def __init__(self, num_class, num_heading_bin, num_size_cluster, mean_size_arr,
input_feature_dim=0, num_proposal=128, vote_factor=1, sampling='vote_fps', scale=1):
super().__init__()
self.num_class = num_class
self.num_heading_bin = num_heading_bin
self.num_size_cluster = num_size_cluster
self.mean_size_arr = mean_size_arr
assert(mean_size_arr.shape[0] == self.num_size_cluster)
self.input_feature_dim = input_feature_dim
self.num_proposal = num_proposal
self.vote_factor = vote_factor
self.sampling=sampling
# Backbone point feature learning
self.backbone_net = Pointnet2Backbone(input_feature_dim=self.input_feature_dim, SCALE=scale)
# Hough voting
if scale <= 2:
self.vgen = VotingModule(self.vote_factor, 512)
else:
self.vgen = VotingModule(self.vote_factor, 256*scale)
# Vote aggregation and detection
if scale <= 2:
self.pnet = ProposalModule(num_class, num_heading_bin, num_size_cluster,
mean_size_arr, num_proposal, sampling, seed_feat_dim=512)
else:
self.pnet = ProposalModule(num_class, num_heading_bin, num_size_cluster,
mean_size_arr, num_proposal, sampling, seed_feat_dim=256*scale)
def __init__(self,
num_class,
num_heading_bin,
num_size_cluster,
mean_size_arr,
input_feature_dim=0,
num_proposal=128,
vote_factor=1,
sampling='vote_fps'):
super().__init__()
self.num_class = num_class
self.num_heading_bin = num_heading_bin
self.num_size_cluster = num_size_cluster
self.mean_size_arr = mean_size_arr
assert (mean_size_arr.shape[0] == self.num_size_cluster)
self.input_feature_dim = input_feature_dim
self.num_proposal = num_proposal
self.vote_factor = vote_factor
self.sampling = sampling
# Backbone point feature learning
self.backbone_net = Pointnet2Backbone(
input_feature_dim=self.input_feature_dim)
# Box proposal, aggregation and detection
self.pnet = ProposalModule(num_class, num_heading_bin,
num_size_cluster, mean_size_arr,
num_proposal, sampling)
def __init__(self,
num_class,
input_feature_dim=0,
num_proposal=128,
vote_factor=1,
sampling='vote_fps'):
super().__init__()
self.num_class = num_class
self.input_feature_dim = input_feature_dim
self.num_proposal = num_proposal
self.vote_factor = vote_factor
self.sampling = sampling
self.backbone_net = Pointnet2Backbone(
input_feature_dim=self.input_feature_dim)
self.vgen = VotingModule(self.vote_factor, 256)
self.pnet = ProposalModule(num_class, num_proposal, sampling)
def __init__(self,
num_class,
num_heading_bin,
num_size_cluster,
mean_size_arr,
input_feature_dim=0,
num_proposal=128,
vote_factor=1,
sampling='vote_fps',
backbone='pointnet2'):
super().__init__()
self.num_class = num_class
self.num_heading_bin = num_heading_bin
self.num_size_cluster = num_size_cluster
self.mean_size_arr = mean_size_arr
assert (mean_size_arr.shape[0] == self.num_size_cluster)
self.input_feature_dim = input_feature_dim
self.num_proposal = num_proposal
self.vote_factor = vote_factor
self.sampling = sampling
self.backbone = backbone
# Backbone point feature learning
if backbone == 'pointnet2':
self.backbone_net = Pointnet2Backbone(
input_feature_dim=self.input_feature_dim)
else:
self.backbone_net = SparseConvBackbone(
input_feature_dim=self.input_feature_dim + 3,
output_feature_dim=256,
num_seed=1024)
# from pdb import set_trace; set_trace()
# Hough voting
self.vgen = VotingModule(self.vote_factor, 256)
# Vote aggregation and detection
self.pnet = ProposalModule(num_class, num_heading_bin,
num_size_cluster, mean_size_arr,
num_proposal, sampling)
def __init__(self, num_class, num_angle_bin, num_viewpoint,
input_feature_dim=0, num_proposal=128, vote_factor=10, sampling='vote_fps'):
super().__init__()
self.num_class = num_class
self.num_angle_bin = num_angle_bin
self.num_viewpoint = num_viewpoint
self.input_feature_dim = input_feature_dim
self.num_proposal = num_proposal
self.vote_factor = vote_factor
self.sampling=sampling
# Backbone point feature learning
self.backbone_net = Pointnet2Backbone(input_feature_dim=self.input_feature_dim)
# Hough voting
self.vgen = VotingModule(self.vote_factor, 256)
# Vote aggregation and detection
self.pnet = ProposalModule(num_class, num_angle_bin, num_viewpoint, num_proposal, sampling)
def __init__(self,
num_class,
num_heading_bin,
num_size_cluster,
mean_size_arr,
input_feature_dim=0,
num_proposal=128,
vote_factor=1,
sampling='vote_fps',
max_imvote_per_pixel=3,
image_feature_dim=18,
image_hidden_dim=256):
super().__init__()
self.num_class = num_class
self.num_heading_bin = num_heading_bin
self.num_size_cluster = num_size_cluster
self.mean_size_arr = mean_size_arr
assert (mean_size_arr.shape[0] == self.num_size_cluster)
self.input_feature_dim = input_feature_dim
self.num_proposal = num_proposal
self.vote_factor = vote_factor
self.sampling = sampling
self.max_imvote_per_pixel = max_imvote_per_pixel
self.image_feature_dim = image_feature_dim
# Backbone point feature learning
self.backbone_net = Pointnet2Backbone(
input_feature_dim=self.input_feature_dim)
# Image feature extractor
self.image_feature_extractor = ImageFeatureModule(
max_imvote_per_pixel=self.max_imvote_per_pixel)
# MLP on image features before fusing with point features
self.image_mlp = ImageMLPModule(image_feature_dim,
image_hidden_dim=image_hidden_dim)
# Hough voting modules
self.img_only_vgen = VotingModule(self.vote_factor, image_hidden_dim)
self.pc_only_vgen = VotingModule(self.vote_factor, 256)
self.pc_img_vgen = VotingModule(self.vote_factor,
image_hidden_dim + 256)
# Vote aggregation and detection
self.img_only_pnet = ProposalModule(num_class,
num_heading_bin,
num_size_cluster,
mean_size_arr,
num_proposal,
sampling,
seed_feat_dim=image_hidden_dim,
key_prefix='img_only_')
self.pc_only_pnet = ProposalModule(num_class,
num_heading_bin,
num_size_cluster,
mean_size_arr,
num_proposal,
sampling,
seed_feat_dim=256,
key_prefix='pc_only_')
self.pc_img_pnet = ProposalModule(num_class,
num_heading_bin,
num_size_cluster,
mean_size_arr,
num_proposal,
sampling,
seed_feat_dim=image_hidden_dim + 256,
key_prefix='pc_img_')
