Skip to content

F
face
Commit 4ef65c1b by vincent
Options

check in latest train scripts

parent 0320da07
Showing with 202 additions and 209 deletions
tools/train/serveTinyYolov2.js
......@@ -25,7 +25,7 @@ const detectionFilenames = fs.readdirSync(detectionsPath)
app.use(express.static(trainDataPath))
app.get('/detection_filenames', (req, res) => res.status(202).send(detectionFilenames))
app.get('/', (req, res) => res.sendFile(path.join(publicDir, 'overfit.html')))
app.get('/', (req, res) => res.sendFile(path.join(publicDir, 'train.html')))
app.get('/verify', (req, res) => res.sendFile(path.join(publicDir, 'verify.html')))
app.listen(3000, () => console.log('Listening on port 3000!'))
\ No newline at end of file
tools/train/tinyYolov2/overfit.html deleted 100644 → 0
<!DOCTYPE html>
<html>
<head>
<link rel="stylesheet" href="styles.css">
<link rel="stylesheet" href="https://cdnjs.cloudflare.com/ajax/libs/materialize/0.100.2/css/materialize.css">
<script type="text/javascript" src="https://code.jquery.com/jquery-2.1.1.min.js"></script>
<script src="https://cdnjs.cloudflare.com/ajax/libs/materialize/0.100.2/js/materialize.min.js"></script>
<script src="face-api.js"></script>
<script src="commons.js"></script>
<script src="FileSaver.js"></script>
<script src="trainUtils.js"></script>
<script src="train.js"></script>
<script src="loss.js"></script>
</head>
<body>
<script>
tf = faceapi.tf
const weightsUrl = '/tmp/initial_tiny_yolov2_glorot_normal.weights'
const fromEpoch = 0
window.debug = false
window.logTrainSteps = false
// hyper parameters
window.objectScale = 5
window.noObjectScale = 1
window.coordScale = 1
window.saveEveryNthIteration = 1
window.trainSteps = 4000
//window.optimizer = tf.train.sgd(0.001)
window.optimizer = tf.train.adam(0.001, 0.9, 0.999, 1e-8)
// all samples
const numTrainSamples = Infinity
async function loadNetWeights(uri) {
return new Float32Array(await (await fetch(uri)).arrayBuffer())
}
async function fetchDetectionFilenames() {
return fetch('/detection_filenames').then(res => res.json())
}
async function run() {
const weights = await loadNetWeights(weightsUrl)
window.net = new faceapi.TinyYolov2(true)
window.net.load(weights)
window.net.variable()
window.detectionFilenames = (await fetchDetectionFilenames()).slice(0, numTrainSamples)
window.lossMap = {}
console.log('ready')
}
const trainSizes = [320]
function logLossChange(lossType) {
const { currentLoss, prevLoss, detectionFilenames } = window
log(`${lossType} : ${faceapi.round(currentLoss[lossType])} (avg: ${faceapi.round(currentLoss[lossType] / detectionFilenames.length)}) (delta: ${currentLoss[lossType] - prevLoss[lossType]})`)
}
async function train(batchSize = 1) {
for (let i = fromEpoch; i < trainSteps; i++) {
log('step', i)
let ts = Date.now()
const batchCreators = createBatchCreators(shuffle(window.detectionFilenames), batchSize)
for (let s = 0; s < trainSizes.length; s++) {
let ts2 = Date.now()
await trainStep(batchCreators, trainSizes[s])
ts2 = Date.now() - ts2
//log('train for size %s done (%s ms)', trainSizes[s], ts2)
}
ts = Date.now() - ts
log()
log('--------------------')
log()
log('step %s done (%s ms)', i, ts)
window.prevLoss = window.currentLoss
window.currentLoss = Object.keys(lossMap)
.map(filename => lossMap[filename])
.reduce((accumulatedLosses, losses) =>
Object.keys(losses)
.map(key => ({
[key]: (accumulatedLosses[key] || 0) + losses[key]
}))
.reduce((map, curr) => ({ ...map, ...curr }), {}),
{}
)
if (window.prevLoss) {
logLossChange('noObjectLoss')
logLossChange('objectLoss')
logLossChange('coordLoss')
logLossChange('totalLoss')
}
log()
log('--------------------')
log()
