11import { Niivue } from '@niivue/niivue'
2- import * as ort from 'onnxruntime-web' ;
2+ // IMPORTANT: we need to import this specific file.
3+ import * as ort from "./node_modules/onnxruntime-web/dist/ort.all.mjs"
4+ console . log ( ort ) ;
35async function main ( ) {
46 aboutBtn . onclick = function ( ) {
57 let url = "https://github.com/axinging/mlmodel-convension-demo/blob/main/onnx/onnx-brainchop.html"
@@ -17,79 +19,103 @@ async function main() {
1719 const nv1 = new Niivue ( defaults )
1820 nv1 . attachToCanvas ( gl1 )
1921 await nv1 . loadVolumes ( [ { url : './t1_crop.nii.gz' } ] )
22+ // FIXME: Do we want to conform?
23+ const conformed = await nv1 . conform (
24+ nv1 . volumes [ 0 ] ,
25+ false ,
26+ true ,
27+ true
28+ )
29+ nv1 . removeVolume ( nv1 . volumes [ 0 ] )
30+ nv1 . addVolume ( conformed )
2031
2132 let feedsInfo = [ ] ;
2233 function getFeedInfo ( feed , type , data , dims ) {
23- const warmupTimes = 0 ;
24- const runTimes = 1 ;
25- for ( let i = 0 ; i < warmupTimes + runTimes ; i ++ ) {
26- let typedArray ;
27- let typeBytes ;
28- if ( type === 'bool' ) {
29- data = [ data ] ;
30- dims = [ 1 ] ;
31- typeBytes = 1 ;
32- } else if ( type === 'int8' ) {
33- typedArray = Int8Array ;
34- } else if ( type === 'float16' ) {
35- typedArray = Uint16Array ;
36- } else if ( type === 'int32' ) {
37- typedArray = Int32Array ;
38- } else if ( type === 'uint32' ) {
39- typedArray = Uint32Array ;
40- } else if ( type === 'float32' ) {
41- typedArray = Float32Array ;
42- } else if ( type === 'int64' ) {
43- typedArray = BigInt64Array ;
44- }
45- if ( typeBytes === undefined ) {
46- typeBytes = typedArray . BYTES_PER_ELEMENT ;
47- }
34+ const warmupTimes = 0 ;
35+ const runTimes = 1 ;
36+ for ( let i = 0 ; i < warmupTimes + runTimes ; i ++ ) {
37+ let typedArray ;
38+ let typeBytes ;
39+ if ( type === 'bool' ) {
40+ data = [ data ] ;
41+ dims = [ 1 ] ;
42+ typeBytes = 1 ;
43+ } else if ( type === 'int8' ) {
44+ typedArray = Int8Array ;
45+ } else if ( type === 'float16' ) {
46+ typedArray = Uint16Array ;
47+ } else if ( type === 'int32' ) {
48+ typedArray = Int32Array ;
49+ } else if ( type === 'uint32' ) {
50+ typedArray = Uint32Array ;
51+ } else if ( type === 'float32' ) {
52+ typedArray = Float32Array ;
53+ } else if ( type === 'int64' ) {
54+ typedArray = BigInt64Array ;
55+ }
56+ if ( typeBytes === undefined ) {
57+ typeBytes = typedArray . BYTES_PER_ELEMENT ;
58+ }
4859
49- let size , _data ;
50- if ( Array . isArray ( data ) || ArrayBuffer . isView ( data ) ) {
51- size = data . length ;
52- _data = data ;
53- } else {
54- size = dims . reduce ( ( a , b ) => a * b ) ;
55- if ( data === 'random' ) {
56- _data = typedArray . from ( { length : size } , ( ) => getRandom ( type ) ) ;
57- } else {
58- _data = typedArray . from ( { length : size } , ( ) => data ) ;
59- }
60- }
60+ let size , _data ;
61+ if ( Array . isArray ( data ) || ArrayBuffer . isView ( data ) ) {
62+ size = data . length ;
63+ _data = data ;
64+ } else {
65+ size = dims . reduce ( ( a , b ) => a * b ) ;
66+ if ( data === 'random' ) {
67+ _data = typedArray . from ( { length : size } , ( ) => getRandom ( type ) ) ;
68+ } else {
69+ _data = typedArray . from ( { length : size } , ( ) => data ) ;
70+ }
71+ }
6172
62- if ( i > feedsInfo . length - 1 ) {
63- feedsInfo . push ( new Map ( ) ) ;
64- }
65- feedsInfo [ i ] . set ( feed , [ type , _data , dims , Math . ceil ( size * typeBytes / 16 ) * 16 ] ) ;
73+ if ( i > feedsInfo . length - 1 ) {
74+ feedsInfo . push ( new Map ( ) ) ;
6675 }
67- return feedsInfo ;
76+ feedsInfo [ i ] . set ( feed , [ type , _data , dims , Math . ceil ( size * typeBytes / 16 ) * 16 ] ) ;
77+ }
78+ return feedsInfo ;
6879 }
6980 const option = {
70- executionProviders : [
71- {
72- //name: 'webgpu',
73- name : 'webgl' ,
74- } ,
75- ] ,
76- graphOptimizationLevel : 'extended' ,
77- optimizedModelFilepath : 'opt.onnx'
81+ executionProviders : [
82+ {
83+ name : 'webgpu' ,
84+ } ,
85+ ] ,
86+ graphOptimizationLevel : 'extended' ,
87+ optimizedModelFilepath : 'opt.onnx'
7888 } ;
7989
8090 const session = await ort . InferenceSession . create ( './model_5_channels.onnx' , option ) ;
8191 const shape = [ 1 , 1 , 256 , 256 , 256 ] ;
82- const temp = getFeedInfo ( "input.1" , "float32" , 0 , shape ) ;
92+ // FIXME: Do we want to use a real image for inference?
93+ const imgData = nv1 . volumes [ 0 ] . img ;
94+ const expectedLength = shape . reduce ( ( a , b ) => a * b ) ;
95+ // FIXME: Do we need want this?
96+ if ( imgData . length !== expectedLength ) {
97+ throw new Error ( `imgData length (${ imgData . length } ${ expectedLength } ) ;
98+ }
99+
100+ const temp = getFeedInfo ( "input.1" , "float32" , imgData , shape ) ;
83101 let dataA = temp [ 0 ] . get ( 'input.1' ) [ 1 ] ;
84- // let dataTemp = await loadJSON("./onnx-branchchop-input64.jsonc");
85- // dataA = dataTemp['data'];
86102 const tensorA = new ort . Tensor ( 'float32' , dataA , shape ) ;
87-
103+
88104 const feeds = { "input.1" : tensorA } ;
89105 // feed inputs and run
90106 console . log ( "before run" ) ;
91107 const results = await session . run ( feeds ) ;
92- console . log ( "after run" ) ;
108+ console . log ( results ) ;
109+ console . log ( "after run" )
110+ // FIXME: is this really the output data? It doesn't make sense when rendered,
111+ // but then again, maybe the input was wrong?
112+ const outData = results [ 39 ] . data
113+ const newImg = nv1 . cloneVolume ( 0 ) ;
114+ newImg . img = outData
115+ // Add the output to niivue
116+ nv1 . addVolume ( newImg )
117+ nv1 . setColormap ( newImg . id , "red" )
118+ nv1 . setOpacity ( 1 , 0.5 )
93119}
94120
95121main ( )
0 commit comments