Backend WASM does not implement kernel SparseToDense

[vladmandic]

attempting to run nanodet object detection model

converted from pytorch - conversion notes and example of model usage are at:
https://gist.github.com/vladmandic/d0115e3899860391382d58550bd49b3f

engine.ts:540 Uncaught (in promise) Error: Kernel 'SparseToDense' not registered for backend 'wasm'
    at Engine.runKernel (engine.ts:540)
    at sparseToDense_ (sparse_to_dense.ts:88)
    at sparseToDense__op (operation.ts:51)
    at executeOp15 (slice_join_executor.ts:183)
    at operation_executor.ts:92
    at engine.ts:467
    at Engine.scopedRun (engine.ts:478)
    at Engine.tidy (engine.ts:465)
    at tidy (globals.ts:192)
    at operation_executor.ts:91

same graph model works without issues in nodejs with tensorflow backend and browser with webgl backend

environment: tfjs 3.3.0 on ubuntu 20.10 with chrome 88

[rthadur]

cc @lina128 @jinjingforever

[CanyonWind]

@rthadur @ahmedsabie A similar thing happened for panoptic segmentation model conversion as well. Any idea where the SparseToDense operation could possibly come from? There is no sparse tensor in the pytorch/onnx models.

Thanks a lot.

[FabioRomagnolo]

Same problem here! I tried to reimplement the operation in a custom way like this:

const customSparseToDense = function(node) {
            let sparseIndices = node.inputs[0];
            let sparseValues = node.inputs[1];
            let outputShape = node.inputs[2].arraySync();
            let defaultValue = node.inputs[3].arraySync();
            return tf.sparseToDense(sparseIndices, sparseValues, outputShape, defaultValue);
          }
          tf.registerOp('SparseToDense', customSparseToDense);

But without success: it gives too many error on the input's validation.

[AdvMaple]

Having this problem when using with @tensorflow-models/universal-sentence-encoder.

Evironment: Node 14, Ubuntu 20.04 Package installed:

    "@tensorflow-models/universal-sentence-encoder": "^1.3.3",
    "@tensorflow/tfjs": "^3.18.0",
    "@tensorflow/tfjs-backend-wasm": "^3.18.0",
    "@tensorflow/tfjs-core": "^3.18.0",
    "@tensorflow/tfjs-node": "^3.18.0",

WASM init:

const tf = require('@tensorflow/tfjs');
import "@tensorflow/tfjs-backend-wasm";
import "@tensorflow/tfjs-backend-webgl";
import "@tensorflow/tfjs-node";
import {getThreadsCount, setThreadsCount} from '@tensorflow/tfjs-backend-wasm';

dotenv.config();
setThreadsCount(2);
async function bootstrap() {
    const use = new UniversalSentenceEncoder();
    await use.load();
    use.embed("string").then(...)
}

tf.setBackend("wasm").then(() => void bootstrap());

Error:

ERROR  2022-07-11T23:44:58.189 middleware/unknown-except:15  | Error: Kernel 'SparseToDense' not registered for backend 'wasm'
    at Engine.runKernel (/home/user/Documents/tfjs-core/src/engine.ts:548:13)
    at sparseToDense_ (/home/user/Documents/tfjs-core/src/ops/sparse_to_dense.ts:88:17)
    at tfOps.sparseToDense (/home/user/Documents/tfjs-core/src/ops/operation.ts:51:24)
    at sliceJoin.executeOp (/home/user/Documents/project/tfjs-converter/src/operations/executors/slice_join_executor.ts:183:19)
    at /home/user/Documents/project/tfjs-converter/src/operations/operation_executor.ts:94:23
    at /home/user/Documents/project/tfjs-core/src/engine.ts:467:20
    at Engine.scopedRun (/home/user/Documents/project/tfjs-core/src/engine.ts:478:19)
    at Engine.tidy (/home/user/Documents/project/tfjs-core/src/engine.ts:465:17)
    at Object.tidy (/home/user/Documents/project/tfjs-core/src/globals.ts:192:17)
    at /home/user/Documents/project/tfjs-converter/src/operations/operation_executor.ts:93:24
    at executeOp (/home/user/Documents/project/tfjs-converter/src/operations/operation_executor.ts:121:8)
    at _loop_1 (/home/user/Documents/project/tfjs-converter/src/executor/graph_executor.ts:551:13)
    at GraphExecutor.processStack (/home/user/Documents/project/node_modules/@tensorflow/tfjs-converter/dist/tf-converter.node.js:29831:13)
    at GraphExecutor.<anonymous> (/home/user/Documents/project/tfjs-converter/src/executor/graph_executor.ts:497:29)
    at step (/home/user/Documents/project/node_modules/tslib/tslib.es6.js:100:23)
    at Object.next (/home/user/Documents/project/node_modules/tslib/tslib.es6.js:81:53) (unknown-exceptions)