if (((i + 1) % saveEveryNthIteration) === 0) {
saveWeights(window.net, 'adam_511_' + (i + 1) + '.weights')
}
}
}
run()
</script>
</body>
</html>
\ No newline at end of file
tools/train/tinyYolov2/train.html
......@@ -17,22 +17,32 @@
<script>
tf = faceapi.tf
const weightsUrl = '/tmp/.weights'
const startIdx224 = 3220
const startIdx320 = 20688
const startIdx416 = 950
const startIdx608 = 15220
const weightsUrl = `/tmp/tmp__224_${startIdx224}__320_${startIdx320}__416_${startIdx416}__608_${startIdx608}.weights`
const fromEpoch = 0
window.debug = false
window.logTrainSteps = true
// hyper parameters
window.objectScale = 5
window.noObjectScale = 0.5
window.noObjectScale = 1
window.coordScale = 1
window.saveEveryNthIteration = 1
window.trainSteps = 100
//window.optimizer = tf.train.sgd(learningRate)
const rescaleEveryNthBatch = Infinity
window.saveEveryNthDataIdx = 100
window.trainSteps = 4000
//window.optimizer = tf.train.sgd(0.001)
window.optimizer = tf.train.adam(0.001, 0.9, 0.999, 1e-8)
function lossFunction(labels, out) {
return tf.losses.meanSquaredError(labels, out)
}
// all samples
const numTrainSamples = Infinity
async function loadNetWeights(uri) {
return new Float32Array(await (await fetch(uri)).arrayBuffer())
......@@ -47,36 +57,69 @@
window.net = new faceapi.TinyYolov2(true)
window.net.load(weights)
window.net.variable()
window.detectionFilenames = await fetchDetectionFilenames()
window.detectionFilenames = (await fetchDetectionFilenames()).slice(0, numTrainSamples)
window.lossMap = {}
console.log('ready')
}
const trainSizes = [608, 416, 320, 224]
//const trainSizes = [224, 320, 416, 608]
const trainSizes = [608]
function logLossChange(lossType) {
const { currentLoss, prevLoss, detectionFilenames } = window
log(`${lossType} : ${faceapi.round(currentLoss[lossType])} (avg: ${faceapi.round(currentLoss[lossType] / detectionFilenames.length)}) (delta: ${currentLoss[lossType] - prevLoss[lossType]})`)
}
function onBatchProcessed(dataIdx, inputSize) {
if (((dataIdx + 1) % saveEveryNthDataIdx) === 0) {
saveWeights(window.net, `tmp__224_${startIdx224}__320_${startIdx320}__416_${startIdx416}__608_${startIdx608 + dataIdx + 1}.weights`)
}
}
async function train(batchSize = 1) {
for (let i = 0; i < trainSteps; i++) {
console.log('step', i)
let ts = Date.now()
async function train() {
const batchCreators = createBatchCreators(shuffle(window.detectionFilenames), batchSize)
const batchSize = 1
for (let s = 0; s < trainSizes.length; s++) {
let ts2 = Date.now()
for (let i = fromEpoch; i < trainSteps; i++) {
log('step', i)
let ts2 = Date.now()
await trainStep(batchCreators, trainSizes[s])
const batchCreators = createBatchCreators(shuffle(window.detectionFilenames), batchSize)
ts2 = Date.now() - ts2
console.log('train for size %s done (%s ms)', trainSizes[s], ts2)
}
await trainStep(batchCreators, trainSizes, rescaleEveryNthBatch, onBatchProcessed)
ts = Date.now() - ts
console.log('step %s done (%s ms)', i, ts)
ts2 = Date.now() - ts2
}
if (((i + 1) % saveEveryNthIteration) === 0) {
saveWeights(window.net, 'tiny_yolov2_separable_model_' + i)
}
ts = Date.now() - ts
log()
log('--------------------')
log()
log('step %s done (%s ms)', i, ts)
window.prevLoss = window.currentLoss
window.currentLoss = Object.keys(lossMap)
.map(filename => lossMap[filename])
.reduce((accumulatedLosses, losses) =>
Object.keys(losses)
.map(key => ({
[key]: (accumulatedLosses[key] || 0) + losses[key]
}))
.reduce((map, curr) => ({ ...map, ...curr }), {}),
{}
)
if (window.prevLoss) {
logLossChange('noObjectLoss')
logLossChange('objectLoss')
logLossChange('coordLoss')
logLossChange('totalLoss')
}
log()
log('--------------------')
log()
}
run()
......
