def __init__(self):
# Create logger
self.__logger = logging.getLogger('file_crawler')
self.__logger.setLevel(logging.INFO)
logging_handler = logging.StreamHandler()
logging_formatter = logging.Formatter(LOGGING_FORMAT)
logging_handler.setFormatter(logging_formatter)
self.__logger.addHandler(logging_handler)
self.__manager = FileCrawlerManager()
self.__manager.start()
# Create process-safe args object
self.__cli_args = get_cli_args(self.__manager)
# FIXME: This doesn't work correctly on Windows.
# According to the documentation here: https://docs.python.org/2.7/library/multiprocessing.html#logging
# None of the configuration will be inherited by the child processes except for the level, but in practice
# I'm seeing that even the level isn't being inherited. I tried a few different ways to pass down the level,
# but none of them worked.
if self.__cli_args.verbose:
self.__logger.setLevel(logging.DEBUG)
self.__timer = Timer(
start_message="Starting to scan %s for files matching %s" %
(self.__cli_args.root_dir, self.__cli_args.keyword.pattern))
self.__dir_queue = self.__manager.Queue()
self.__file_queue = self.__manager.Queue()
self.__results = self.__manager.FileCrawlerResults(
self.__logger.getEffectiveLevel())
self.__processes = list()
self._create_processes()
def mode_train():
"""The main training mode entrypoint"""
start_time = time.time()
io = get_io()
logline("using GPU?", tf.test.is_gpu_available())
logline("train")
enter_group()
logline("loading preprocessed data")
preprocessed = load_preprocessed(io)
logline("creating models")
train_model = create_model(batch_size=io.get("batch_size"))
logline("fitting model")
enter_group()
fit_model(io, train_model, preprocessed)
exit_group()
logline("exporting model")
export_model(train_model, io)
exit_group()
logline("done training, runtime is {}".format(
Timer.stringify_time(Timer.format_time(time.time() - start_time))))
def mode_test():
"""The main testing mode entrypoint"""
start_time = time.time()
io = get_io()
logline("test")
enter_group()
logline("reconstructing model")
model = create_model(1)
logline("applying learned weights")
model = apply_weights(model, io)
logline("reading testing files")
test_files = read_test_files(io)
logline("running testing data")
enter_group()
run_tests(io, model, test_files)
exit_group()
exit_group()
logline("done training, runtime is {}".format(
Timer.stringify_time(Timer.format_time(time.time() - start_time))))
def test_timer_automatic(self):
timer = Timer()
with freeze_time('2012-01-14 03:21:34.100'):
timer.__enter__()
with freeze_time('2012-01-14 03:21:34.255'):
timer.__exit__()
self.assertEqual(timer.interval, 155)
def test_timer_manual(self):
timer = Timer()
with freeze_time('2012-01-14 03:21:34.100'):
timer.start()
with freeze_time('2012-01-14 03:21:34.255'):
timer.stop()
self.assertEqual(timer.interval, 155)
def __init__(self, sess, model, data, logger):
super(CTPNTrainer, self).__init__(sess, model, data, logger)
self.imdb = data.load_imdb('voc_2007_trainval')
self.roidb = data.get_training_roidb(self.imdb)
self.pretrained_model = cfg.PRETRAINED_MODEL if cfg.PRETRAINED_MODEL else None
# print('Computing bounding-box regression targets...')
# if cfg.TRAIN.BBOX_REG:
# self.bbox_means, self.bbox_stds = data.add_bbox_regression_targets(self.roidb)
# print('done')
self.timer = Timer()
def mode_preprocess():
"""The main preprocessing entrypoint"""
start_time = time.time()
preprocessed = []
io = get_io()
logline("preprocessing")
enter_group()
logline("reading input paths")
enter_group()
input_paths = collect_input_paths(io)
for input_path in input_paths:
logline('found path: "{}"'.format(input_path))
exit_group()
logline("iterating files")
enter_group()
for file in get_files(input_paths):
if not file:
error("no files")
return None
features = gen_features(file)
outputs = gen_outputs(file, io)
feature_arr = list(map(lambda x: x.to_arr(), features))
output_arr = list(map(lambda x: x.to_arr(), outputs))
assert np.array(feature_arr).shape[1] == Features.length()
assert np.array(output_arr).shape[1] == OUT_VEC_SIZE
preprocessed.append({
"file_name": file.name,
"features": feature_arr,
"outputs": output_arr
})
logline('done with file: "{}"'.format(file.name))
file.close()
exit_group()
logline("done iterating files")
with open(io.get("output_file"), "wb+") as file:
pickle.dump(preprocessed, file)
logline("wrote output to file: {}".format(io.get("output_file")))
exit_group()
logline("done preprocessing, runtime is {}".format(
Timer.stringify_time(Timer.format_time(time.time() - start_time))))
def extract_features_batch(model, config, source_path, target_path, voxel_size,
device):
folders = get_folder_list(source_path)
assert len(
folders) > 0, f"Could not find 3DMatch folders under {source_path}"
logging.info(folders)
list_file = os.path.join(target_path, "list.txt")
f = open(list_file, "w")
timer, tmeter = Timer(), AverageMeter()
num_feat = 0
model.eval()
for fo in folders:
if 'evaluation' in fo:
continue
files = get_file_list(fo, ".ply")
fo_base = os.path.basename(fo)
f.write("%s %d\n" % (fo_base, len(files)))
for i, fi in enumerate(files):
# Extract features from a file
pcd = o3d.io.read_point_cloud(fi)
save_fn = "%s_%03d" % (fo_base, i)
if i % 100 == 0:
logging.info(f"{i} / {len(files)}: {save_fn}")
timer.tic()
xyz_down, feature = extract_features(model,
xyz=np.array(pcd.points),
rgb=None,
normal=None,
voxel_size=voxel_size,
device=device,
skip_check=True)
t = timer.toc()
if i > 0:
tmeter.update(t)
num_feat += len(xyz_down)
np.savez_compressed(os.path.join(target_path, save_fn),
points=np.array(pcd.points),
xyz=xyz_down,
feature=feature.detach().cpu().numpy())
if i % 20 == 0 and i > 0:
# 最后一项算的是每个点的特征提取时间
logging.info(
f'Average time: {tmeter.avg}, FPS: {num_feat / tmeter.sum}, time / feat: {tmeter.sum / num_feat}, '
)
f.close()
def run_inspect_sfm_perspective(self):
work_dir = os.path.abspath(self.config['work_dir'])
log_file = os.path.join(work_dir, 'logs/log_inspect_sfm_perspective.txt')
self.logger.set_log_file(log_file)
local_timer = Timer('inspect sfm')
local_timer.start()
# inspect sfm perspective
sfm_dir = os.path.join(work_dir, 'colmap/sfm_perspective')
for subdir in ['tri', 'tri_ba']:
dir = os.path.join(sfm_dir, subdir)
logging.info('\ninspecting {} ...'.format(dir))
inspect_dir = os.path.join(sfm_dir, 'inspect_' + subdir)
if os.path.exists(inspect_dir):
shutil.rmtree(inspect_dir)
db_path = os.path.join(sfm_dir, 'database.db')
sfm_inspector = SparseInspector(dir, db_path, inspect_dir, camera_model='PERSPECTIVE')
sfm_inspector.inspect_all()
# stop local timer
local_timer.mark('inspect sfm perspective done')
logging.info(local_timer.summary())
def clean_data_general(self, ift):
dataset_dir = self.config["dataset_dir"]
work_dir = self.config["work_dir"]
# set log file and timer
log_file = os.path.join(work_dir, "logs/log_clean_data.txt")
self.logger.set_log_file(log_file)
# create a local timer
local_timer = Timer("Data cleaning Module")
local_timer.start()
# clean data
cleaned_data_dir = os.path.join(work_dir, "cleaned_data")
if os.path.exists(cleaned_data_dir): # remove cleaned_data_dir
shutil.rmtree(cleaned_data_dir)
os.mkdir(cleaned_data_dir)
# check if dataset_dir is a list or tuple
if not (isinstance(dataset_dir, list)
or isinstance(dataset_dir, tuple)):
dataset_dir = [
dataset_dir,
]
clean_data_general(dataset_dir, cleaned_data_dir, ift=ift)
# stop local timer
local_timer.mark("Data cleaning done")
logging.info(local_timer.summary())
def run_crop_image_general(
self,
ift,
oft,
execute_parallel,
remove_aux_file,
apply_tone_mapping,
joint_tone_mapping,
):
work_dir = self.config["work_dir"]
# set log file
log_file = os.path.join(work_dir, "logs/log_crop_image.txt")
self.logger.set_log_file(log_file)
# create a local timer
local_timer = Timer("Image cropping module")
local_timer.start()
# crop image and tone map
image_crop_general(
work_dir,
ift,
oft,
execute_parallel,
remove_aux_file,
apply_tone_mapping,
joint_tone_mapping,
)
# stop local timer
local_timer.mark("image cropping done")
logging.info(local_timer.summary())
def clean_data(self):
dataset_dir = self.config['dataset_dir']
work_dir = self.config['work_dir']
# set log file and timer
log_file = os.path.join(work_dir, 'logs/log_clean_data.txt')
self.logger.set_log_file(log_file)
# create a local timer
local_timer = Timer('Data cleaning Module')
local_timer.start()
# clean data
cleaned_data_dir = os.path.join(work_dir, 'cleaned_data')
if os.path.exists(cleaned_data_dir): # remove cleaned_data_dir
shutil.rmtree(cleaned_data_dir)
os.mkdir(cleaned_data_dir)
# check if dataset_dir is a list or tuple
if not (isinstance(dataset_dir, list) or isinstance(dataset_dir, tuple)):
dataset_dir = [dataset_dir, ]
clean_data(dataset_dir, cleaned_data_dir)
# stop local timer
local_timer.mark('Data cleaning done')
logging.info(local_timer.summary())
def train(self):
curr_iter = self.curr_iter
data_loader = self.data_loader
data_loader_iter = self.data_loader.__iter__()
data_meter, data_timer, total_timer = AverageMeter(), Timer(), Timer()
total_loss = 0
total_num = 0.0
while (curr_iter < self.config.opt.max_iter):
curr_iter += 1
epoch = curr_iter / len(self.data_loader)
batch_loss, batch_pos_loss, batch_neg_loss = self._train_iter(
data_loader_iter, [data_meter, data_timer, total_timer])
total_loss += batch_loss
total_num += 1
if curr_iter % self.lr_update_freq == 0 or curr_iter == 1:
lr = self.scheduler.get_last_lr()
self.scheduler.step()
if self.is_master:
logging.info(f" Epoch: {epoch}, LR: {lr}")
self._save_checkpoint(curr_iter,
'checkpoint_' + str(curr_iter))
if curr_iter % self.config.trainer.stat_freq == 0 and self.is_master:
self.writer.add_scalar('train/loss', batch_loss, curr_iter)
self.writer.add_scalar('train/pos_loss', batch_pos_loss,
curr_iter)
self.writer.add_scalar('train/neg_loss', batch_neg_loss,
curr_iter)
logging.info(
"Train Epoch: {:.3f} [{}/{}], Current Loss: {:.3e}".format(
epoch, curr_iter, len(self.data_loader), batch_loss) +
"\tData time: {:.4f}, Train time: {:.4f}, Iter time: {:.4f}, LR: {}"
.format(data_meter.avg, total_timer.avg - data_meter.avg,
total_timer.avg, self.scheduler.get_last_lr()))
data_meter.reset()
total_timer.reset()
def run_aggregate_3d(self):
work_dir = self.config['work_dir']
# set log file
log_file = os.path.join(work_dir, 'logs/log_aggregate_3d.txt')
self.logger.set_log_file(log_file)
# create a local timer
local_timer = Timer('3D aggregation module')
local_timer.start()
aggregate_3d.run_fuse(work_dir)
# stop local timer
local_timer.mark('3D aggregation done')
logging.info(local_timer.summary())
def run_crop_image(self):
work_dir = self.config['work_dir']
# set log file
log_file = os.path.join(work_dir, 'logs/log_crop_image.txt')
self.logger.set_log_file(log_file)
# create a local timer
local_timer = Timer('Image cropping module')
local_timer.start()
# crop image and tone map
image_crop(work_dir)
# stop local timer
local_timer.mark('image cropping done')
logging.info(local_timer.summary())
def run_aggregate_2p5d(self):
work_dir = self.config['work_dir']
# set log file
log_file = os.path.join(work_dir, 'logs/log_aggregate_2p5d.txt')
self.logger.set_log_file(log_file)
# create a local timer
local_timer = Timer('2.5D aggregation module')
local_timer.start()
max_processes = -1
if 'aggregate_max_processes' in self.config:
max_processes = self.config['aggregate_max_processes']
aggregate_2p5d.run_fuse(work_dir, max_processes=max_processes)
# stop local timer
local_timer.mark('2.5D aggregation done')
logging.info(local_timer.summary())
def run_derive_approx(self):
work_dir = self.config['work_dir']
# set log file to 'logs/log_derive_approx.txt'
log_file = os.path.join(work_dir, 'logs/log_derive_approx.txt')
self.logger.set_log_file(log_file)
# create a local timer
local_timer = Timer('Derive Approximation Module')
local_timer.start()
# derive approximations for later uses
appr = CameraApprox(work_dir)
appr.approx_affine_latlonalt()
appr.approx_perspective_enu()
# stop local timer
local_timer.mark('Derive approximation done')
logging.info(local_timer.summary())
def run_colmap_mvs(self, window_radius=3):
work_dir = self.config['work_dir']
mvs_dir = os.path.join(work_dir, 'colmap/mvs')
# set log file
log_file = os.path.join(work_dir, 'logs/log_mvs.txt')
self.logger.set_log_file(log_file)
# create a local timer
local_timer = Timer('Colmap MVS Module')
local_timer.start()
# first run PMVS without filtering
run_photometric_mvs(mvs_dir, window_radius)
# next do forward-backward checking and filtering
run_consistency_check(mvs_dir, window_radius)
# stop local timer
local_timer.mark('Colmap MVS done')
logging.info(local_timer.summary())
def run_colmap_sfm_perspective(self, weight=0.01):
work_dir = os.path.abspath(self.config['work_dir'])
sfm_dir = os.path.join(work_dir, 'colmap/sfm_perspective')
if not os.path.exists(sfm_dir):
os.mkdir(sfm_dir)
log_file = os.path.join(work_dir, 'logs/log_sfm_perspective.txt')
self.logger.set_log_file(log_file)
# create a local timer
local_timer = Timer('Colmap SfM Module, perspective camera')
local_timer.start()
# create a hard link to avoid copying of images
if os.path.exists(os.path.join(sfm_dir, 'images')):
os.unlink(os.path.join(sfm_dir, 'images'))
os.symlink(os.path.relpath(os.path.join(work_dir, 'colmap/subset_for_sfm/images'), sfm_dir),
os.path.join(sfm_dir, 'images'))
init_camera_file = os.path.join(work_dir, 'colmap/subset_for_sfm/perspective_dict.json')
colmap_sfm_perspective.run_sfm(work_dir, sfm_dir, init_camera_file, weight)
# stop local timer
local_timer.mark('Colmap SfM done')
logging.info(local_timer.summary())
def calibrate_neighbors(dataset,
config,
collate_fn,
keep_ratio=0.8,
samples_threshold=2000):
timer = Timer()
last_display = timer.total_time
# From config parameter, compute higher bound of neighbors number in a neighborhood
hist_n = int(np.ceil(4 / 3 * np.pi * (config.deform_radius + 1)**3))
neighb_hists = np.zeros((config.num_layers, hist_n), dtype=np.int32)
# Get histogram of neighborhood sizes i in 1 epoch max.
for i in range(len(dataset)):
timer.tic()
batched_input = collate_fn([dataset[i]],
config,
neighborhood_limits=[hist_n] * 5)
# update histogram
counts = [
torch.sum(neighb_mat < neighb_mat.shape[0], dim=1).numpy()
for neighb_mat in batched_input['neighbors']
]
hists = [np.bincount(c, minlength=hist_n)[:hist_n] for c in counts]
neighb_hists += np.vstack(hists)
timer.toc()
if timer.total_time - last_display > 0.1:
last_display = timer.total_time
print(f"Calib Neighbors {i:08d}: timings {timer.total_time:4.2f}s")
if np.min(np.sum(neighb_hists, axis=1)) > samples_threshold:
break
cumsum = np.cumsum(neighb_hists.T, axis=0)
percentiles = np.sum(cumsum < (keep_ratio * cumsum[hist_n - 1, :]), axis=0)
neighborhood_limits = percentiles
print('\n')
return neighborhood_limits
def ctpn(self, sess, net, image_name):
"""
:param sess: 会话
:param net: 创建的测试网络
:param image_name: 所要测试的单张图片的目录
:return:
"""
timer = Timer()
timer.tic()
# 读取图片
image = cv2.imread(image_name)
shape = image.shape[:2] # 获取高，宽
# resize_im，返回缩放后的图片和相应的缩放比。缩放比定义为 修改后的图/原图
img, scale = TestClass.resize_im(image,
scale=self._cfg.TEST.SCALE,
max_scale=self._cfg.TEST.MAX_SCALE)
# 将图片去均值化
im_orig = img.astype(np.float32, copy=True)
im_orig -= self._cfg.TRAIN.PIXEL_MEANS
# 将缩放和去均值化以后的图片，放入网络进行前向计算，获取分数和对应的文本片段，该片段为映射到最原始图片的坐标
scores, boxes = TestClass.test_ctpn(sess, net, im_orig, scale)
# 此处调用了一个文本检测器
textdetector = TextDetector(self._cfg)
"""
输入参数分别为：
N×4矩阵，每行为一个已经映射回最初的图片的文字片段坐标
N维向量，对应的分数
两维向量，分别为最原始图片的高宽
返回：
一个N×9的矩阵，表示N个拼接以后的完整的文本框。每一行，前八个元素一次是左上，右上，左下，右下的坐标，最后一个元素是文本框的分数
"""
boxes = textdetector.detect(boxes, scores, shape)
