search

intel / webml-polyfill

Deprecated, the Web Neural Network Polyfill project has been moved to https://github.com/webmachinelearning/webnn-polyfill
Apache License 2.0
161 stars 42 forks source link

[Investigation] CPU custom ops #564

Closed huningxin closed 5 years ago

huningxin commented 5 years ago

https://github.com/webmachinelearning/webnn/issues/6

huningxin commented 5 years ago

Layout a draft plan:

  1. Integrate WebNN into ONNX.js #350 (by focusing on WebAssembly backend as first step)
  2. Use ResNet50 as workload
  3. Execute ResNet50 (e.g. modifying demo) with different options 3.1 all wasm op 3.2 some wasm ops (e.g. BN, RELU) + some WebNN ops (CONV, GEMM, POOL) (allow to configure for different combinations) 3.3 all WebNN ops
  4. Implement ADD (#456) and FC (#460) in MKL-DNN backend
  5. Move MKL-DNN backend into renderer process from GPU process (#569)
pinzhenx commented 5 years ago

@huningxin I'm wondering if we can close this issue?

huningxin commented 5 years ago

3. Execute ResNet50 (e.g. modifying demo) with different options 3.1 all wasm op 3.2 some wasm ops (e.g. BN, RELU) + some WebNN ops (CONV, GEMM, POOL) (allow to configure for different combinations) 3.3 all WebNN ops

This part has been covered by https://github.com/intel/webml-polyfill/issues/594