christorange / qc-cnn Goto Github PK

老哥

运行项目的时候 Devices是不是不能选择cuda
我看选用cuda后 optimizer.step()那边报错：
RuntimeError: Expected all tensors to be on the same device, but found at least two devices, cuda:0 and cpu!

量子微分是不是目前不能在gpu上 【基于pennylane】

老哥我又来了 我有个新问题

我有读Hybrid quantum-classical convolutional neural networks这篇论文
Hybrid QCCNN. The input demonstrated here is a two-dimensional array, which is sent to a quantum convolutional layer of 6
filters. 这个多个滤波器的概念如果要实现应该怎么修改呢 因为我看现在数据的通道数是从1变到N量子位 且受限于量子位（虽然量子机器学习的优势就是减少参数数量 但我总感觉用一个量子电路会不会不够） 正常我们Conv2d会有输入输出的通道数 也就是说滤波核是三维的（+通道） 我看现在您复现的QCCNN里面好像不是 我看其他复现QCCNN的好像也没通道这个概念 我不清楚我这样理解是否正确 希望能够得到您的指导

非常感谢
Pengl

Where need make modification?

Thanks you

Thanks you for you make this code, is very good for me understanding.

Here:
https://pennylane.ai/qml/demos/tutorial_quantum_transfer_learning.html

@qml.qnode(dev, interface="torch")
def quantum_net(q_input_features, q_weights_flat):
    
    # Reshape weights
    q_weights = q_weights_flat.reshape(q_depth, n_qubits)

    # Start from state |+> , unbiased w.r.t. |0> and |1>
    H_layer(n_qubits)

    # Embed features in the quantum node
    RY_layer(q_input_features)

    # Sequence of trainable variational layers
    for k in range(q_depth):
        entangling_layer(n_qubits)
        RY_layer(q_weights[k])

    # Expectation values in the Z basis
    exp_vals = [[qml.expval](https://docs.pennylane.ai/en/stable/code/api/pennylane.expval.html#pennylane.expval)([qml.PauliZ]

(https://docs.pennylane.ai/en/stable/code/api/pennylane.PauliZ.html#pennylane.PauliZ)(position)) for position in range(n_qubits)]
return tuple(exp_vals)
He make use of reshape weights, but your code no make use. Why? Whts is difference?

Also here:
https://github.com/christorange/QC-CNN/blob/6f498d25de4187b730f7b146ad691e95520ed2d9/models/single_encoding.py#LL36C4-L42C43

 x_lst = []
for i in range(0, x.shape[2]-1,2):
    for j in range(0, x.shape[3]-1,2):
        x_lst.append(self.ql1(torch.flatten(x[:, :, i:i + self.kernel_size, j:j + self.kernel_size], start_dim=1)))
x = torch.cat(x_lst,dim=1)  # .view(bs,n_qubits,14,14)

What is the prupose of x_lst and make loop after? I no understand.

If want make process 64x64 image, what change neem make?

Thanks you and sorry for bad english.

Hi @christorange ,
You have done a fabulous work. I would like to run on my custom image dataset where image size is quite larger 416x416. But if i convert to csv its very large, can you please help me to execute in such configuration.

因为最近导师给的方向是经典神经网络加一点量子，那么对于传统的目标检测，比如ssd算法的话，我现在想的是将传统的2d卷积换成量子卷积，但是我觉得因为输入会很大，比如对一个3030256的输入进行量子卷积，输出成1515256的，使用量子卷积的话，可以实现出来嘛