在VGG卷积块中添加CBAM注意力模块，结果优化器报错，优化器代码如下：
def cbam(ipt,num_channels,reduction):
#通道注意力
avg_pool = fluid.layers.pool2d(
input=ipt,
pool_size=1,
pool_type='avg',
)
fc_1 = fluid.layers.conv2d(input=avg_pool,num_filters=num_channels//reduction, filter_size=1, stride=1, padding=0, act=None)
avg = fluid.layers.relu(fc_1)
avg = fluid.layers.conv2d(input=avg,num_filters=num_channels, filter_size=1, stride=1, padding=0, act=None)
max_pool = fluid.layers.pool2d(
input=ipt,
pool_size=1,
pool_type='max'
)
fc1_2 = fluid.layers.conv2d(input=max_pool,num_filters=num_channels//reduction, filter_size=1, stride=1, padding=0, act=None)
mx = fluid.layers.relu(fc1_2)
mx = fluid.layers.conv2d(input=mx,num_filters=num_channels, filter_size=1, stride=1, padding=0, act=None)
out1 = avg+mx
out1 = fluid.layers.sigmoid(out1)
#空间注意力
ipt2 = ipt*out1
avg = fluid.layers.reduce_mean(ipt2,1,keep_dim=True)
mx = fluid.layers.reduce_max(ipt2,1,keep_dim=True)
#print(avg.shape, mx.shape)
out2 = fluid.layers.concat([avg, mx], axis=1)
#print(out2.shape)
out2 = fluid.layers.conv2d(input=out2,num_filters=1, filter_size=7, stride=1, padding=3, act=None)
out2 = fluid.layers.sigmoid(out2)
#print(ipt2.shape, out2.shape)
out3 = ipt2*out2
return out3
报错信息：375 #optimizer = fluid.optimizer.MomentumOptimizer(learning_rate=0.001, momentum=0.9)
376 #optimizer = fluid.optimizer.AdagradOptimizer(0.001)
--> 377 opts = optimizer.minimize(avg_cost)
378 DATADIR = '/home/aistudio/dataset/train'
379 test_path = 'dataset/val'
in minimize(self, loss, startup_program, parameter_list, no_grad_set)
/opt/conda/envs/python35-paddle120-env/lib/python3.7/site-packages/paddle/fluid/dygraph/base.py in __impl__(func, *args, **kwargs)
201 def __impl__(func, *args, **kwargs):
202 with _switch_tracer_mode_guard_(is_train=False):
--> 203 return func(*args, **kwargs)
204
205 return __impl__(func)
/opt/conda/envs/python35-paddle120-env/lib/python3.7/site-packages/paddle/fluid/optimizer.py in minimize(self, loss, startup_program, parameter_list, no_grad_set)
832 startup_program=startup_program,
833 parameter_list=parameter_list,
--> 834 no_grad_set=no_grad_set)
835
836 optimize_ops = self.apply_optimize(
/opt/conda/envs/python35-paddle120-env/lib/python3.7/site-packages/paddle/fluid/optimizer.py in backward(self, loss, startup_program, parameter_list, no_grad_set, callbacks)
675 with program_guard(program, startup_program):
676 params_grads = append_backward(loss, parameter_list,
--> 677 act_no_grad_set, callbacks)
678 # Note: since we can't use all_reduce_op now,
679 # dgc_op should be the last op of one grad.
/opt/conda/envs/python35-paddle120-env/lib/python3.7/site-packages/paddle/fluid/backward.py in append_backward(loss, parameter_list, no_grad_set, callbacks, checkpoints)
1417
1418 program.current_block_idx = current_block_idx
-> 1419 program._sync_with_cpp()
1420
1421 if parameter_list is not None:
/opt/conda/envs/python35-paddle120-env/lib/python3.7/site-packages/paddle/fluid/framework.py in _sync_with_cpp(self)
4555 self.blocks.append(Block(self, block_idx))
4556 for block in self.blocks:
-> 4557 block._sync_with_cpp()
4558
4559 def _copy_param_info_from(self, other):
/opt/conda/envs/python35-paddle120-env/lib/python3.7/site-packages/paddle/fluid/framework.py in _sync_with_cpp(self)
2744 ops_in_python_index += 1
2745
-> 2746 assert len(self.ops) == len(ops_in_cpp)
2747 for index in range(len(self.ops)):
2748 assert self.ops[index].desc == ops_in_cpp[index]
AssertionError: