// META: title=test WebNN API instanceNormalization operation
// META: global=window,dedicatedworker
// META: variant=?cpu
// META: variant=?gpu
// META: variant=?npu
// META: script=../resources/utils.js
// META: timeout=long
'use strict';
// https://www.w3.org/TR/webnn/#api-mlgraphbuilder-instancenorm
// Normalize the input using Instance-Normalization.
//
// dictionary MLInstanceNormalizationOptions {
// MLOperand scale;
// MLOperand bias;
// double epsilon = 1e-5;
// MLInputOperandLayout layout = "nchw";
// };
//
// MLOperand instanceNormalization(
// MLOperand input, optional MLInstanceNormalizationOptions options = {});
const getInstanceNormPrecisionTolerance = (graphResources) => {
// according to
// https://github.com/web-platform-tests/wpt/pull/43891#discussion_r1457026316
const toleranceValueDict = {float32: 840, float16: 8400};
const expectedDataType =
getExpectedDataTypeOfSingleOutput(graphResources.expectedOutputs);
return {metricType: 'ULP', value: toleranceValueDict[expectedDataType]};
};
const instanceNormTests = [
{
'name': 'instanceNormalization float32 4D tensor default options',
'graph': {
'inputs': {
'instanceNormInput': {
'data': [
-97.949951171875, 29.44037628173828, -73.92131042480469,
-38.11185836791992, 41.33772659301758, -59.77853012084961,
-74.66901397705078, -68.16508483886719, 35.82481384277344,
-6.948329448699951, 54.42462158203125, 47.53074645996094,
66.93562316894531, 76.74034881591797, 5.6758809089660645,
25.68659210205078, 37.37651062011719, 56.252689361572266,
-16.574905395507812, 42.949893951416016, 73.8739242553711,
-99.00035095214844, -33.11322784423828, -17.380685806274414
],
'descriptor': {'dimensions': [2, 3, 2, 2], 'dataType': 'float32'}
}
},
'operators': [{
'name': 'instanceNormalization',
'arguments': [{'input': 'instanceNormInput'}],
'outputs': 'instanceNormOutput'
}],
'expectedOutputs': {
'instanceNormOutput': {
'data': [
-1.0995290279388428, 1.5525832176208496, -0.5992818474769592,
0.14622758328914642, 1.72129487991333, -0.41020718216896057,
-0.7240943908691406, -0.586993396282196, 0.13073226809501648,
-1.6633318662643433, 0.9108771681785583, 0.6217224597930908,
0.7947131395339966, 1.1309205293655396, -1.3059037923812866,
-0.6197298169136047, 0.2657700479030609, 0.9459608793258667,
-1.6783342361450195, 0.46660327911376953, 1.5037200450897217,
-1.2981476783752441, -0.2302791178226471, 0.024706769734621048
],
'descriptor': {'dimensions': [2, 3, 2, 2], 'dataType': 'float32'}
}
}
}
},
{
'name': 'instanceNormalization float32 4D tensor options.scale',
'graph': {
'inputs': {
'instanceNormInput': {
'data': [
-97.949951171875, 29.44037628173828, -73.92131042480469,
-38.11185836791992, 41.33772659301758, -59.77853012084961,
-74.66901397705078, -68.16508483886719, 35.82481384277344,
-6.948329448699951, 54.42462158203125, 47.53074645996094,
66.93562316894531, 76.74034881591797, 5.6758809089660645,
25.68659210205078, 37.37651062011719, 56.252689361572266,
-16.574905395507812, 42.949893951416016, 73.8739242553711,
-99.00035095214844, -33.11322784423828, -17.380685806274414
],
'descriptor': {'dimensions': [2, 3, 2, 2], 'dataType': 'float32'}
},
'instanceNormScale': {
'data': [-94.42772674560547, 66.69620513916016, -98.56572723388672],
'descriptor': {'dimensions': [3], 'dataType': 'float32'},
'constant': true
}
},
'operators': [{
'name': 'instanceNormalization',
'arguments': [
{'input': 'instanceNormInput'},
{'options': {'scale': 'instanceNormScale'}}
],
'outputs': 'instanceNormOutput'
}],
'expectedOutputs': {
'instanceNormOutput': {
'data': [
103.8260269165039, -146.60690307617188, 56.58882141113281,
-13.807937622070312, 114.80384063720703, -27.359262466430664,
-48.29434585571289, -39.150230407714844, -12.885721206665039,
163.94752502441406, -89.78126525878906, -61.2805290222168,
-75.04296112060547, -106.79025268554688, 123.31352996826172,
58.51968002319336, 17.725852966308594, 63.09199905395508,
-111.93852233886719, 31.120668411254883, -148.2152557373047,
127.95286560058594, 22.697628021240234, -2.4352407455444336
],
'descriptor': {'dimensions': [2, 3, 2, 2], 'dataType': 'float32'}
}
}
}
},
{
'name': 'instanceNormalization float32 4D tensor options.bias',
'graph': {
'inputs': {
'instanceNormInput': {
'data': [
-97.949951171875, 29.44037628173828, -73.92131042480469,
-38.11185836791992, 41.33772659301758, -59.77853012084961,
-74.66901397705078, -68.16508483886719, 35.82481384277344,
-6.948329448699951, 54.42462158203125, 47.53074645996094,
66.93562316894531, 76.74034881591797, 5.6758809089660645,
25.68659210205078, 37.37651062011719, 56.252689361572266,
-16.574905395507812, 42.949893951416016, 73.8739242553711,
-99.00035095214844, -33.11322784423828, -17.380685806274414
],
'descriptor': {'dimensions': [2, 3, 2, 2], 'dataType': 'float32'}
},
'instanceNormBias': {
'data': [-33.048641204833984, 4.511423587799072, -37.93617248535156],
'descriptor': {'dimensions': [3], 'dataType': 'float32'},
'constant': true
}
},
'operators': [{
'name': 'instanceNormalization',
'arguments': [
{'input': 'instanceNormInput'},
{'options': {'bias': 'instanceNormBias'}}
],
'outputs': 'instanceNormOutput'
}],
'expectedOutputs': {
'instanceNormOutput': {
'data': [
-34.148170471191406, -31.496057510375977, -33.64792251586914,
-32.90241241455078, 6.232718467712402, 4.1012163162231445,
3.7873291969299316, 3.9244301319122314, -37.80543899536133,
-39.59950256347656, -37.02529525756836, -37.314449310302734,
-32.253929138183594, -31.917720794677734, -34.35454559326172,
-33.66836929321289, 4.777193546295166, 5.4573845863342285,
2.8330893516540527, 4.978026866912842, -36.43245315551758,
-39.23432159423828, -38.16645050048828, -37.91146469116211
],
'descriptor': {'dimensions': [2, 3, 2, 2], 'dataType': 'float32'}
}
}
}
},
{
'name': 'instanceNormalization float32 4D tensor options.epsilon',
'graph': {
'inputs': {
'instanceNormInput': {
'data': [
-97.949951171875, 29.44037628173828, -73.92131042480469,
-38.11185836791992, 41.33772659301758, -59.77853012084961,
-74.66901397705078, -68.16508483886719, 35.82481384277344,
-6.948329448699951, 54.42462158203125, 47.53074645996094,
66.93562316894531, 76.74034881591797, 5.6758809089660645,
25.68659210205078, 37.37651062011719, 56.252689361572266,
-16.574905395507812, 42.949893951416016, 73.8739242553711,
-99.00035095214844, -33.11322784423828, -17.380685806274414
],
'descriptor': {'dimensions': [2, 3, 2, 2], 'dataType': 'float32'}
}
},
'operators': [{
'name': 'instanceNormalization',
'arguments': [
{'input': 'instanceNormInput'}, {'options': {'epsilon': 0.000001}}
],
'outputs': 'instanceNormOutput'
}],
'expectedOutputs': {
'instanceNormOutput': {
'data': [
-1.0995290279388428, 1.5525832176208496, -0.5992818474769592,
0.14622758328914642, 1.72129487991333, -0.41020718216896057,
-0.7240943908691406, -0.586993396282196, 0.13073226809501648,
-1.6633318662643433, 0.9108771681785583, 0.6217224597930908,
0.7947131991386414, 1.1309205293655396, -1.3059037923812866,
-0.6197298765182495, 0.2657700479030609, 0.9459608793258667,
-1.6783342361450195, 0.46660327911376953, 1.5037200450897217,
-1.2981476783752441, -0.2302791178226471, 0.024706769734621048
],
'descriptor': {'dimensions': [2, 3, 2, 2], 'dataType': 'float32'}
}
}
}
},
{
'name':
'instanceNormalization float32 4D tensor explict options.layout=\'nchw\'',
'graph': {
'inputs': {
'instanceNormInput': {
'data': [
-97.949951171875, 29.44037628173828, -73.92131042480469,
-38.11185836791992, 41.33772659301758, -59.77853012084961,
-74.66901397705078, -68.16508483886719, 35.82481384277344,
-6.948329448699951, 54.42462158203125, 47.53074645996094,
66.93562316894531, 76.74034881591797, 5.6758809089660645,
25.68659210205078, 37.37651062011719, 56.252689361572266,
-16.574905395507812, 42.949893951416016, 73.8739242553711,
-99.00035095214844, -33.11322784423828, -17.380685806274414
],
'descriptor': {'dimensions': [2, 3, 2, 2], 'dataType': 'float32'}
}
},
'operators': [{
'name': 'instanceNormalization',
'arguments':
[{'input': 'instanceNormInput'}, {'options': {'layout': 'nchw'}}],
'outputs': 'instanceNormOutput'
}],
'expectedOutputs': {
'instanceNormOutput': {
'data': [
-1.0995290279388428, 1.5525832176208496, -0.5992818474769592,
