webgl seems not to work (or be used)

[weingaunity]

Hi,

i'm using onnx.js V0.1.8 from to run mobilenetv2 exported from pytorch. But i had to do some modifications because of unsupported operators created due to an adapative average pooling2d (see #265). But now it works and create the same output compared to pytorch when using the same image.

New issue: webgl seems not to work.
A tensorflow.js version of mobilenetv2 achieved approx. 10fps inference.

onnx.js - backend - cpu: Approx. 0.5fps
onnx.js - backend - wasm: Approx. 0.9fps
onnx.js - backend - webgl: Approx. 0.5fps .... so i thing webgl is not used there.

I used following code to activate webgl.

What is wrong?

Thx. Klaus

Using <script src="./onnx.min.js"></script> and download all the files from cdn and placed the files in the same directory like index.html

var mdl="./pytorch_mobilenetv2.onnx";
const myOnnxSession = new onnx.InferenceSession({ backendHint: 'webgl' });
myOnnxSession.loadModel(mdl).then(() => {});

[weingaunity]

When using webgl i get following warnings, but as mentioned no speed improvement compared to cpu.

WebGL warning: drawArraysInstanced: Using format enabled by implicitly enabled extension: EXT_float_blend. For maximal portability enable it explicitly. 31 webgl-context.ts:191:12
WebGL: No further warnings will be reported for this WebGL context. (already reported 32 warnings)

[smilemakc]

@weingaunity you are running on MacBook?

[weingaunity]

@weingaunity you are running on MacBook?

No, i'm using a Laptop with Linux. My solution was now converting the ONNX file to a tensorflow.js based on this i'm using now following lines:

pip install tensorflowjs
don't use: pip install onnx_tf
pip install git+https://github.com/onnx/onnx-tensorflow.git

import tensorflow as tf
import tensorflowjs as tfjs
import onnx
from onnx_tf.backend import prepare
from tensorflow.python.tools.import_pb_to_tensorboard import import_to_tensorboard

onnx_model = onnx.load("output/pytorch_mobilenetv2.onnx")
tf_rep = prepare(onnx_model) # returns: A TensorflowRep class object representing the ONNX model
tf_rep.export_graph("output/pytorch_mobilenetv2.pb")
#import_to_tensorboard("output/pytorch_mobilenetv2.pb", "tb_log","")
mdl2=tf.saved_model.load("output/pytorch_mobilenetv2.pb")
tfjs.converters.convert_tf_saved_model("output/pytorch_mobilenetv2.pb", "output/js")