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

[Windows/clDNN] Fail to switch to other models in DeepLab example on Windows #451

Open Christywl opened 5 years ago

Christywl commented 5 years ago

Test Env: Chromium Version: nightly build 70.0.3503.0 (4da550a) Platform: Windows 10(Dell XPS 13)[CPU: Intel i5-8250U, GPU: Intel UHD Craphics 620(driver: 25.20.100.6471), Memory: 8GB]

Expected Result: DeepLab example works.

Actual Result: Fail to switch to other models in DeepLab example, and see a black screen in the browser :

deeplab

How to Reproduce:

  1. Launch chromium and visit https://pinzhenx.github.io/webml-demo/examples/deeplab/index.html
  2. The default model is DeepLab 513, try to switch to other model
Christywl commented 5 years ago

In the default model DeepLab 513, tried to switch to WebGL, it also failed. WASM is ok.

Christywl commented 5 years ago

Found a way to test other DeepLab models on WebML backend: For example: Default model: DeepLab 513 Firstly, switch to WASM backend for the default model, and then select DeepLab 321 model, finally switch to WebML backend for DeepLab 321.

huningxin commented 5 years ago

@Christywl , please help verify whether this issue has been fixed https://github.com/otcshare/chromium-src/pull/86. Thanks!

Christywl commented 5 years ago

On the latest build https://github.com/otcshare/chromium-src/commit/20fce9c04a478d8e81e868223bb02e1cd79f0a45 and examples https://github.com/intel/webml-polyfill/commit/d67757271fdddcb1159f4bf1aa6d81689b6701d6, mostly I can switch the deeplab models successfully, sometimes it happens when switch models frequently. The reproducibility is lower. image

huningxin commented 5 years ago

Thanks for re-testing!

sometimes it happens when switch models frequently. The reproducibility is lower.

Can you elaborate the the details? How to reproduce? and how often to happen?

Christywl commented 5 years ago

Steps to reproduce:

  1. git clone https://github.com/intel/webml-polyfill, setup and start the server
  2. Visit http://localhost:8080/examples/semantic_segmentation/index.html
  3. Switch to SUSTAINED_SPEED
  4. Select one model, wait for the result, and then switch to other model
  5. Repeat Step 4 for several times

For example: