@vladmandic/face-api
Version:
FaceAPI: AI-powered Face Detection & Rotation Tracking, Face Description & Recognition, Age & Gender & Emotion Prediction for Browser and NodeJS using TensorFlow/JS
93 lines (83 loc) • 3.91 kB
text/typescript
import * as tf from '../../dist/tfjs.esm';
import { extractConvParamsFactory } from '../common/index';
import { extractSeparableConvParamsFactory } from '../common/extractSeparableConvParamsFactory';
import { extractWeightsFactory } from '../common/extractWeightsFactory';
import { ExtractWeightsFunction, ParamMapping } from '../common/types';
import { TinyYolov2Config } from './config';
import { BatchNorm, ConvWithBatchNorm, TinyYolov2NetParams } from './types';
function extractorsFactory(extractWeights: ExtractWeightsFunction, paramMappings: ParamMapping[]) {
const extractConvParams = extractConvParamsFactory(extractWeights, paramMappings);
function extractBatchNormParams(size: number, mappedPrefix: string): BatchNorm {
const sub = tf.tensor1d(extractWeights(size));
const truediv = tf.tensor1d(extractWeights(size));
paramMappings.push(
{ paramPath: `${mappedPrefix}/sub` },
{ paramPath: `${mappedPrefix}/truediv` },
);
return { sub, truediv };
}
function extractConvWithBatchNormParams(channelsIn: number, channelsOut: number, mappedPrefix: string): ConvWithBatchNorm {
const conv = extractConvParams(channelsIn, channelsOut, 3, `${mappedPrefix}/conv`);
const bn = extractBatchNormParams(channelsOut, `${mappedPrefix}/bn`);
return { conv, bn };
}
const extractSeparableConvParams = extractSeparableConvParamsFactory(extractWeights, paramMappings);
return {
extractConvParams,
extractConvWithBatchNormParams,
extractSeparableConvParams,
};
}
export function extractParams(
weights: Float32Array,
config: TinyYolov2Config,
boxEncodingSize: number,
filterSizes: number[],
): { params: TinyYolov2NetParams, paramMappings: ParamMapping[] } {
const {
extractWeights,
getRemainingWeights,
} = extractWeightsFactory(weights);
const paramMappings: ParamMapping[] = [];
const {
extractConvParams,
extractConvWithBatchNormParams,
extractSeparableConvParams,
} = extractorsFactory(extractWeights, paramMappings);
let params: TinyYolov2NetParams;
if (config.withSeparableConvs) {
const [s0, s1, s2, s3, s4, s5, s6, s7, s8] = filterSizes;
const conv0 = config.isFirstLayerConv2d
? extractConvParams(s0, s1, 3, 'conv0')
: extractSeparableConvParams(s0, s1, 'conv0');
const conv1 = extractSeparableConvParams(s1, s2, 'conv1');
const conv2 = extractSeparableConvParams(s2, s3, 'conv2');
const conv3 = extractSeparableConvParams(s3, s4, 'conv3');
const conv4 = extractSeparableConvParams(s4, s5, 'conv4');
const conv5 = extractSeparableConvParams(s5, s6, 'conv5');
const conv6 = s7 ? extractSeparableConvParams(s6, s7, 'conv6') : undefined;
const conv7 = s8 ? extractSeparableConvParams(s7, s8, 'conv7') : undefined;
const conv8 = extractConvParams(s8 || s7 || s6, 5 * boxEncodingSize, 1, 'conv8');
params = {
conv0, conv1, conv2, conv3, conv4, conv5, conv6, conv7, conv8,
};
} else {
const [s0, s1, s2, s3, s4, s5, s6, s7, s8] = filterSizes;
const conv0 = extractConvWithBatchNormParams(s0, s1, 'conv0');
const conv1 = extractConvWithBatchNormParams(s1, s2, 'conv1');
const conv2 = extractConvWithBatchNormParams(s2, s3, 'conv2');
const conv3 = extractConvWithBatchNormParams(s3, s4, 'conv3');
const conv4 = extractConvWithBatchNormParams(s4, s5, 'conv4');
const conv5 = extractConvWithBatchNormParams(s5, s6, 'conv5');
const conv6 = extractConvWithBatchNormParams(s6, s7, 'conv6');
const conv7 = extractConvWithBatchNormParams(s7, s8, 'conv7');
const conv8 = extractConvParams(s8, 5 * boxEncodingSize, 1, 'conv8');
params = {
conv0, conv1, conv2, conv3, conv4, conv5, conv6, conv7, conv8,
};
}
if (getRemainingWeights().length !== 0) {
throw new Error(`weights remaing after extract: ${getRemainingWeights().length}`);
}
return { params, paramMappings };
}