tools/train/tinyYolov2/train.js
async function trainStep(batchCreators, inputSize) {
await promiseSequential(batchCreators.map((batchCreator, dataIdx) => async () => {
// TODO: skip if groundTruthBoxes are too tiny
const { imgs, groundTruthBoxes, filenames } = await batchCreator()
const batchInput = (await faceapi.toNetInput(imgs)).managed()
let ts = Date.now()
const loss = optimizer.minimize(() => {
// TBD: batch loss
const batchIdx = 0
const outTensor = window.net.forwardInput(batchInput, inputSize)
const {
noObjectLoss,
objectLoss,
coordLoss,
totalLoss
} = computeLoss(
outTensor,
groundTruthBoxes[batchIdx],
batchInput.getReshapedInputDimensions(batchIdx),
batchInput.getRelativePaddings(batchIdx)
)
const losses = {
totalLoss: totalLoss.dataSync()[0],
noObjectLoss: noObjectLoss.dataSync()[0],
objectLoss: objectLoss.dataSync()[0],
coordLoss: coordLoss.dataSync()[0]
}
const batchIdx = 0
if (window.logTrainSteps) {
log(`ground truth boxes: ${groundTruthBoxes[batchIdx].length}`)
log(`noObjectLoss[${dataIdx}]: ${losses.noObjectLoss}`)
log(`objectLoss[${dataIdx}]: ${losses.objectLoss}`)
log(`coordLoss[${dataIdx}]: ${losses.coordLoss}`)
log(`totalLoss[${dataIdx}]: ${losses.totalLoss}`)
function minimize(groundTruthBoxes, batchInput, inputSize, batch) {
const filename = batch.filenames[batchIdx]
const { dataIdx } = batch
if (window.lossMap[filenames]) {
log(`loss change: ${losses.totalLoss - window.lossMap[filenames].totalLoss}`)
}
}
return optimizer.minimize(() => {
window.lossMap[filenames] = losses
const outTensor = window.net.forwardInput(batchInput, inputSize)
return totalLoss
}, true)
const {
noObjectLoss,
objectLoss,
coordLoss,
totalLoss
} = computeLoss(
outTensor,
groundTruthBoxes,
batchInput.getReshapedInputDimensions(batchIdx),
batchInput.getRelativePaddings(batchIdx)
)
const losses = {
totalLoss: totalLoss.dataSync()[0],
noObjectLoss: noObjectLoss.dataSync()[0],
objectLoss: objectLoss.dataSync()[0],
coordLoss: coordLoss.dataSync()[0]
}
const lossKey = `${filename}_${inputSize}`
ts = Date.now() - ts
if (window.logTrainSteps) {
log(`trainStep time for dataIdx ${dataIdx} (${inputSize}): ${ts} ms (${ts / batchInput.batchSize} ms / batch element)`)
log(`ground truth boxes: ${groundTruthBoxes.length}`)
log(`noObjectLoss[${dataIdx}]: ${losses.noObjectLoss}`)
log(`objectLoss[${dataIdx}]: ${losses.objectLoss}`)
log(`coordLoss[${dataIdx}]: ${losses.coordLoss}`)
log(`totalLoss[${dataIdx}]: ${losses.totalLoss}`)
if (window.lossMap[lossKey]) {
log(`loss change: ${losses.totalLoss - window.lossMap[lossKey].totalLoss}`)
}
}
window.lossMap[lossKey] = losses
return totalLoss
}, true)
}
async function trainStep(batchCreators, inputSizes, rescaleEveryNthBatch, onBatchProcessed = () => {}) {
async function step(currentBatchCreators) {
if (!currentBatchCreators.batchCreators.length) {
return
}
loss.dispose()
await tf.nextFrame()
//console.log(tf.memory())
}))
await promiseSequential(inputSizes.map(inputSize => async () => {
await promiseSequential(currentBatchCreators.batchCreators.map(batchCreator => async () => {
