def train():
pt_num = 4096
voxel_num = FLAGS.split_num * FLAGS.split_num * FLAGS.split_num
train_list, test_list = get_train_test_split()
train_list = [
'data/S3DIS/folding/block_v2/' + fn + '.h5' for fn in train_list
]
test_list = [
'data/S3DIS/folding/block_v2/' + fn + '.h5' for fn in test_list
]
train_provider = ProviderV2(train_list, 'train',
FLAGS.batch_size * FLAGS.num_gpus, batch_fn,
read_fn, 2)
test_provider = ProviderV2(test_list, 'test',
FLAGS.batch_size * FLAGS.num_gpus, batch_fn,
read_fn, 2)
try:
pls = {}
pls['points'] = tf.placeholder(tf.float32, [None, pt_num, 6], 'points')
pls['covars'] = tf.placeholder(tf.float32, [None, pt_num, 9], 'covars')
pls['rpoints'] = tf.placeholder(tf.float32, [None, pt_num, 3],
'rpoints')
pls['labels'] = tf.placeholder(tf.int64, [None, pt_num], 'labels')
pls['voxel_state'] = tf.placeholder(tf.float32, [None, voxel_num],
'voxel_state')
pls['voxel_color'] = tf.placeholder(tf.float32, [None, voxel_num, 3],
'voxel_color')
ops = train_ops(pls['points'], pls['covars'], pls['rpoints'],
pls['labels'], pls['voxel_state'], pls['voxel_color'],
FLAGS.num_classes, voxel_num,
26000 / (FLAGS.batch_size * FLAGS.num_gpus))
feed_dict = {}
config = tf.ConfigProto()
config.gpu_options.allow_growth = True
config.allow_soft_placement = True
config.log_device_placement = False
sess = tf.Session(config=config)
sess.run(tf.global_variables_initializer())
saver = tf.train.Saver(max_to_keep=500)
# saver.restore(sess,'model/voxel_label_sync/unsupervise9.ckpt')
summary_writer = tf.summary.FileWriter(FLAGS.train_dir,
graph=sess.graph)
for epoch_num in xrange(FLAGS.train_epoch_num):
train_one_epoch(ops, pls, sess, summary_writer, train_provider,
epoch_num, feed_dict)
test_one_epoch(ops, pls, sess, saver, test_provider, epoch_num,
feed_dict)
finally:
train_provider.close()
test_provider.close()
def train():
pt_num = 4096
train_list, test_list = get_train_test_split()
train_list += test_list
train_list = [
'data/S3DIS/folding/block_v2/' + fn + '.h5' for fn in train_list
]
test_list = [
'data/S3DIS/folding/block_v2/' + fn + '.h5' for fn in test_list[:5]
]
read_fn = lambda model, fn: read_block_v2(fn)[:2]
train_provider = ProviderV2(train_list, 'train',
FLAGS.batch_size * FLAGS.num_gpus, batch_fn,
read_fn, 2)
test_provider = ProviderV2(test_list, 'test',
FLAGS.batch_size * FLAGS.num_gpus, batch_fn,
read_fn, 2)
try:
pls = {}
pls['points'] = tf.placeholder(tf.float32, [None, pt_num, 3], 'points')
pls['covars'] = tf.placeholder(tf.float32, [None, pt_num, 9], 'covars')
pls['grids'] = tf.placeholder(tf.float32, [4096, 3], 'grids')
ops = train_ops(pls['points'], pls['covars'], pls['grids'],
22000 / (FLAGS.batch_size * FLAGS.num_gpus))
# compute grids
feed_dict = {}
feed_dict[pls['grids']] = generate_grids()
config = tf.ConfigProto()
config.gpu_options.allow_growth = True
config.allow_soft_placement = True
config.log_device_placement = False
sess = tf.Session(config=config)
sess.run(tf.global_variables_initializer())
saver = tf.train.Saver(max_to_keep=500)
summary_writer = tf.summary.FileWriter(FLAGS.train_dir,
graph=sess.graph)
saver.restore(sess, 'model/unsupervise255.ckpt')
for epoch_num in xrange(255, FLAGS.train_epoch_num):
train_one_epoch(ops, pls, sess, summary_writer, train_provider,
epoch_num, feed_dict)
test_one_epoch(ops, pls, sess, saver, test_provider, epoch_num,
feed_dict)
finally:
train_provider.close()
test_provider.close()
def train():
pt_num = 2048
voxel_num = FLAGS.split_num * FLAGS.split_num * FLAGS.split_num
# train_list,test_list=get_train_test_split()
train_list = [
'data/ModelNetTrain/nidxs/ply_data_train{}.h5'.format(i)
for i in xrange(5)
]
train_list += ['data/ModelNetTrain/voxel/ply_data_test_voxel0.h5']
test_list = ['data/ModelNetTrain/voxel/ply_data_test_voxel1.h5']
train_provider = ProviderV2(train_list, 'train',
FLAGS.batch_size * FLAGS.num_gpus, batch_fn,
read_fn, 2)
test_provider = ProviderV2(test_list, 'test',
FLAGS.batch_size * FLAGS.num_gpus, batch_fn,
read_fn, 2)
try:
pls = {}
pls['points'] = tf.placeholder(tf.float32, [None, pt_num, 3], 'points')
pls['covars'] = tf.placeholder(tf.float32, [None, pt_num, 9], 'covars')
pls['labels'] = tf.placeholder(tf.int64, [None], 'labels')
pls['is_training'] = tf.placeholder(tf.bool, name='is_training')
pls['voxel_state'] = tf.placeholder(tf.float32, [None, voxel_num],
'voxel_state')
ops = train_ops(pls['points'], pls['covars'], pls['labels'],
pls['voxel_state'], pls['is_training'],
FLAGS.num_classes, voxel_num,
10000 / (FLAGS.batch_size * FLAGS.num_gpus))
feed_dict = {}
config = tf.ConfigProto()
config.gpu_options.allow_growth = True
config.allow_soft_placement = True
config.log_device_placement = False
sess = tf.Session(config=config)
sess.run(tf.global_variables_initializer())
saver = tf.train.Saver(max_to_keep=500)
summary_writer = tf.summary.FileWriter(FLAGS.train_dir,
graph=sess.graph)
for epoch_num in xrange(FLAGS.train_epoch_num):
train_one_epoch(ops, pls, sess, summary_writer, train_provider,
epoch_num, feed_dict)
test_one_epoch(ops, pls, sess, saver, test_provider, epoch_num,
feed_dict)
finally:
train_provider.close()
test_provider.close()
def test_data_iter():
train_list, test_list = get_train_test_split()
train_list += test_list
train_list = [
'data/S3DIS/folding/block_v2/{}.h5'.format(fn) for fn in train_list
]
# test_list=['data/S3DIS/folding/block_v2/{}.h5'.format(fn) for fn in test_list]
train_provider = ProviderV2(train_list, 'train', 32, batch_fn, read_fn, 2)
begin = time.time()
for data in train_provider:
for item in data:
print item.shape
print 'cost {} s'.format(time.time() - begin)
begin = time.time()
# print data[0].shape,data[1].shape
begin = time.time()
for data in train_provider:
print 'cost {} s'.format(time.time() - begin)
begin = time.time()
# print data[0].shape, data[1].shape
train_provider.close()
def test_data_iter():
train_list, test_list = get_train_test_split()
train_list = [
'data/S3DIS/folding/block_v2/' + fn + '.h5' for fn in train_list
]
read_fn = lambda fn: read_block_v2(fn)[:2]
train_provider = ProviderV2(train_list, 'train', 20, batch_fn, read_fn, 2)
for data in train_provider:
print data[0].shape, data[1].shape
for data in train_provider:
print data[0].shape, data[1].shape
train_provider.close()
def test_data_iter():
train_list, test_list = get_train_test_split()
train_list = [
'data/S3DIS/folding/block_v2/' + fn + '.h5' for fn in train_list
]
# test_list=['data/S3DIS/folding/block_v2/'+fn+'.h5' for fn in test_list]
train_provider = ProviderV2(train_list, 'train',
FLAGS.batch_size * FLAGS.num_gpus, batch_fn,
read_block_v2, 2)
max_label = 0
for data in train_provider:
# print data[0].shape,data[1].shape
max_label = max(np.max(data[3]), max_label)
print max_label
train_provider.close()
def test_data_iter():
train_list = [
'data/ModelNetTrain/nidxs/ply_data_train{}.h5'.format(i)
for i in xrange(5)
]
train_list += ['data/ModelNetTrain/voxel/ply_data_test_voxel0.h5']
test_list = ['data/ModelNetTrain/voxel/ply_data_test_voxel1.h5']
train_provider = ProviderV2(train_list, 'train',
FLAGS.batch_size * FLAGS.num_gpus, batch_fn,
read_fn, 2)
# test_provider = ProviderV2(test_list,'test',FLAGS.batch_size*FLAGS.num_gpus,batch_fn,read_fn,2)
begin = time.time()
for data in train_provider:
print 'cost {} s'.format(time.time() - begin)
begin = time.time()
def test_data_iter():
train_list=['data/ModelNetTrain/nidxs/ply_data_train{}.h5'.format(i) for i in xrange(5)]
test_list=['data/ModelNetTrain/nidxs/ply_data_test{}.h5'.format(i) for i in xrange(2)]
train_provider = ProviderV2(train_list,'test',FLAGS.batch_size*FLAGS.num_gpus,batch_fn,read_fn,2)
# test_provider = ProviderV2(test_list,'test',FLAGS.batch_size*FLAGS.num_gpus,batch_fn,read_fn,2)
begin=time.time()
for data in train_provider:
for item in data:
print item.shape
break
# true_state_pts = voxel2points(data[3][0])
# fn = os.path.join(FLAGS.dump_dir, 'state_true.txt')
# output_points(fn, true_state_pts)
# print np.min(data[0],axis=(0,1))
# print 'cost {} s'.format(time.time()-begin)
# break
train_provider.close()
def eval():
pt_num = 4096
voxel_num = FLAGS.split_num * FLAGS.split_num * FLAGS.split_num
train_list, test_list = get_train_test_split()
test_list = [
'data/S3DIS/folding/block_v2/' + fn + '.h5' for fn in test_list
]
test_provider = ProviderV2(test_list, 'test',
FLAGS.batch_size * FLAGS.num_gpus, batch_fn,
read_block_v2, 2)
try:
pls = {}
pls['points'] = tf.placeholder(tf.float32, [None, pt_num, 6], 'points')
pls['covars'] = tf.placeholder(tf.float32, [None, pt_num, 9], 'covars')
pls['rpoints'] = tf.placeholder(tf.float32, [None, pt_num, 3],
'rpoints')
pls['labels'] = tf.placeholder(tf.int64, [None, pt_num], 'labels')
pls['voxel_state'] = tf.placeholder(tf.float32, [None, voxel_num],
'voxel_state')
pls['voxel_color'] = tf.placeholder(tf.float32, [None, voxel_num, 3],
'voxel_color')
ops = train_ops(pls['points'], pls['covars'], pls['rpoints'],
pls['labels'], pls['voxel_state'], pls['voxel_color'],
FLAGS.num_classes, voxel_num,
22000 / (FLAGS.batch_size * FLAGS.num_gpus))
feed_dict = {}
config = tf.ConfigProto()
config.gpu_options.allow_growth = True
config.allow_soft_placement = True
config.log_device_placement = False
sess = tf.Session(config=config)
saver = tf.train.Saver(max_to_keep=500)
saver.restore(sess, FLAGS.eval_model)
test_one_epoch(ops, pls, sess, saver, test_provider, 0, feed_dict)
finally:
test_provider.close()
def eval():
pt_num = 2048
# train_list,test_list=get_train_test_split()
# train_list = ['data/ModelNetTrain/nidxs/ply_data_train{}.h5'.format(i) for i in xrange(5)]
test_list = ['data/ModelNetTrain/nidxs/ply_data_test{}.h5'.format(i) for i in xrange(2)]
# train_provider = ProviderV2(train_list, 'train', FLAGS.batch_size * FLAGS.num_gpus, batch_fn, read_fn, 2)
test_provider = ProviderV2(test_list, 'test', FLAGS.batch_size * FLAGS.num_gpus, read_fn)
try:
pls = {}
pls['points'] = tf.placeholder(tf.float32, [None, pt_num, 3], 'points')
pls['covars'] = tf.placeholder(tf.float32, [None, pt_num, 9], 'covars')
pls['labels'] = tf.placeholder(tf.int64, [None], 'labels')
pls['is_training'] = tf.placeholder(tf.bool, name='is_training')
ops = train_ops(pls['points'], pls['covars'],
pls['labels'], pls['is_training'], FLAGS.num_classes,
8000 / (FLAGS.batch_size * FLAGS.num_gpus))
feed_dict = {}
config = tf.ConfigProto()
config.gpu_options.allow_growth = True
config.allow_soft_placement = True
config.log_device_placement = False
sess = tf.Session(config=config)
sess.run(tf.global_variables_initializer())
saver = tf.train.Saver(max_to_keep=500)
saver.restore(sess,FLAGS.eval_model)
test_one_epoch(ops, pls, sess, saver, test_provider, 0, feed_dict)
finally:
# train_provider.close()
test_provider.close()
def test_points2voxel():
import Points2Voxel
from provider import ProviderV2, ProviderV3
from s3dis.block_util import read_block_v2
from s3dis.data_util import get_train_test_split
from s3dis.voxel_util import points2voxel_color_gpu
from s3dis.voxel_util import point2voxel
from s3dis.draw_util import output_points
def read_fn(filename):
points, covars = read_block_v2(filename)[:2]
voxel_state, voxel_color = points2voxel_color_gpu(points, 30)
return points, covars, voxel_state, voxel_color
def batch_fn(file_data, cur_idx, data_indices, require_size):
points, covars, voxel_state, voxel_color = file_data
end_idx = min(cur_idx + require_size, points.shape[0])
return [
points[data_indices[cur_idx:end_idx], :, :],
covars[data_indices[cur_idx:end_idx], :, :],
voxel_state[data_indices[cur_idx:end_idx], :],
voxel_color[data_indices[cur_idx:end_idx], :, :]
], end_idx - cur_idx
train_list, _ = get_train_test_split()
train_list = [
'data/S3DIS/folding/block_v2/{}.h5'.format(fn) for fn in train_list
]
train_provider = ProviderV2(train_list, 'train', 32, batch_fn, read_fn, 2)
begin = time.time()
total_begin = time.time()
for data in train_provider:
# for pts_i,pts in enumerate(data[0][:2]):
# print np.min(pts,axis=0)
# pts[:,3:]+=1.0
# pts[:,3:]*=128
# output_points('original{}.txt'.format(pts_i),pts)
#
# for vi,v in enumerate(data[2][:2]):
# vpts=voxel2points(v)
# vpts[:,:3]/=np.max(vpts[:,:3],axis=0,keepdims=True)
# print np.min(vpts,axis=0)
# output_points('voxels{}.txt'.format(vi),vpts)
#
# for vi,v in enumerate(data[3][:2]):
# vpts=voxel2points(v)
# vpts[:,:3]/=np.max(vpts[:,:3],axis=0,keepdims=True)
# print np.min(vpts,axis=0)
# output_points('voxels_color{}.txt'.format(vi),vpts)
# time.sleep(0.1)
print 'cost {} s'.format(time.time() - begin)
begin = time.time()
# train_provider.close()
# exit(0)
print 'total cost {} s'.format(time.time() - total_begin)
train_provider.close()
print ' exit '
exit(0)