first commit

modules/pointnet_encoder.py

@@ -0,0 +1,107 @@
+from __future__ import print_function
+import torch
+import torch.nn as nn
+import torch.nn.parallel
+import torch.utils.data
+from torch.autograd import Variable
+import numpy as np
+import torch.nn.functional as F
+
+import PytorchBoot.stereotype as stereotype
+@stereotype.module("pointnet_encoder")
+class PointNetEncoder(nn.Module):
+
+    def __init__(self, config:dict):
+        super(PointNetEncoder, self).__init__()
+        
+        self.out_dim = config["out_dim"]
+        self.in_dim = config["in_dim"]
+        self.feature_transform = config.get("feature_transform", False)
+        self.stn = STNkd(k=self.in_dim)
+        self.conv1 = torch.nn.Conv1d(self.in_dim , 64, 1)
+        self.conv2 = torch.nn.Conv1d(64, 128, 1)
+        self.conv3 = torch.nn.Conv1d(128, 512, 1)
+        self.conv4 = torch.nn.Conv1d(512, self.out_dim , 1)
+        if self.feature_transform:
+            self.f_stn = STNkd(k=64)
+
+    def forward(self, x):
+        trans = self.stn(x)
+        x = x.transpose(2, 1)
+        x = torch.bmm(x, trans)
+        x = x.transpose(2, 1)
+        x = F.relu(self.conv1(x))
+
+        if self.feature_transform:
+            trans_feat = self.f_stn(x)
+            x = x.transpose(2, 1)
+            x = torch.bmm(x, trans_feat)
+            x = x.transpose(2, 1)
+
+        point_feat = x
+        x = F.relu(self.conv2(x))
+        x = F.relu(self.conv3(x))
+        x = self.conv4(x)
+        x = torch.max(x, 2, keepdim=True)[0]
+        x = x.view(-1, self.out_dim)
+        return x, point_feat
+
+    def encode_points(self, pts, require_per_point_feat=False):
+        pts = pts.transpose(2, 1)
+        global_pts_feature, per_point_feature = self(pts)
+        if require_per_point_feat:
+            return global_pts_feature, per_point_feature.transpose(2, 1)
+        else:
+            return global_pts_feature
+
+class STNkd(nn.Module):
+    def __init__(self, k=64):
+        super(STNkd, self).__init__()
+        self.conv1 = torch.nn.Conv1d(k, 64, 1)
+        self.conv2 = torch.nn.Conv1d(64, 128, 1)
+        self.conv3 = torch.nn.Conv1d(128, 1024, 1)
+        self.fc1 = nn.Linear(1024, 512)
+        self.fc2 = nn.Linear(512, 256)
+        self.fc3 = nn.Linear(256, k * k)
+        self.relu = nn.ReLU()
+
+        self.k = k
+
+    def forward(self, x):
+        batchsize = x.size()[0]
+        x = F.relu(self.conv1(x))
+        x = F.relu(self.conv2(x))
+        x = F.relu(self.conv3(x))
+        x = torch.max(x, 2, keepdim=True)[0]
+        x = x.view(-1, 1024)
+
+        x = F.relu(self.fc1(x))
+        x = F.relu(self.fc2(x))
+        x = self.fc3(x)
+
+        iden = (
+            Variable(torch.from_numpy(np.eye(self.k).flatten().astype(np.float32)))
+            .view(1, self.k * self.k)
+            .repeat(batchsize, 1)
+        )
+        if x.is_cuda:
+            iden = iden.to(x.get_device())
+        x = x + iden
+        x = x.view(-1, self.k, self.k)
+        return x
+
+if __name__ == "__main__":
+    sim_data = Variable(torch.rand(32, 2500, 3))
+    config = {
+        "in_dim": 3,
+        "out_dim": 1024,
+        "feature_transform": False
+    }
+    pointnet = PointNetEncoder(config)
+    out = pointnet.encode_points(sim_data)
+
+    print("global feat", out.size())
+
+    out, per_point_out = pointnet.encode_points(sim_data, require_per_point_feat=True)
+    print("point feat", out.size())
+    print("per point feat", per_point_out.size())