search

hyperconnect / MMNet

Code for Towards Real-Time Automatic Portrait Matting on Mobile Devices
https://arxiv.org/abs/1904.03816v1
Apache License 2.0
177 stars 36 forks source link

Pre-trained model #1

Closed SanthoshRajendiran closed 5 years ago

SanthoshRajendiran commented 5 years ago

We went through your model. The outputs were surprising. Could you share with us your pre-trained model? Is the model giving these results in Mobile CPU?? You have handled quantized models, but how did you manage to preserve the accuracy to good extent.

shurain-hpcnt commented 5 years ago

Could you share with us your pre-trained model?

We don't have a plan to release a pre-trained model yet.

Is the model giving these results in Mobile CPU??

Yes, the model is intended to work best on mobile CPUs. To reduce latency, we had to implement custom operations that support SIMD instructions.

You have handled quantized models, but how did you manage to preserve the accuracy to good extent.

As mentioned in the supplementary material, we've used fake quantization for quantization-aware training.