Empty module name at to_hls.InferStreamingMaxPool operation

[whitelambs]

Dear FINN team,

I have a custom CNV model with different dimension and input_size.
`import torch
from torch.nn import Module, ModuleList, BatchNorm2d, MaxPool2d, BatchNorm1d

from brevitas.nn import QuantConv2d, QuantIdentity, QuantLinear
from brevitas.core.restrict_val import RestrictValueType
from .tensor_norm import TensorNorm
from .common import CommonWeightQuant, CommonActQuant

CNV_OUT_CH_POOL = [(64, True), (128, True), (256, False), (256, True), (512, False), (512, True)]
INTERMEDIATE_FC_FEATURES = [(51200, 512), (512, 512)]
LAST_FC_IN_FEATURES = 512
LAST_FC_PER_OUT_CH_SCALING = False
POOL_SIZE = 2
KERNEL_SIZE = 3

class CNV(Module):

def __init__(self, num_classes, weight_bit_width, act_bit_width, in_bit_width, in_ch):
    super(CNV, self).__init__()

    self.conv_features = ModuleList()
    self.linear_features = ModuleList()

    self.conv_features.append(QuantIdentity( # for Q1.7 input format
        act_quant=CommonActQuant,
        bit_width=in_bit_width,
        min_val=- 1.0,
        max_val=1.0 - 2.0 ** (-7),
        narrow_range=False,
        restrict_scaling_type=RestrictValueType.POWER_OF_TWO))

    for out_ch, is_pool_enabled in CNV_OUT_CH_POOL:
        self.conv_features.append(QuantConv2d(
            kernel_size=KERNEL_SIZE,
            in_channels=in_ch,
            out_channels=out_ch,
            bias=False,
            weight_quant=CommonWeightQuant,
            weight_bit_width=weight_bit_width))
        in_ch = out_ch
        self.conv_features.append(BatchNorm2d(in_ch, eps=1e-4))
        self.conv_features.append(QuantIdentity(
            act_quant=CommonActQuant,
            bit_width=act_bit_width))
        if is_pool_enabled:
            self.conv_features.append(MaxPool2d(kernel_size=2))

    for in_features, out_features in INTERMEDIATE_FC_FEATURES:
        self.linear_features.append(QuantLinear(
            in_features=in_features,
            out_features=out_features,
            bias=False,
            weight_quant=CommonWeightQuant,
            weight_bit_width=weight_bit_width))
        self.linear_features.append(BatchNorm1d(out_features, eps=1e-4))
        self.linear_features.append(QuantIdentity(
            act_quant=CommonActQuant,
            bit_width=act_bit_width))

    self.linear_features.append(QuantLinear(
        in_features=LAST_FC_IN_FEATURES,
        out_features=num_classes,
        bias=False,
        weight_quant=CommonWeightQuant,
        weight_bit_width=weight_bit_width))
    self.linear_features.append(TensorNorm())
    
    for m in self.modules():
      if isinstance(m, QuantConv2d) or isinstance(m, QuantLinear):
        torch.nn.init.uniform_(m.weight.data, -1, 1)


def clip_weights(self, min_val, max_val):
    for mod in self.conv_features:
        if isinstance(mod, QuantConv2d):
            mod.weight.data.clamp_(min_val, max_val)
    for mod in self.linear_features:
        if isinstance(mod, QuantLinear):
            mod.weight.data.clamp_(min_val, max_val)

def forward(self, x):
    x = 2.0 * x - torch.tensor([1.0], device=x.device)
    for mod in self.conv_features:
        x = mod(x)
    x = x.view(x.shape[0], -1)
    for mod in self.linear_features:
        x = mod(x)
    return x

def cnv(cfg):
weight_bit_width = cfg.getint('QUANT', 'WEIGHT_BIT_WIDTH')
act_bit_width = cfg.getint('QUANT', 'ACT_BIT_WIDTH')
in_bit_width = cfg.getint('QUANT', 'IN_BIT_WIDTH')
num_classes = cfg.getint('MODEL', 'NUM_CLASSES')
in_channels = cfg.getint('MODEL', 'IN_CHANNELS')
net = CNV(weight_bit_width=weight_bit_width,
act_bit_width=act_bit_width,
in_bit_width=in_bit_width,
num_classes=num_classes,
in_ch=in_channels)
return net`

I've trained my model mainly using https://github.com/Xilinx/brevitas/tree/master/src/brevitas_examples/bnn_pynq.

Then, I exported my model using FINNManager.export. During exportation, I received these warnings below.

After that, I've used same steps in end2end_example/bnn-pynq/cnv_end2end_example to build CNV using FINN. After streamline operation, my nodes was not correct like in the picture above.

Therefore I've removed extra transpose layers and apply MakeMaxPoolNHWC again. As a result, I've received this onnx model.

I've started to apply Maxpool operation, but I'm constantly receiving this error.

Thanks in advance for your help!