akirasosa
akirasosa
Hi, I have rewrite code using PyTorch and tf-lite is TODO now.
It's TODO now.
@aminm2a I tried it by using U-Net like architecture. It works well. The tradeoff of accuracy vs speed is almost same with the classification problem. https://github.com/akirasosa/mobile-semantic-segmentation
@hkashlan I will consider it. But can I ask how you utilize it more concretely? I would like to hear your ideas.
@hkashlan I understand your idea. Thanks,
Yes. I will update. Though I don't have so much tricks, Ranger and Mish might work well.