0.14622758328914642, 1.72129487991333, -0.41020718216896057,
-0.7240943908691406, -0.586993396282196, 0.13073226809501648,
-1.6633318662643433, 0.9108771681785583, 0.6217224597930908,
0.7947131395339966, 1.1309205293655396, -1.3059037923812866,
-0.6197298169136047, 0.2657700479030609, 0.9459608793258667,
-1.6783342361450195, 0.46660327911376953, 1.5037200450897217,
-1.2981476783752441, -0.2302791178226471, 0.024706769734621048
],
'descriptor': {'dimensions': [2, 3, 2, 2], 'dataType': 'float32'}
}
}
}
},
{
'name': 'instanceNormalization float32 4D tensor options.layout=\'nhwc\'',
'graph': {
'inputs': {
'instanceNormInput': {
'data': [
-97.949951171875, 41.33772659301758, 35.82481384277344,
29.44037628173828, -59.77853012084961, -6.948329448699951,
-73.92131042480469, -74.66901397705078, 54.42462158203125,
-38.11185836791992, -68.16508483886719, 47.53074645996094,
66.93562316894531, 37.37651062011719, 73.8739242553711,
76.74034881591797, 56.252689361572266, -99.00035095214844,
5.6758809089660645, -16.574905395507812, -33.11322784423828,
25.68659210205078, 42.949893951416016, -17.380685806274414
],
'descriptor': {'dimensions': [2, 2, 2, 3], 'dataType': 'float32'}
}
},
'operators': [{
'name': 'instanceNormalization',
'arguments':
[{'input': 'instanceNormInput'}, {'options': {'layout': 'nhwc'}}],
'outputs': 'instanceNormOutput'
}],
'expectedOutputs': {
'instanceNormOutput': {
'data': [
-1.0995290279388428, 1.72129487991333, 0.13073226809501648,
1.5525832176208496, -0.41020718216896057, -1.6633318662643433,
-0.5992818474769592, -0.7240943908691406, 0.9108771681785583,
0.14622758328914642, -0.586993396282196, 0.6217224597930908,
0.7947131395339966, 0.2657700479030609, 1.5037200450897217,
1.1309205293655396, 0.9459608793258667, -1.2981476783752441,
-1.3059037923812866, -1.6783342361450195, -0.2302791178226471,
-0.6197298169136047, 0.46660327911376953, 0.024706769734621048
],
'descriptor': {'dimensions': [2, 2, 2, 3], 'dataType': 'float32'}
}
}
}
},
{
'name': 'instanceNormalization float32 4D tensor all options',
'graph': {
'inputs': {
'instanceNormInput': {
'data': [
-97.949951171875, 41.33772659301758, 35.82481384277344,
29.44037628173828, -59.77853012084961, -6.948329448699951,
-73.92131042480469, -74.66901397705078, 54.42462158203125,
-38.11185836791992, -68.16508483886719, 47.53074645996094,
66.93562316894531, 37.37651062011719, 73.8739242553711,
76.74034881591797, 56.252689361572266, -99.00035095214844,
5.6758809089660645, -16.574905395507812, -33.11322784423828,
25.68659210205078, 42.949893951416016, -17.380685806274414
],
'descriptor': {'dimensions': [2, 2, 2, 3], 'dataType': 'float32'}
},
'instanceNormScale': {
'data': [-94.42772674560547, 66.69620513916016, -98.56572723388672],
'descriptor': {'dimensions': [3], 'dataType': 'float32'},
'constant': true
},
'instanceNormBias': {
'data': [-33.048641204833984, 4.511423587799072, -37.93617248535156],
'descriptor': {'dimensions': [3], 'dataType': 'float32'},
'constant': true
}
},
'operators': [{
'name': 'instanceNormalization',
'arguments': [
{'input': 'instanceNormInput'}, {
'options': {
'scale': 'instanceNormScale',
'bias': 'instanceNormBias',
'epsilon': 0.000001,
'layout': 'nhwc'
}
}
],
'outputs': 'instanceNormOutput'
}],
'expectedOutputs': {
'instanceNormOutput': {
'data': [
70.77738189697266, 119.31526184082031, -50.821895599365234,
-179.65554809570312, -22.847837448120117, 126.01134490966797,
23.540178298950195, -43.782920837402344, -127.71744537353516,
-46.8565788269043, -34.6388053894043, -99.2166976928711,
-108.09159851074219, 22.237276077270508, -186.15142822265625,
-139.83889770507812, 67.60342407226562, 90.01669311523438,
90.26488494873047, -107.4271011352539, -15.238543510437012,
25.471038818359375, 35.6320915222168, -40.37141418457031
],
'descriptor': {'dimensions': [2, 2, 2, 3], 'dataType': 'float32'}
}
}
}
}
];
if (navigator.ml) {
instanceNormTests.forEach((test) => {
webnn_conformance_test(
buildGraphAndCompute, getInstanceNormPrecisionTolerance, test);
});
} else {
test(() => assert_implements(navigator.ml, 'missing navigator.ml'));
}
Codebase Browser
chromium