Unknown op '_FusedConv2D' with tfjs-automl using object-detection demo

[gitunit]

I've exported a model from AutoML and now trying to make it run with Tensorflow.js.
Im using the object_detection demo code and i replaced the link to my local model.
It has been loaded successfully but when it comes to the prediction part, it fails with the error message:

Uncaught (in promise) TypeError: Unknown op '_FusedConv2D'. File an issue at https://github.com/tensorflow/tfjs/issues so we can add it, or register a custom execution with tf.registerOp()
at operation_executor.ts:95
at executeOp$16 (operation_executor.ts:52)
at c (graph_executor.ts:351)
at e.processStack (graph_executor.ts:337)
at e. (graph_executor.ts:299)
at tensor.ts:397
at Object.next (tensor.ts:397)
at o (tensor.ts:397)

i saw a similar issue but this was a different model and i believe running the latest with the demo code from this repo.

edit: it works with this snippet though

[pyu10055]

@gitunit The _FusedConv2d op is supported after 1.13, please load the latest tfjs union package. thanks.

[gitunit]

@gitunit The _FusedConv2d op is supported after 1.13, please load the latest tfjs union package. thanks.

isn't the object_detection demo already loading the latest? if not, how would i do that?

edit: i figured it out. i had to run the following to update the required packages:

npm update @tensorflow/tfjs-core
npm update @tensorflow/tfjs-converter

this will update those packages the tfjs-automl is based on (i guess) to version 1.3.1
@pyu10055 you might have misspelled the version number, since i couldn't find v1.13

furthermore with this it works but it seems very slow. could it be that it uses CPU as backend? how would i set WEBGL as backend?
i've tried by using tf.setbackend('webgl') but this has no effect. did i miss anything? has this something to do with automl?

edit: setting the backend to 'webgl' actually works. first prediction of a frame takes longer than of subsequent frames. see answer here.

[Wlisfes]

您好！我已经更新了依赖包还是出现这个错误

"@tensorflow/tfjs-core": "^1.3.1", "@tensorflow/tfjs-automl": "^1.3.1"
operation_executor.ts:95 Uncaught (in promise) TypeError: Unknown op '_FusedConv2D'. File an issue at https://github.com/tensorflow/tfjs/issues so we can add it, or register a custom execution with tf.registerOp()
at operation_executor.ts:95
at executeOp$16 (operation_executor.ts:52)
at c (graph_executor.ts:351)
at e.processStack (graph_executor.ts:337)
at e. (graph_executor.ts:299)
at add.ts:17
at Object.next (add.ts:17)
at o (add.ts:17)
``

[alexhmyang]

您好！我已经更新了依赖包还是出现这个错误

"@tensorflow/tfjs-core": "^1.3.1", "@tensorflow/tfjs-automl": "^1.3.1" operation_executor.ts:95 Uncaught (in promise) TypeError: Unknown op '_FusedConv2D'. File an issue at https://github.com/tensorflow/tfjs/issues so we can add it, or register a custom execution with tf.registerOp() at operation_executor.ts:95 at executeOp$16 (operation_executor.ts:52) at c (graph_executor.ts:351) at e.processStack (graph_executor.ts:337) at e. (graph_executor.ts:299) at add.ts:17 at Object.next (add.ts:17) at o (add.ts:17) ``

same problem, anyone solve it?
npm info @tensorflow/tfjs-converter

@tensorflow/tfjs-converter@4.2.0 | Apache-2.0 | deps: none | versions: 123