Some questions about MobileNet and DCGAN implemented on VTA

Questions
#1

I try my best to implement the mobilenet and dcgan mentioned in VTA: An Open Hardware-Software Stack for Deep Learning.
I used the standard model in the module zoo, but it didn’t achieve the desired results. The implementation of resnet18 on vta is open source, but MobileNet and DCGAN are not. So anyone can tell me how to implement the above two neural networks on vta. Or, if the authors open up their implementation, please tell me. I will be very grateful！

Thank you!

#2

Hi @varinic, thank you for your interest in our work. We’ll be releasing our experiments very soon as we are finalizing Relay compilation support. We’ll share the models we used in the paper. Indeed these aren’t in the gluon model zoo, and are instead customized models

#3

Thanks! I am looking forward to your next work.

#4

Hi, I am wondering if there is any update on the release of MobileNet and DCGAN implementation on VTA?

#5

Hello, Is there any update regarding releasing the model used in the paper especially MobileNetGC? Any help would be greatly appreciated.

#6

any update on this issue?

#7

@thierry @zhanghaohit @varinic @mrb256 @joyliu37 Hello everyone, I have been trying to implement the network on FPGA through VTA and mobilenetV2, where I set nn. Conv2d as the following format:

groups = int(in_planes/16)
    return nn.Conv2d(in_planes, out_planes, kernel_size=3, groups=groups ,stride=stride ,padding=1, bias=False)

I found that although he eventually implemented it, the classification output I got from the VTA was different or even completely different from the classification output I got from the software. Have you ever tried to implement grouping convolution, and is the result of it fought for? I’d like to hear from you,

#8

Hello, have you implemented the deployment of mobilenetv2 on FPGA?