const batch = await batchCreator()
const { imgs, groundTruthBoxes, filenames, dataIdx } = batch
const batchInput = await faceapi.toNetInput(imgs)
const [imgHeight, imgWidth] = batchInput.inputs[batchIdx].shape
// skip groundTruthBoxes, which are too tiny
const scaleFactor = inputSize / Math.max(imgHeight, imgWidth)
const filteredGroundTruthBoxes = groundTruthBoxes[batchIdx].filter(({ x, y, width, height }) => {
const box = (new faceapi.Rect(x, y, width, height))
.toBoundingBox()
.rescale({ height: imgHeight, width: imgWidth })
.rescale(scaleFactor)
const isTooTiny = box.width < 50 || box.height < 50
if (isTooTiny) {
log(`skipping box for input size ${inputSize}: (${Math.floor(box.width)} x ${Math.floor(box.height)})`)
}
return !isTooTiny
})
if (!filteredGroundTruthBoxes.length) {
log(`no boxes for input size ${inputSize}, ${groundTruthBoxes[batchIdx].length} boxes were too small`)
batchInput.dispose()
onBatchProcessed(dataIdx, inputSize)
return
}
let ts = Date.now()
const loss = minimize(filteredGroundTruthBoxes, batchInput, inputSize, batch)
ts = Date.now() - ts
if (window.logTrainSteps) {
log(`trainStep time for dataIdx ${dataIdx} (${inputSize}): ${ts} ms`)
}
loss.dispose()
batchInput.dispose()
onBatchProcessed(dataIdx, inputSize)
await tf.nextFrame()
}))
}))
await step(currentBatchCreators.next(rescaleEveryNthBatch))
}
await step(batchCreators.next(rescaleEveryNthBatch))
}
function createBatchCreators(detectionFilenames, batchSize) {
function createBatchCreators(detectionFilenames, batchSize, ) {
if (batchSize < 1) {
throw new Error('invalid batch size: ' + batchSize)
}
......@@ -75,7 +125,7 @@ function createBatchCreators(detectionFilenames, batchSize) {
pushToBatch(detectionFilenames)
const batchCreators = batches.map(filenamesForBatch => async () => {
const batchCreators = batches.map((filenamesForBatch, dataIdx) => async () => {
const groundTruthBoxes = await Promise.all(filenamesForBatch.map(
file => fetch(file).then(res => res.json())
))
......@@ -87,9 +137,24 @@ function createBatchCreators(detectionFilenames, batchSize) {
return {
imgs,
groundTruthBoxes,
filenames: filenamesForBatch
filenames: filenamesForBatch,
dataIdx
}
})
return batchCreators
let idx = 0
function next(n) {
const nextBatchCreators = batchCreators.slice(idx, idx + n)
idx += n
return {
batchCreators: nextBatchCreators,
next
}
}
return {
data: batchCreators,
next
}
}
\ No newline at end of file
tools/train/tinyYolov2/verify.html
......@@ -138,7 +138,14 @@
async function run() {
$('#imgByNr').keydown(onKeyDown)
const weights = await loadNetWeights('/tmp/test_n100_320_114.weights')
const startIdx224 = 3220
const startIdx320 = 20688
const startIdx416 = 950
const startIdx608 = 15220
const weightsUrl = `/tmp/tmp__224_${startIdx224}__320_${startIdx320}__416_${startIdx416}__608_${startIdx608}.weights`
const weights = await loadNetWeights(weightsUrl)
window.net = new faceapi.TinyYolov2(true)
await window.net.load(weights)
......
Markdown is supported
0% or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment