targ, test_pred)
print('Test confusion matrix: \n',confmatrix_test)
# Log confusion matrix.
fig, ax = plot_confusion_matrix(confmatrix_test, args.categories, normalize=False, title='Test Confusion Matrix')
# Log normalized confusion matrix.
fig_n, ax_n = plot_confusion_matrix(confmatrix_test, args.categories, normalize=True, title='Test Confusion Matrix - Normalized')
# Add the "confusion matrix" "normalized confusion matrix" into tensorboard figure which will be shown in tensorboard.
tb_writer.add_figure('test_confusion_matrix/abs', fig, global_step=iteration, close=True)
tb_writer.add_figure('test_confusion_matrix/norm', fig_n, global_step=iteration, close=True)
# Log precision recall curves.
for idx, clsname in enumerate(args.categories):
# Convert log_softmax to softmax(which is actual probability) and select the desired class.
test_pred_binary = torch.exp(test_pred[:,idx])
test_targ_binary = test_targ.eq(idx)
# Add the "precision recall curves" which will be shown in tensorboard.
tb_writer.add_pr_curve(tag='pr_curves/'+clsname, labels=test_targ_binary, predictions=test_pred_binary, global_step=iteration)
# Store the best test accuracy
if (test_acc >= best_test_acc):
best_test_acc = max([best_test_acc, test_acc])
# NOTE: we possibly want to save only when when the best test accuracy is surpassed. For now lets save every X epoch
''' --- SAVE NETWORK --- '''
if not epoch % int(args.save_every_X_epochs):
print('Best Accuracy: %f'%best_test_acc)
state = {
'epoch': epoch,
'iteration': iteration,
'test_accuracy': best_test_acc,
'model_state_dict': classifier.state_dict(),
'optimizer_state_dict': optimizer.state_dict(),
}
torch.save(state, saveloadpath)
print('Model saved!!!')
# epoch += 1
# print('Epoch %d/%s:' % (epoch, args.epoch))
scheduler.step()
tb_writer.close()
def main(args):
'''HYPER PARAMETER'''
os.environ["CUDA_VISIBLE_DEVICES"] = args.gpu
# datapath = './data/ModelNet/'
datapath = './data/modelnet40_ply_hdf5_2048/'
if args.rotation is not None:
ROTATION = (int(args.rotation[0:2]), int(args.rotation[3:5]))
else:
ROTATION = None
'''CREATE DIR'''
experiment_dir = Path('./experiment/')
experiment_dir.mkdir(exist_ok=True)
checkpoints_dir = Path('./experiment/checkpoints/')
checkpoints_dir.mkdir(exist_ok=True)
log_dir = Path('./experiment/logs/')
log_dir.mkdir(exist_ok=True)
'''LOG'''
args = parse_args()
logger = logging.getLogger("PointNet2")
logger.setLevel(logging.INFO)
formatter = logging.Formatter(
'%(asctime)s - %(name)s - %(levelname)s - %(message)s')
file_handler = logging.FileHandler(
'./experiment/logs/test_%s_' % args.model_name +
str(datetime.datetime.now().strftime('%Y-%m-%d %H-%M')) + '.txt')
file_handler.setLevel(logging.INFO)
file_handler.setFormatter(formatter)
logger.addHandler(file_handler)
logger.info(
'---------------------------------------------------Test---------------------------------------------------'
)
logger.info('PARAMETER ...')
logger.info(args)
'''DATA LOADING'''
logger.info('Load dataset ...')
train_data, train_label, test_data, test_label = load_data(
datapath, classification=True)
logger.info("The number of training data is: %d", train_data.shape[0])
logger.info("The number of test data is: %d", test_data.shape[0])
trainDataset = ModelNetDataLoader(train_data,
train_label,
rotation=ROTATION)
if ROTATION is not None:
print('The range of training rotation is', ROTATION)
testDataset = ModelNetDataLoader(test_data, test_label, rotation=ROTATION)
trainDataLoader = torch.utils.data.DataLoader(trainDataset,
batch_size=args.batchsize,
shuffle=True)
testDataLoader = torch.utils.data.DataLoader(testDataset,
batch_size=args.batchsize,
shuffle=False) # 不打乱
'''MODEL LOADING'''
num_class = 40
###################### PointNetCls ######################
classifier = PointNetCls(num_class, args.feature_transform).cuda(
) if args.model_name == 'pointnet' else PointNet2ClsMsg().cuda()
if args.pretrain is not None:
print('Use pretrain model...')
logger.info('Use pretrain model')
checkpoint = torch.load(args.pretrain)
start_epoch = checkpoint['epoch']
classifier.load_state_dict(checkpoint['model_state_dict'])
else:
print('Please Input the pretrained model ***.pth')
return
if args.optimizer == 'SGD':
optimizer = torch.optim.SGD(classifier.parameters(),
lr=0.01,
momentum=0.9)
elif args.optimizer == 'Adam':
optimizer = torch.optim.Adam(classifier.parameters(),
lr=args.learning_rate,
betas=(0.9, 0.999),
eps=1e-08,
weight_decay=args.decay_rate)
scheduler = torch.optim.lr_scheduler.StepLR(optimizer,
step_size=20,
gamma=0.5)
global_epoch = 0
global_step = 0
best_tst_accuracy = 0.0
blue = lambda x: '\033[94m' + x + '\033[0m'
'''TestING'''
logger.info('Start testing...')
scheduler.step()
acc = test(classifier.eval(), testDataLoader)
print('\r Test %s: %f' % (blue('Accuracy'), acc))
logger.info('Test Accuracy: %f', acc)
logger.info('End of testing...')
def main(args):
'''HYPER PARAMETER'''
os.environ["CUDA_VISIBLE_DEVICES"] = args.gpu
# datapath = './data/ModelNet/'
datapath = './data/objecnn20_data_hdf5_2048/'
if args.rotation is not None:
ROTATION = (int(args.rotation[0:2]), int(args.rotation[3:5]))
else:
ROTATION = None
'''CREATE DIR'''
experiment_dir = Path('./experiment/')
experiment_dir.mkdir(exist_ok=True)
file_dir = Path(
str(experiment_dir) + '/Test_%sObjectNNClf-' % args.model_name +
str(datetime.datetime.now().strftime('%Y-%m-%d_%H-%M')))
file_dir.mkdir(exist_ok=True)
checkpoints_dir = file_dir.joinpath('checkpoints/')
checkpoints_dir.mkdir(exist_ok=True)
log_dir = file_dir.joinpath('logs/')
log_dir.mkdir(exist_ok=True)
'''LOG'''
args = parse_args()
logger = logging.getLogger(args.model_name)
logger.setLevel(logging.INFO)
formatter = logging.Formatter(
'%(asctime)s - %(name)s - %(levelname)s - %(message)s')
file_handler = logging.FileHandler(
str(log_dir) + '/test_%s_ObjectNNClf.txt' % args.model_name)
file_handler.setLevel(logging.INFO)
file_handler.setFormatter(formatter)
logger.addHandler(file_handler)
logger.info(
'---------------------------------------------------Test---------------------------------------------------'
)
logger.info('PARAMETER ...')
logger.info(args)
'''DATA LOADING'''
logger.info('Load dataset ...')
database_data, database_label, query_data, query_label = load_data(
datapath, classification=True)
print(">>>>>>>>>database_data:", database_data.shape)
print(">>>>>>>>>query_data:", query_data.shape)
logger.info("The number of database_data data is: %d",
database_data.shape[0])
logger.info("The number of query_data data is: %d", query_data.shape[0])
###################### 加载 database 和 query ######################
databaseDataset = TestQueryObjectNNDataLoader(database_data,
database_label,
rotation=ROTATION)
if ROTATION is not None:
print('The range of training rotation is', ROTATION)
queryDataset = TestQueryObjectNNDataLoader(query_data,
query_label,
rotation=ROTATION)
databaseDataLoader = torch.utils.data.DataLoader(databaseDataset,
batch_size=args.batchsize,
shuffle=False)
queryDataLoader = torch.utils.data.DataLoader(queryDataset,
batch_size=args.batchsize,
shuffle=False) # 不打乱
'''MODEL LOADING'''
num_class = 20
###################### PointNetCls ######################
classifier = PointNetCls(num_class, args.feature_transform).cuda(
) if args.model_name == 'pointnet' else PointNet2ClsMsg().cuda()
# classifier = PointNetCls(num_class,args.feature_transform) if args.model_name == 'pointnet' else PointNet2ClsMsg()
if args.pretrain is not None:
print('Use pretrain model...')
logger.info('Use pretrain model')
checkpoint = torch.load(args.pretrain)
start_epoch = checkpoint['epoch']
classifier.load_state_dict(checkpoint['model_state_dict'])
else:
print('Please Input the pretrained model ***.pth')
return
if args.optimizer == 'SGD':
optimizer = torch.optim.SGD(classifier.parameters(),
lr=0.01,
momentum=0.9)
elif args.optimizer == 'Adam':
optimizer = torch.optim.Adam(classifier.parameters(),
lr=args.learning_rate,
betas=(0.9, 0.999),
eps=1e-08,
weight_decay=args.decay_rate)
scheduler = torch.optim.lr_scheduler.StepLR(optimizer,
step_size=20,
gamma=0.5)
global_epoch = 0
global_step = 0
best_tst_accuracy = 0.0
blue = lambda x: '\033[94m' + x + '\033[0m'
'''QueryING'''
logger.info('Start query...')
scheduler.step()
###################### Query ######################
classifier.eval()
_acc, database_feature_martix = getGlobalFeature('database',
classifier.eval(),
databaseDataLoader)
_acc, query_feature_martix = getGlobalFeature('query', classifier.eval(),
queryDataLoader)
