Linchenn issues

Results 11 issues of


                                            Linchenn

[browserstack benchmark tool] Move CodeSnippet from server to client

This PR moves CodeSnippet to benchmark from server sider to client side. Specifically: Previously, CodeSnippet is hard-coded in a server file (benchmark_model.js), which will be executed by BrowserStack instances. This...

[WebGL backend] Wrong results for AutoML Object model

When executing AutoML Object model with the following code snippet, it works for `3.19.0` version of WebGL backend, but it does not work for the [current github repo](https://github.com/tensorflow/tfjs/tree/f5b469bca042cf7d4144eb6f65e143a15c975805). ```js const...

type:bug

[Optimization target] The inference time of DeepLabV3's cityscapes architecture is long

DeepLabV3 with cityscapes takes long time to execute and we need to figure out the reason and then add it.

type:performance

[local-benchmark tool] Generates URL with parameter setting in the page

We could add a function for the local-benchmark tool: Provides a button and when users click it, it would encode all paramter setting in the current page-view and construct an...

type:feature

[WebGL] Compare setting inputChannel-loop as inner loop against filterHeight-loop as inner loop for Conv2d

This template is for miscellaneous issues not covered by the other issue categories. For questions on how to work with TensorFlow.js, or support for problems that are not verified bugs...

type:others

Dense support and conv2d implementation on dense

This PR is just for proof-of-concept use and is not intended to be merged. This PR: * Assumes the size of the last dimension of all tensor are multiples of...

[WebGL] Benchmark code snippet for conv2d

**Will not merge this into code base** Will use this code snippet to compare the conv2d's performance of the experimental dense implementation, https://github.com/tensorflow/tfjs/pull/6566, against the current implementation. ## Browserstack benchmark...

Linchenn