1 | "use strict";
|
2 |
|
3 |
|
4 |
|
5 |
|
6 |
|
7 |
|
8 |
|
9 |
|
10 |
|
11 |
|
12 |
|
13 |
|
14 |
|
15 |
|
16 |
|
17 |
|
18 | Object.defineProperty(exports, "__esModule", { value: true });
|
19 | var tfjs_1 = require("@tensorflow/tfjs");
|
20 | var DepthwiseConv2dNative_1 = require("./DepthwiseConv2dNative");
|
21 | exports.fusedDepthwiseConv2DConfig = {
|
22 | kernelName: tfjs_1.FusedDepthwiseConv2D,
|
23 | backendName: 'tensorflow',
|
24 | kernelFunc: function (args) {
|
25 | var _a = args.inputs, x = _a.x, filter = _a.filter, bias = _a.bias, preluActivationWeights = _a.preluActivationWeights;
|
26 | var backend = args.backend;
|
27 | var _b = args.attrs, strides = _b.strides, pad = _b.pad, dilations = _b.dilations, dimRoundingMode = _b.dimRoundingMode, activation = _b.activation, leakyreluAlpha = _b.leakyreluAlpha;
|
28 | var $dilations = dilations;
|
29 | if ($dilations == null) {
|
30 | $dilations = [1, 1];
|
31 | }
|
32 | var convInfo = tfjs_1.backend_util.computeConv2DInfo(x.shape, filter.shape, strides, $dilations, pad, dimRoundingMode, true );
|
33 | var result = DepthwiseConv2dNative_1.depthwiseConv2dNativeImpl(x, filter, convInfo, backend);
|
34 | var toDispose = [];
|
35 | if (bias != null) {
|
36 | toDispose.push(result);
|
37 | result = tfjs_1.add(result, bias);
|
38 | }
|
39 | var temp = result;
|
40 | result = backend.applyActivation(result, activation, preluActivationWeights, leakyreluAlpha);
|
41 | if (temp !== result) {
|
42 | toDispose.push(temp);
|
43 | }
|
44 | toDispose.forEach(function (t) { return t.dispose(); });
|
45 | return result;
|
46 | }
|
47 | };
|