英文:
Why am I getting a TypeError during ML Model initialisation?
问题
错误信息中显示了一个问题：mlp2 参数被设置为 None，但你不知道如何解决它。根据你提供的代码，这个参数确实被设置为 None，但这似乎是在代码中有意为之的。
在你的 PointNetSetAbstraction 类中，有这个条件：
for out_channel in mlp2:self.mlp2_convs.append(nn.Sequential(nn.Conv1d(last_channel, out_channel, 1, bias=False),nn.BatchNorm1d(out_channel)))last_channel = out_channel
这段代码只有在 mlp2 不为 None 时才会执行，否则它不会添加任何层。因此，将 mlp2 参数设置为 None 是合理的，只要在使用这个类时确保你不传递任何 mlp2 参数即可。
如果你想为 mlp2 参数传递一个非空的列表，你需要在创建 PointNetSetAbstraction 的实例时将其传递给该类的构造函数。如果你需要进一步的帮助或有其他问题，请提出。
英文:
I am running a deep learning project at https://github.com/hyangwinter/flownet3d_pytorch. There are some problems during operation, and I don't know how to solve them.
The error is as follows：
Namespace(batch_size=64, cycle=False, dataset='SceneflowDataset', dataset_path='data/data_processed_maxcut_35_20k_2k_8192/', dropout=0.5, emb_dims=512, epochs=250, eval=False, exp_name='flownet3d', gaussian_noise=False, lr=0.001, model='flownet', model_path='', momentum=0.9, no_cuda=False, num_points=2048, seed=1234, test_batch_size=32, unseen=False, use_sgd=False)train : 20006test : 2007Traceback (most recent call last):File "main.py", line 282, in <module>main()File "main.py", line 254, in mainnet = FlowNet3D(args).cuda()File "/home/ubuntu/project/flownet3d_pytorch/model.py", line 13, in __init__self.sa1 = PointNetSetAbstraction(npoint=1024, radius=0.5, nsample=16, in_channel=3, mlp=[32,32,64], group_all=False)File "/home/ubuntu/project/flownet3d_pytorch/util.py", line 225, in __init__for out_channel in mlp2:TypeError: 'NoneType' object is not iterable`
My code:
class FlowNet3D(nn.Module):def __init__(self,args):super(FlowNet3D,self).__init__()self.sa1 = PointNetSetAbstraction(npoint=1024, radius=0.5, nsample=16, in_channel=3, mlp=[32,32,64], group_all=False)self.sa2 = PointNetSetAbstraction(npoint=256, radius=1.0, nsample=16, in_channel=64, mlp=[64, 64, 128], group_all=False)self.sa3 = PointNetSetAbstraction(npoint=64, radius=2.0, nsample=8, in_channel=128, mlp=[128, 128, 256], group_all=False)self.sa4 = PointNetSetAbstraction(npoint=16, radius=4.0, nsample=8, in_channel=256, mlp=[256,256,512], group_all=False)self.fe_layer = FlowEmbedding(radius=10.0, nsample=64, in_channel = 128, mlp=[128, 128, 128], pooling='max', corr_func='concat')self.su1 = PointNetSetUpConv(nsample=8, radius=2.4, f1_channel = 256, f2_channel = 512, mlp=[], mlp2=[256, 256])self.su2 = PointNetSetUpConv(nsample=8, radius=1.2, f1_channel = 128+128, f2_channel = 256, mlp=[128, 128, 256], mlp2=[256])self.su3 = PointNetSetUpConv(nsample=8, radius=0.6, f1_channel = 64, f2_channel = 256, mlp=[128, 128, 256], mlp2=[256])self.fp = PointNetFeaturePropogation(in_channel = 256+3, mlp = [256, 256])self.conv1 = nn.Conv1d(256, 128, kernel_size=1, bias=False)self.bn1 = nn.BatchNorm1d(128)self.conv2=nn.Conv1d(128, 3, kernel_size=1, bias=True)def forward(self, pc1, pc2, feature1, feature2):l1_pc1, l1_feature1 = self.sa1(pc1, feature1)l2_pc1, l2_feature1 = self.sa2(l1_pc1, l1_feature1)l1_pc2, l1_feature2 = self.sa1(pc2, feature2)l2_pc2, l2_feature2 = self.sa2(l1_pc2, l1_feature2)_, l2_feature1_new = self.fe_layer(l2_pc1, l2_pc2, l2_feature1, l2_feature2)l3_pc1, l3_feature1 = self.sa3(l2_pc1, l2_feature1_new)l4_pc1, l4_feature1 = self.sa4(l3_pc1, l3_feature1)l3_fnew1 = self.su1(l3_pc1, l4_pc1, l3_feature1, l4_feature1)l2_fnew1 = self.su2(l2_pc1, l3_pc1, torch.cat([l2_feature1, l2_feature1_new], dim=1), l3_fnew1)l1_fnew1 = self.su3(l1_pc1, l2_pc1, l1_feature1, l2_fnew1)l0_fnew1 = self.fp(pc1, l1_pc1, feature1, l1_fnew1)x = F.relu(self.bn1(self.conv1(l0_fnew1)))sf = self.conv2(x)return sfclass PointNetSetAbstraction(nn.Module):def __init__(self, npoint, radius, nsample, in_channel, mlp, mlp2 = None, group_all = False):super(PointNetSetAbstraction, self).__init__()self.npoint = npointself.radius = radiusself.nsample = nsampleself.group_all = group_allself.mlp_convs = nn.ModuleList()self.mlp_bns = nn.ModuleList()self.mlp2_convs = nn.ModuleList()last_channel = in_channel+3for out_channel in mlp:self.mlp_convs.append(nn.Conv2d(last_channel, out_channel, 1, bias = False))self.mlp_bns.append(nn.BatchNorm2d(out_channel))last_channel = out_channelfor out_channel in mlp2:self.mlp2_convs.append(nn.Sequential(nn.Conv1d(last_channel, out_channel, 1, bias=False),nn.BatchNorm1d(out_channel)))last_channel = out_channelif group_all:self.queryandgroup = pointutils.GroupAll()else:self.queryandgroup = pointutils.QueryAndGroup(radius, nsample)def forward(self, xyz, points):"""Input:xyz: input points position data, [B, C, N]points: input points data, [B, D, N]Return:new_xyz: sampled points position data, [B, S, C]new_points_concat: sample points feature data, [B, S, D']"""device = xyz.deviceB, C, N = xyz.shapexyz_t = xyz.permute(0, 2, 1).contiguous()# if points is not None:# points = points.permute(0, 2, 1).contiguous()if self.group_all == False:fps_idx = pointutils.furthest_point_sample(xyz_t, self.npoint) # [B, N]new_xyz = pointutils.gather_operation(xyz, fps_idx) # [B, C, N]else:new_xyz = xyznew_points = self.queryandgroup(xyz_t, new_xyz.transpose(2, 1).contiguous(), points) # [B, 3+C, N, S]# new_xyz: sampled points position data, [B, C, npoint]# new_points: sampled points data, [B, C+D, npoint, nsample]for i, conv in enumerate(self.mlp_convs):bn = self.mlp_bns[i]new_points = F.relu(bn(conv(new_points)))new_points = torch.max(new_points, -1)[0]for i, conv in enumerate(self.mlp2_convs):new_points = F.relu(conv(new_points))return new_xyz, new_points`
I noticed that the mlp2 parameter is set to NONE but I don't know how to solve it. In the source code implementation, this parameter is also NONE
答案1
得分: 1
由于您的 self.sa1 没有名为 mlp2 的参数，它将使用默认值，即您在以下代码中定义的 None：
class PointNetSetAbstraction(nn.Module):def __init__(self, npoint, radius, nsample, in_channel, mlp, mlp2=None, group_all=False):
当然，None 不能迭代，为了解决这个问题，您可以在 for 循环之前添加一个 if 条件。另外，您也可以像您已经为 self.su1 做的那样传递 mlp2=[]。
英文:
Since your self.sa1 do not have parameters named mlp2, it will use the default one, None as you defined in
class PointNetSetAbstraction(nn.Module):def init(self, npoint, radius, nsample, in_channel, mlp, mlp2 = None, group_all = False):
And of cause None is not iterable, to solve this, you can just add a if-condition before the for-loop. Otherwise, you can also just pass mlp2=[] as you already done for self.su1.
答案2
得分: 0
如您在问题中指出的那样，在函数调用中mlp2被设置为None。尝试迭代mlp2时：
for out_channel in mlp2:
代码正确地引发了TypeError。
很难理解您试图做什么，特别是因为mlp2不是一个好的变量名（它应该是什么？）。根据您的代码上下文，我会假设mlp2是一个包含您层大小的列表。您可以尝试类似以下的内容：
mlp2 = [16, 16, 16, 16]
这将初始化模型以具有大小为16的层（如果代码的其余部分正常工作的话）。
英文:
As you pointed out in your question, mlp2 is set to None in the function call. When trying to iterate over mlp2 with
for out_channel in mlp2:
the code corectly throws a TypeError.
It is hard to understand what you are trying to do, especially because mlp2 is not a good variable name (what even should that be?). From the context of your code I would assume that mpl2 is a list that holds the sizes of your layers. You might try something like:
mlp2 = [16,16,16,16]
This will initialize the Model to have layers of size 16 (if the rest of the code is working).
通过集体智慧和协作来改善编程学习和解决问题的方式。致力于成为全球开发者共同参与的知识库，让每个人都能够通过互相帮助和分享经验来进步。
评论