self.draw_boxes(image, image_name, boxes, scale)
timer.toc()
print(('Detection took {:.3f}s for '
'{:d} object proposals').format(timer.total_time,
boxes.shape[0]))
def _train_epoch(self, epoch):
config = self.config
gc.collect()
self.model.train()
# Epoch starts from 1
total_loss = 0
total_num = 0.0
data_loader = self.data_loader
data_loader_iter = self.data_loader.__iter__()
iter_size = self.iter_size
data_meter, data_timer, total_timer = AverageMeter(), Timer(), Timer()
pos_dist_meter, neg_dist_meter = AverageMeter(), AverageMeter()
start_iter = (epoch - 1) * (len(data_loader) // iter_size)
for curr_iter in range(len(data_loader) // iter_size):
self.optimizer.zero_grad()
batch_loss = 0
data_time = 0
total_timer.tic()
for iter_idx in range(iter_size):
data_timer.tic()
input_dict = data_loader_iter.next()
data_time += data_timer.toc(average=False)
# pairs consist of (xyz1 index, xyz0 index)
sinput0 = ME.SparseTensor(
input_dict['sinput0_F'], coords=input_dict['sinput0_C']).to(self.device)
F0 = self.model(sinput0).F
sinput1 = ME.SparseTensor(
input_dict['sinput1_F'], coords=input_dict['sinput1_C']).to(self.device)
F1 = self.model(sinput1).F
pos_pairs = input_dict['correspondences']
loss, pos_dist, neg_dist = self.triplet_loss(
F0,
F1,
pos_pairs,
num_pos=config.triplet_num_pos * config.batch_size,
num_hn_samples=config.triplet_num_hn * config.batch_size,
num_rand_triplet=config.triplet_num_rand * config.batch_size)
loss /= iter_size
loss.backward()
batch_loss += loss.item()
pos_dist_meter.update(pos_dist)
neg_dist_meter.update(neg_dist)
self.optimizer.step()
gc.collect()
torch.cuda.empty_cache()
total_loss += batch_loss
total_num += 1.0
total_timer.toc()
data_meter.update(data_time)
if curr_iter % self.config.stat_freq == 0:
self.writer.add_scalar('train/loss', batch_loss, start_iter + curr_iter)
logging.info(
"Train Epoch: {} [{}/{}], Current Loss: {:.3e}, Pos dist: {:.3e}, Neg dist: {:.3e}"
.format(epoch, curr_iter,
len(self.data_loader) //
iter_size, batch_loss, pos_dist_meter.avg, neg_dist_meter.avg) +
"\tData time: {:.4f}, Train time: {:.4f}, Iter time: {:.4f}".format(
data_meter.avg, total_timer.avg - data_meter.avg, total_timer.avg))
pos_dist_meter.reset()
neg_dist_meter.reset()
data_meter.reset()
total_timer.reset()
def _train_epoch(self, epoch):
gc.collect()
self.model.train()
# Epoch starts from 1
total_loss = 0
total_num = 0.0
data_loader = self.data_loader
data_loader_iter = self.data_loader.__iter__()
iter_size = self.iter_size
data_meter, data_timer, total_timer = AverageMeter(), Timer(), Timer()
start_iter = (epoch - 1) * (len(data_loader) // iter_size)
for curr_iter in range(len(data_loader) // iter_size):
self.optimizer.zero_grad()
batch_pos_loss, batch_neg_loss, batch_loss = 0, 0, 0
data_time = 0
total_timer.tic()
for iter_idx in range(iter_size):
data_timer.tic()
input_dict = data_loader_iter.next()
data_time += data_timer.toc(average=False)
sinput0 = ME.SparseTensor(
input_dict['sinput0_F'], coords=input_dict['sinput0_C']).to(self.device)
F0 = self.model(sinput0).F
sinput1 = ME.SparseTensor(
input_dict['sinput1_F'], coords=input_dict['sinput1_C']).to(self.device)
F1 = self.model(sinput1).F
pos_pairs = input_dict['correspondences']
pos_loss, neg_loss = self.contrastive_hardest_negative_loss(
F0,
F1,
pos_pairs,
num_pos=self.config.num_pos_per_batch * self.config.batch_size,
num_hn_samples=self.config.num_hn_samples_per_batch *
self.config.batch_size)
pos_loss /= iter_size
neg_loss /= iter_size
loss = pos_loss + self.neg_weight * neg_loss
loss.backward()
batch_loss += loss.item()
batch_pos_loss += pos_loss.item()
batch_neg_loss += neg_loss.item()
self.optimizer.step()
gc.collect()
torch.cuda.empty_cache()
total_loss += batch_loss
total_num += 1.0
total_timer.toc()
data_meter.update(data_time)
if curr_iter % self.config.stat_freq == 0:
self.writer.add_scalar('train/loss', batch_loss, start_iter + curr_iter)
self.writer.add_scalar('train/pos_loss', batch_pos_loss, start_iter + curr_iter)
self.writer.add_scalar('train/neg_loss', batch_neg_loss, start_iter + curr_iter)
logging.info(
"Train Epoch: {} [{}/{}], Current Loss: {:.3e} Pos: {:.3f} Neg: {:.3f}"
.format(epoch, curr_iter,
len(self.data_loader) //
iter_size, batch_loss, batch_pos_loss, batch_neg_loss) +
"\tData time: {:.4f}, Train time: {:.4f}, Iter time: {:.4f}".format(
data_meter.avg, total_timer.avg - data_meter.avg, total_timer.avg))
data_meter.reset()
total_timer.reset()
def _valid_epoch(self):
# Change the network to evaluation mode
self.model.eval()
self.val_data_loader.dataset.reset_seed(0)
num_data = 0
hit_ratio_meter, feat_match_ratio, loss_meter, rte_meter, rre_meter = AverageMeter(
), AverageMeter(), AverageMeter(), AverageMeter(), AverageMeter()
data_timer, feat_timer, matching_timer = Timer(), Timer(), Timer()
tot_num_data = len(self.val_data_loader.dataset)
if self.val_max_iter > 0:
tot_num_data = min(self.val_max_iter, tot_num_data)
data_loader_iter = self.val_data_loader.__iter__()
for batch_idx in range(tot_num_data):
data_timer.tic()
input_dict = data_loader_iter.next()
data_timer.toc()
# pairs consist of (xyz1 index, xyz0 index)
feat_timer.tic()
sinput0 = ME.SparseTensor(
input_dict['sinput0_F'], coords=input_dict['sinput0_C']).to(self.device)
F0 = self.model(sinput0).F
sinput1 = ME.SparseTensor(
input_dict['sinput1_F'], coords=input_dict['sinput1_C']).to(self.device)
F1 = self.model(sinput1).F
feat_timer.toc()
matching_timer.tic()
xyz0, xyz1, T_gt = input_dict['pcd0'], input_dict['pcd1'], input_dict['T_gt']
xyz0_corr, xyz1_corr = self.find_corr(xyz0, xyz1, F0, F1, subsample_size=5000)
T_est = te.est_quad_linear_robust(xyz0_corr, xyz1_corr)
loss = corr_dist(T_est, T_gt, xyz0, xyz1, weight=None)
loss_meter.update(loss)
rte = np.linalg.norm(T_est[:3, 3] - T_gt[:3, 3])
rte_meter.update(rte)
rre = np.arccos((np.trace(T_est[:3, :3].t() @ T_gt[:3, :3]) - 1) / 2)
if not np.isnan(rre):
rre_meter.update(rre)
hit_ratio = self.evaluate_hit_ratio(
xyz0_corr, xyz1_corr, T_gt, thresh=self.config.hit_ratio_thresh)
hit_ratio_meter.update(hit_ratio)
feat_match_ratio.update(hit_ratio > 0.05)
matching_timer.toc()
num_data += 1
torch.cuda.empty_cache()
if batch_idx % 100 == 0 and batch_idx > 0:
logging.info(' '.join([
f"Validation iter {num_data} / {tot_num_data} : Data Loading Time: {data_timer.avg:.3f},",
f"Feature Extraction Time: {feat_timer.avg:.3f}, Matching Time: {matching_timer.avg:.3f},",
f"Loss: {loss_meter.avg:.3f}, RTE: {rte_meter.avg:.3f}, RRE: {rre_meter.avg:.3f},",
f"Hit Ratio: {hit_ratio_meter.avg:.3f}, Feat Match Ratio: {feat_match_ratio.avg:.3f}"
]))
data_timer.reset()
logging.info(' '.join([
f"Final Loss: {loss_meter.avg:.3f}, RTE: {rte_meter.avg:.3f}, RRE: {rre_meter.avg:.3f},",
f"Hit Ratio: {hit_ratio_meter.avg:.3f}, Feat Match Ratio: {feat_match_ratio.avg:.3f}"
]))
return {
"loss": loss_meter.avg,
"rre": rre_meter.avg,
"rte": rte_meter.avg,
'feat_match_ratio': feat_match_ratio.avg,
'hit_ratio': hit_ratio_meter.avg
}
def _train_epoch(self, epoch):
gc.collect()
self.model.train()
# Epoch starts from 1
total_loss = 0
total_num = 0.0
data_loader = self.data_loader
data_loader_iter = self.data_loader.__iter__()
iter_size = self.iter_size
start_iter = (epoch - 1) * (len(data_loader) // iter_size)
data_meter, data_timer, total_timer = AverageMeter(), Timer(), Timer()
# Main training
for curr_iter in range(len(data_loader) // iter_size):
self.optimizer.zero_grad()
batch_pos_loss, batch_neg_loss, batch_loss = 0, 0, 0
data_time = 0
total_timer.tic()
for iter_idx in range(iter_size):
# Caffe iter size
data_timer.tic()
input_dict = data_loader_iter.next()
data_time += data_timer.toc(average=False)
# pairs consist of (xyz1 index, xyz0 index)
sinput0 = ME.SparseTensor(
input_dict['sinput0_F'], coords=input_dict['sinput0_C']).to(self.device)
F0 = self.model(sinput0).F
sinput1 = ME.SparseTensor(
input_dict['sinput1_F'], coords=input_dict['sinput1_C']).to(self.device)
F1 = self.model(sinput1).F
N0, N1 = len(sinput0), len(sinput1)
pos_pairs = input_dict['correspondences']
neg_pairs = self.generate_rand_negative_pairs(pos_pairs, max(N0, N1), N0, N1)
pos_pairs = pos_pairs.long().to(self.device)
neg_pairs = torch.from_numpy(neg_pairs).long().to(self.device)
neg0 = F0.index_select(0, neg_pairs[:, 0])
neg1 = F1.index_select(0, neg_pairs[:, 1])
pos0 = F0.index_select(0, pos_pairs[:, 0])
pos1 = F1.index_select(0, pos_pairs[:, 1])
# Positive loss
pos_loss = (pos0 - pos1).pow(2).sum(1)
# Negative loss
neg_loss = F.relu(self.neg_thresh -
((neg0 - neg1).pow(2).sum(1) + 1e-4).sqrt()).pow(2)
pos_loss_mean = pos_loss.mean() / iter_size
neg_loss_mean = neg_loss.mean() / iter_size
# Weighted loss
loss = pos_loss_mean + self.neg_weight * neg_loss_mean
loss.backward(
) # To accumulate gradient, zero gradients only at the begining of iter_size
batch_loss += loss.item()
batch_pos_loss += pos_loss_mean.item()
batch_neg_loss += neg_loss_mean.item()
self.optimizer.step()
torch.cuda.empty_cache()
total_loss += batch_loss
total_num += 1.0
total_timer.toc()
data_meter.update(data_time)
# Print logs
if curr_iter % self.config.stat_freq == 0:
self.writer.add_scalar('train/loss', batch_loss, start_iter + curr_iter)
self.writer.add_scalar('train/pos_loss', batch_pos_loss, start_iter + curr_iter)
self.writer.add_scalar('train/neg_loss', batch_neg_loss, start_iter + curr_iter)
logging.info(
"Train Epoch: {} [{}/{}], Current Loss: {:.3e} Pos: {:.3f} Neg: {:.3f}"
.format(epoch, curr_iter,
len(self.data_loader) //
iter_size, batch_loss, batch_pos_loss, batch_neg_loss) +
"\tData time: {:.4f}, Train time: {:.4f}, Iter time: {:.4f}".format(
data_meter.avg, total_timer.avg - data_meter.avg, total_timer.avg))
data_meter.reset()
total_timer.reset()
def _valid_epoch(self, data_loader_iter):
# Change the network to evaluation mode
self.model.eval()
num_data = 0
hit_ratio_meter, reciprocity_ratio_meter = AverageMeter(
), AverageMeter()
reciprocity_hit_ratio_meter = AverageMeter()
data_timer, feat_timer = Timer(), Timer()
tot_num_data = len(self.val_data_loader.dataset)
if self.val_max_iter > 0:
tot_num_data = min(self.val_max_iter, tot_num_data)
for curr_iter in range(tot_num_data):
data_timer.tic()
input_dict = self.get_data(data_loader_iter)
data_timer.toc()
# pairs consist of (xyz1 index, xyz0 index)
feat_timer.tic()
with torch.no_grad():
F0 = self.model(input_dict['img0'].to(self.device))
F1 = self.model(input_dict['img1'].to(self.device))
feat_timer.toc()
# Test self.num_pos_per_batch * self.batch_size features only.
_, _, H0, W0 = F0.shape
_, _, H1, W1 = F1.shape
for batch_idx, pair in enumerate(input_dict['pairs']):
N = len(pair)
sel = np.random.choice(N,
min(N, self.config.num_pos_per_batch),
replace=False)
curr_pair = pair[sel]
w0, h0, w1, h1 = torch.floor(curr_pair.t() /
self.out_tensor_stride).long()
feats0 = F0[batch_idx, :, h0, w0]
nn_inds1 = find_nn_gpu(feats0,
F1[batch_idx, :].view(F1.shape[1], -1),
nn_max_n=self.config.nn_max_n,
transposed=True)
# Convert the index to coordinate: BxCxHxW
xs1 = nn_inds1 % W1
ys1 = nn_inds1 // W1
# Test reciprocity
nn_inds0 = find_nn_gpu(F1[batch_idx, :, ys1, xs1],
F0[batch_idx, :].view(F0.shape[1], -1),
nn_max_n=self.config.nn_max_n,
transposed=True)
# Convert the index to coordinate: BxCxHxW
xs0 = nn_inds0 % W0
ys0 = nn_inds0 // W0
dist_sq = (w1 - xs1)**2 + (h1 - ys1)**2
is_correct = dist_sq < (self.config.ucn_inlier_threshold_pixel
/ self.out_tensor_stride)**2
hit_ratio_meter.update(is_correct.sum().item() /
len(is_correct))
# Recipocity test result
dist_sq_nn = (w0 - xs0)**2 + (h0 - ys0)**2
mask = dist_sq_nn < (self.config.ucn_inlier_threshold_pixel /
self.out_tensor_stride)**2
reciprocity_ratio_meter.update(mask.sum().item() /
float(len(mask)))
reciprocity_hit_ratio_meter.update(
is_correct[mask].sum().item() / (mask.sum().item() + eps))
torch.cuda.empty_cache()
# visualize_image_correspondence(input_dict['img0'][batch_idx, 0].numpy() + 0.5,
# input_dict['img1'][batch_idx, 0].numpy() + 0.5,
# F0[batch_idx], F1[batch_idx], curr_iter,
# self.config)
num_data += 1
if num_data % 100 == 0:
logging.info(', '.join([
f"Validation iter {num_data} / {tot_num_data} : Data Loading Time: {data_timer.avg:.3f}",
f"Feature Extraction Time: {feat_timer.avg:.3f}, Hit Ratio: {hit_ratio_meter.avg}",
f"Reciprocity Ratio: {reciprocity_ratio_meter.avg}, Reci Filtered Hit Ratio: {reciprocity_hit_ratio_meter.avg}"
]))
data_timer.reset()
logging.info(', '.join([
f"Validation : Data Loading Time: {data_timer.avg:.3f}",
f"Feature Extraction Time: {feat_timer.avg:.3f}, Hit Ratio: {hit_ratio_meter.avg}",
f"Reciprocity Ratio: {reciprocity_ratio_meter.avg}, Reci Filtered Hit Ratio: {reciprocity_hit_ratio_meter.avg}"
]))
return {
'hit_ratio': hit_ratio_meter.avg,
'reciprocity_ratio': reciprocity_ratio_meter.avg,
'reciprocity_hit_ratio': reciprocity_hit_ratio_meter.avg,
}
def _train_epoch(self, epoch, data_loader_iter):
# Epoch starts from 1
total_loss = 0
total_num = 0.0
iter_size = self.iter_size
data_meter, data_timer, total_timer = AverageMeter(), Timer(), Timer()
for curr_iter in range(self.train_max_iter):
self.optimizer.zero_grad()
batch_pos_loss, batch_neg_loss, batch_loss = 0, 0, 0
data_time = 0
total_timer.tic()
for iter_idx in range(iter_size):
data_timer.tic()
input_dict = self.get_data(data_loader_iter)
data_time += data_timer.toc(average=False)
F0 = self.model(input_dict['img0'].to(self.device))
F1 = self.model(input_dict['img1'].to(self.device))
pos_loss, neg_loss = self.contrastive_loss(
input_dict['img0'].numpy() + 0.5,
input_dict['img1'].numpy() + 0.5,
F0,
F1,
input_dict['pairs'],
num_pos=self.config.num_pos_per_batch,
num_hn_samples=self.config.num_hn_samples_per_batch)
pos_loss /= iter_size
neg_loss /= iter_size
loss = pos_loss + self.neg_weight * neg_loss
loss.backward()
batch_loss += loss.item()
batch_pos_loss += pos_loss.item()
batch_neg_loss += neg_loss.item()
self.optimizer.step()
gc.collect()
torch.cuda.empty_cache()
total_loss += batch_loss
total_num += 1.0
total_timer.toc()
data_meter.update(data_time)
torch.cuda.empty_cache()
if curr_iter % self.config.stat_freq == 0:
self.writer.add_scalar('train/loss', batch_loss, curr_iter)
self.writer.add_scalar('train/pos_loss', batch_pos_loss,
curr_iter)
self.writer.add_scalar('train/neg_loss', batch_neg_loss,
curr_iter)
logging.info(
"Train epoch {}, iter {}, Current Loss: {:.3e} Pos: {:.3f} Neg: {:.3f}"
.format(epoch, curr_iter, batch_loss, batch_pos_loss,
batch_neg_loss) +
"\tData time: {:.4f}, Train time: {:.4f}, Iter time: {:.4f}"
.format(data_meter.avg, total_timer.avg -
data_meter.avg, total_timer.avg))
data_meter.reset()
total_timer.reset()
class CTPNTrainer(BaseTrain):
def __init__(self, sess, model, data, logger):
super(CTPNTrainer, self).__init__(sess, model, data, logger)
self.imdb = data.load_imdb('voc_2007_trainval')
self.roidb = data.get_training_roidb(self.imdb)
self.pretrained_model = cfg.PRETRAINED_MODEL if cfg.PRETRAINED_MODEL else None
# print('Computing bounding-box regression targets...')
# if cfg.TRAIN.BBOX_REG:
# self.bbox_means, self.bbox_stds = data.add_bbox_regression_targets(self.roidb)
# print('done')
self.timer = Timer()
def get_train_op(self, loss):
lr = tf.Variable(cfg.TRAIN.LEARNING_RATE, trainable=False)
if cfg.TRAIN.SOLVER == 'Adam':
opt = tf.train.AdamOptimizer(cfg.TRAIN.LEARNING_RATE)
elif cfg.TRAIN.SOLVER == 'RMS':
opt = tf.train.RMSPropOptimizer(cfg.TRAIN.LEARNING_RATE)
else:
momentum = cfg.TRAIN.MOMENTUM
opt = tf.train.MomentumOptimizer(lr, momentum)
if cfg.TRAIN.WITH_CLIP:
tvars = tf.trainable_variables()
grads, _ = tf.clip_by_global_norm(tf.gradients(loss, tvars), 10.0)
train_op = opt.apply_gradients(list(zip(grads, tvars)),
global_step=self.global_step)
else:
train_op = opt.minimize(loss, global_step=self.global_step)
return train_op, lr
def load_model(self, restore):
restore_iter = 0
if self.pretrained_model is not None and not restore:
try:
print(('Loading pretrained model weights from {:s}').format(
self.pretrained_model))
self.model.load_npz(self.pretrained_model, self.sess, True)
except:
raise 'Check your pretrained model {:s}'.format(
self.pretrained_model)
# resuming a trainer
if restore:
ckpt_path = self.model.load_ckpt(self.sess)
stem = os.path.splitext(os.path.basename(ckpt_path))[0]
restore_iter = int(stem.split('_')[-1])
self.sess.run(self.global_step.assign(restore_iter))
return restore_iter
def train(self, max_iters, restore=False):
"""Network training loop."""
data_layer = DataGenerator(self.roidb, self.imdb.nrof_classes,
self.data)
total_loss, model_loss, rpn_cross_entropy, rpn_loss_box = self.model.build_loss(
ohem=cfg.TRAIN.OHEM)
summary_op, log_image, log_image_data, log_image_name = self.logger.init_summary(
rpn_reg_loss=rpn_loss_box,
rpn_cls_loss=rpn_cross_entropy,
model_loss=model_loss,
total_loss=total_loss)
train_op, lr = self.get_train_op(total_loss)
# intialize variables
self.sess.run(tf.global_variables_initializer())
restore_iter = self.load_model(restore)
fetch_list = [
total_loss, model_loss, rpn_cross_entropy, rpn_loss_box,
summary_op, train_op
]
print(restore_iter, max_iters)
for _iter in range(restore_iter, max_iters, cfg.TRAIN.EPOCH_SIZE):
losses = self.train_epoch(_iter, lr, data_layer, fetch_list)
print('iter: %d / %d, total loss: %.4f, model loss: %.4f, rpn_loss_cls: %.4f, rpn_loss_box: %.4f, lr: %f' % \
(_iter+cfg.TRAIN.EPOCH_SIZE, max_iters, losses[0], losses[1], losses[2], losses[3], losses[5].eval()))
self.logger.summarize(losses[4], self.global_step.eval())
self.save(_iter + cfg.TRAIN.EPOCH_SIZE)
def train_epoch(self, tm_iter, lr, data_layer, fetch_list):
loop = tqdm(range(cfg.TRAIN.EPOCH_SIZE))
for _iter in loop:
tm_iter += _iter
if tm_iter != 0 and tm_iter % cfg.TRAIN.STEPSIZE == 0:
self.sess.run(tf.assign(lr, lr.eval() * cfg.TRAIN.GAMMA))
self.timer.tic()
total_loss_val, model_loss_val, rpn_loss_cls_val, rpn_loss_box_val, summary_str, _ = \
self.train_step(data_layer,fetch_list)
_diff_time = self.timer.toc(average=False)
print('speed: {:.3f}s / iter'.format(_diff_time))
return total_loss_val, model_loss_val, rpn_loss_cls_val, rpn_loss_box_val, summary_str, lr
def train_step(self, data_layer, fetch_list):
blobs = data_layer.forward()
feed_dict = {
self.model.data: blobs['data'],
self.model.im_info: blobs['im_info'],
self.model.keep_prob: 0.5,
self.model.gt_boxes: blobs['gt_boxes'],
self.model.gt_ishard: blobs['gt_ishard'],
self.model.dontcare_areas: blobs['dontcare_areas']
}
return self.sess.run(fetches=fetch_list, feed_dict=feed_dict)
from lib.spark import spark, sc
from lib.plotly import py
from lib.timer import Timer
import plotly.graph_objs as go
from pyspark.ml.feature import VectorAssembler
from scipy.spatial import ConvexHull
with Timer('read', 'Reading data'):
df = df_base = spark.read.csv('data/yellow_tripdata_2016-01.csv', header=True, inferSchema=True)
with Timer('sample', 'Sampling data'):
df = df.sample(False, 0.005)
from lib.process import process
with Timer('process', 'Cleaning invalid data'):
df = process(df)
from pyspark.sql.functions import col, udf
from pyspark.sql.types import IntegerType
from pyspark import StorageLevel
K = 6
N = 24//K
groups = {i: range(i*N, i*N+N) for i in range(K)}
@udf(returnType=IntegerType())
def get_group(d):
import numpy as np
from lib.timer import Timer
import breakout_detection
import runners.data_loader as data_loader
import matplotlib.pyplot as plt
SAMPLE_FILE_PATH = '../data/demo7.csv'
if __name__ == '__main__':
sw = Timer()
data = data_loader.load_data(SAMPLE_FILE_PATH)
sw.start()
edm_multi = breakout_detection.EdmMulti()
