Browser based Inference is crashing

[10dimensions]

I tried to replicate the node implementation on tfjs (browser) itself.

Basically, with regards to the modifications,

var tfimage = tf.browser.fromPixels(imageData);
where, imageData refers to HTML Canvas ImageData.

And for the inference,

var preProcessedData = preprocessTf();
var output = session_tfjs.predict(preProcessedData);

The browser seems to be completely unresponsive, and crashes eventually. Could you please let me know if there are any other changes needed?

Also, can I expect the average inference time to be around 1500 ms ? (As opposed to ~40 ms fastdepth metrics on Nvidia Jetson hardware)

[anargu]

I've got the same average inference time (adapted for browser). I tried with fastdepth_opset9_tfjs and fastdepth_opset11_tfjs but the time just increases. I'm figuring out how to improve. Did you find a workaround to get a shorter inference time?