@hoff97/tensor-js
Version:
PyTorch like deep learning inferrence library
50 lines • 2.32 kB
JavaScript
var __awaiter = (this && this.__awaiter) || function (thisArg, _arguments, P, generator) {
function adopt(value) { return value instanceof P ? value : new P(function (resolve) { resolve(value); }); }
return new (P || (P = Promise))(function (resolve, reject) {
function fulfilled(value) { try { step(generator.next(value)); } catch (e) { reject(e); } }
function rejected(value) { try { step(generator["throw"](value)); } catch (e) { reject(e); } }
function step(result) { result.done ? resolve(result.value) : adopt(result.value).then(fulfilled, rejected); }
step((generator = generator.apply(thisArg, _arguments || [])).next());
});
};
import { glContext } from '../../../tensor/gpu/gl';
import { OnnxNode } from '../../node';
export class InstanceNormalizationNode extends OnnxNode {
constructor(attributes, inputs, outputs, constants, onnxVersion, mode) {
super(attributes, inputs, outputs, constants, onnxVersion, mode);
this.epsilon = this.getAttributeFloat('epsilon') || 1e-5;
//TODO: Handle onnx versions < 6 here
}
forward(inputs) {
return __awaiter(this, void 0, void 0, function* () {
const x = inputs[0];
let scale = inputs[1];
let B = inputs[2];
const dataRank = x.getShape().length - 2;
const C = scale.getShape()[0];
const newShape = [1, C, ...new Array(dataRank).fill(1)];
scale = scale.reshape(newShape, false);
B = B.reshape(newShape, false);
const reduceAxes = new Array(x.getShape().length - 2);
for (let i = 0; i < dataRank; i++) {
reduceAxes[i] = i + 2;
}
const mean = x.reduceMean(reduceAxes, true);
glContext.flush();
const diff = x.subtract(mean);
glContext.flush();
const variance = diff.reduceMeanSquare(reduceAxes, true);
glContext.flush();
const result = x.normalize(mean, variance, this.epsilon, scale, B);
mean.delete();
diff.delete();
variance.delete();
return [result];
});
}
getType() {
return 'InstanceNormalization';
}
delete() { }
}
//# sourceMappingURL=instanceNormalization.js.map