max_snp = max(max(data.values), 1)
# Z = [x/float(max_snp) for x in data.values]
Z = [x for x in data.values]
edm_multi.evaluate(Z, min_size=24, beta=0.001, degree=1)
print(sw.elapsed(f'data length: {len(data.values)}, using time:'))
plt.plot(np.asarray(data.index).tolist(), Z)
result = edm_multi.getLoc()
print(result)
for i in result:
plt.axvline(np.asarray(data.index).tolist()[i], color='#FF4E24')
# plt.plot(np.asarray(data.index).tolist()[i], np.asarray(data.values).tolist()[i], 'ro')
plt.show()
from lib.spark import spark, sc
from lib.plotly import py
from lib.timer import Timer
from lib.process import process
import plotly.graph_objs as go
from pyspark.ml.feature import VectorAssembler
from scipy.spatial import ConvexHull
import pickle
import numpy as np
with Timer('read', 'Reading data'):
df = df_base = spark.read.csv('data/yellow_tripdata_2016-01.csv',
header=True,
inferSchema=True)
with Timer('process', 'Cleaning invalid data'):
df = process(df)
from pyspark.sql.functions import col, udf
from pyspark.sql.types import IntegerType
from pyspark import StorageLevel
from pyspark.ml.clustering import GaussianMixture
gmm = GaussianMixture(k=1000)
result = []
with Timer('clustering', 'Computing clusters'):
for weekday in range(7):
for hour in range(24):
with Timer('clustering',
def main(config):
test_loader = make_data_loader(
config, config.test_phase, 1, num_threads=config.test_num_workers, shuffle=True)
num_feats = 1
device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
Model = load_model(config.model)
model = Model(
num_feats,
config.model_n_out,
bn_momentum=config.bn_momentum,
conv1_kernel_size=config.conv1_kernel_size,
normalize_feature=config.normalize_feature)
checkpoint = torch.load(config.save_dir + '/checkpoint.pth')
model.load_state_dict(checkpoint['state_dict'])
model = model.to(device)
model.eval()
success_meter, rte_meter, rre_meter = AverageMeter(), AverageMeter(), AverageMeter()
data_timer, feat_timer, reg_timer = Timer(), Timer(), Timer()
test_iter = test_loader.__iter__()
N = len(test_iter)
n_gpu_failures = 0
# downsample_voxel_size = 2 * config.voxel_size
for i in range(len(test_iter)):
data_timer.tic()
try:
data_dict = test_iter.next()
except ValueError:
n_gpu_failures += 1
logging.info(f"# Erroneous GPU Pair {n_gpu_failures}")
continue
data_timer.toc()
xyz0, xyz1 = data_dict['pcd0'], data_dict['pcd1']
T_gth = data_dict['T_gt']
xyz0np, xyz1np = xyz0.numpy(), xyz1.numpy()
pcd0 = make_open3d_point_cloud(xyz0np)
pcd1 = make_open3d_point_cloud(xyz1np)
with torch.no_grad():
feat_timer.tic()
sinput0 = ME.SparseTensor(
data_dict['sinput0_F'].to(device), coordinates=data_dict['sinput0_C'].to(device))
F0 = model(sinput0).F.detach()
sinput1 = ME.SparseTensor(
data_dict['sinput1_F'].to(device), coordinates=data_dict['sinput1_C'].to(device))
F1 = model(sinput1).F.detach()
feat_timer.toc()
feat0 = make_open3d_feature(F0, 32, F0.shape[0])
feat1 = make_open3d_feature(F1, 32, F1.shape[0])
reg_timer.tic()
distance_threshold = config.voxel_size * 1.0
ransac_result = o3d.registration.registration_ransac_based_on_feature_matching(
pcd0, pcd1, feat0, feat1, distance_threshold,
o3d.registration.TransformationEstimationPointToPoint(False), 4, [
o3d.registration.CorrespondenceCheckerBasedOnEdgeLength(0.9),
o3d.registration.CorrespondenceCheckerBasedOnDistance(
distance_threshold)
], o3d.registration.RANSACConvergenceCriteria(4000000, 10000))
T_ransac = torch.from_numpy(
ransac_result.transformation.astype(np.float32))
reg_timer.toc()
# Translation error
rte = np.linalg.norm(T_ransac[:3, 3] - T_gth[:3, 3])
rre = np.arccos((np.trace(T_ransac[:3, :3].t() @ T_gth[:3, :3]) - 1) / 2)
# Check if the ransac was successful. successful if rte < 2m and rre < 5◦
# http://openaccess.thecvf.com/content_ECCV_2018/papers/Zi_Jian_Yew_3DFeat-Net_Weakly_Supervised_ECCV_2018_paper.pdf
if rte < 2:
rte_meter.update(rte)
if not np.isnan(rre) and rre < np.pi / 180 * 5:
rre_meter.update(rre)
if rte < 2 and not np.isnan(rre) and rre < np.pi / 180 * 5:
success_meter.update(1)
else:
success_meter.update(0)
logging.info(f"Failed with RTE: {rte}, RRE: {rre}")
if i % 10 == 0:
logging.info(
f"{i} / {N}: Data time: {data_timer.avg}, Feat time: {feat_timer.avg}," +
f" Reg time: {reg_timer.avg}, RTE: {rte_meter.avg}," +
f" RRE: {rre_meter.avg}, Success: {success_meter.sum} / {success_meter.count}"
+ f" ({success_meter.avg * 100} %)")
data_timer.reset()
feat_timer.reset()
reg_timer.reset()
logging.info(
f"RTE: {rte_meter.avg}, var: {rte_meter.var}," +
f" RRE: {rre_meter.avg}, var: {rre_meter.var}, Success: {success_meter.sum} " +
f"/ {success_meter.count} ({success_meter.avg * 100} %)")
