86 lines
3.2 KiB
Python
86 lines
3.2 KiB
Python
from argparse import Namespace
|
|
|
|
import torch.nn.functional as F
|
|
|
|
import torch
|
|
from torch import nn
|
|
|
|
from datasets.full_pointclouds import FullCloudsDataset
|
|
from ml_lib.modules.geometric_blocks import SAModule, GlobalSAModule, MLP, FPModule
|
|
from ml_lib.modules.util import LightningBaseModule, F_x
|
|
|
|
from utils.module_mixins import BaseValMixin, BaseTrainMixin, BaseOptimizerMixin, BaseDataloadersMixin, DatasetMixin
|
|
from utils.project_config import GlobalVar
|
|
|
|
|
|
class PointNet2(BaseValMixin,
|
|
BaseTrainMixin,
|
|
BaseOptimizerMixin,
|
|
DatasetMixin,
|
|
BaseDataloadersMixin,
|
|
LightningBaseModule
|
|
):
|
|
|
|
def __init__(self, hparams):
|
|
super(PointNet2, self).__init__(hparams=hparams)
|
|
|
|
# Dataset
|
|
# =============================================================================
|
|
self.dataset = self.build_dataset(FullCloudsDataset)
|
|
|
|
# Model Paramters
|
|
# =============================================================================
|
|
# Additional parameters
|
|
|
|
# Modules
|
|
self.sa1_module = SAModule(0.2, 0.2, MLP([3 + 3, 64, 64, 128]))
|
|
self.sa2_module = SAModule(0.25, 0.4, MLP([128 + 3, 128, 128, 256]))
|
|
self.sa3_module = GlobalSAModule(MLP([256 + 3, 256, 512, 1024]))
|
|
|
|
self.fp3_module = FPModule(1, MLP([1024 + 256, 256, 256]))
|
|
self.fp2_module = FPModule(3, MLP([256 + 128, 256, 128]))
|
|
self.fp1_module = FPModule(3, MLP([128 + 3, 128, 128, 128]))
|
|
|
|
self.lin1 = torch.nn.Linear(128, 128)
|
|
self.lin2 = torch.nn.Linear(128, 128)
|
|
self.lin3 = torch.nn.Linear(128, len(GlobalVar.classes))
|
|
|
|
# Utility
|
|
self.dropout = nn.Dropout(self.params.dropout) if self.params.dropout else F_x(None)
|
|
self.activation = self.params.activation()
|
|
self.log_softmax = nn.LogSoftmax(dim=-1)
|
|
|
|
def forward(self, data, **kwargs):
|
|
"""
|
|
data: a batch of input torch_geometric.data.Data type
|
|
- torch_geometric.data.Data, as torch_geometric batch input:
|
|
data.x: (batch_size * ~num_points, C), batch nodes/points feature,
|
|
~num_points means each sample can have different number of points/nodes
|
|
|
|
data.pos: (batch_size * ~num_points, 3)
|
|
|
|
data.batch: (batch_size * ~num_points,), a column vector of graph/pointcloud
|
|
idendifiers for all nodes of all graphs/pointclouds in the batch. See
|
|
pytorch_gemometric documentation for more information
|
|
"""
|
|
|
|
sa0_out = (data.x, data.pos, data.batch)
|
|
sa1_out = self.sa1_module(*sa0_out)
|
|
sa2_out = self.sa2_module(*sa1_out)
|
|
sa3_out = self.sa3_module(*sa2_out)
|
|
|
|
fp3_out = self.fp3_module(*sa3_out, *sa2_out)
|
|
fp2_out = self.fp2_module(*fp3_out, *sa1_out)
|
|
tensor, _, _ = self.fp1_module(*fp2_out, *sa0_out)
|
|
|
|
tensor = tensor.float()
|
|
|
|
tensor = self.activation(tensor)
|
|
tensor = self.lin1(tensor)
|
|
tensor = self.dropout(tensor)
|
|
tensor = self.lin2(tensor)
|
|
tensor = self.dropout(tensor)
|
|
tensor = self.lin3(tensor)
|
|
tensor = self.log_softmax(tensor)
|
|
return Namespace(main_out=tensor)
|