Releases: ml5js/ml5-library
Releases · ml5js/ml5-library
v0.4.1
change log
- makes testing more efficient - removes redundant model instantiations
- adds
ml5.flipImage()
function to allow flipping of video and images - removes console.log from imageClassifier test
- fixes URLs to p5 web editor examples
- allows passing video to face-api .detect() and .detectSingle()
- fixes bug when passing in teachable machine URL
- adds test for
sentiment
- cleans up devOps scripts
v0.4.0
Major Notes:
- added
ml5.neuralNetwork()
- added
ml5.faceApi()
- added
ml5.kmeans()
- p5
preload()
is now supported for all features (exceptml5.neuralNetwork()
) - p5 instance mode can now be used with ml5
- updates posenet to v2.1.3
- updates mobilenet to v2.0.3
- updates tensorflow to 1.2.1
- updates DCGAN to accept latent vector
- Adds imageData handling for image inputs
- Adds support for calling
.segment()
and.segmentWithParts()
directly on video - changes documentation to use docsify directly in ml5-library rather than using website
Fixes:
- Fixes posenet parameter inputs
Additional Notes:
v0.3.1
Major Notes:
- Added
ml5.imageClassifier('DoodleNet)
by @yining1023 - allows for classifying doodles trained on Google's QuickDraw dataset - Added
ml5.soundClassifier()
by @yining1023 - allows for use of the SpeechCommands - https://github.com/tensorflow/tfjs-models/tree/master/speech-commands - as well as using trained audio from google's Teachable Machine. - Added support for models trained in Google's new Teachable Machine as part of
ml5.imageClassifier()
andml5.soundClassifier()
by @yining1023 - Added BodyPix - https://github.com/tensorflow/tfjs-models/tree/master/body-pix by @joeyklee
- Added all-contributors for highlighting ml5 contributors
Models Added/Updates:
- Added
ml5.imageClassifier('DoodleNet)
by @yining1023 - Added
ml5.soundClassifier()
by @yining1023 - Added
ml5.bodyPix()
by @joeyklee - Updated DCGAN model loading and support for
preload() in p5js
. Now add amanifest.json
with the url of your model. by @joeyklee
Examples added:
- Added examples for
ml5.imageClassifier()
for: doodleNet - Canvas, doodleNet - video, loading custom model from Teachable Machine - added examples for
ml5.soundClassification()
for: speechCommands and speechCommands loaded from custom model - Added [BodyPix] Examples - https://github.com/ml5js/ml5-examples/tree/development/p5js/BodyPix
- Updated DCGAN example with urban aerial image generator
Handy things:
- @shiffman and @handav are presenting at EyeO 2019 - http://eyeofestival.com/, http://eyeofestival.com/speaker/hannah-davis/, http://eyeofestival.com/speaker/daniel-shiffman/
- New ml5.js website alpha version is launched - https://ml5js.org
v0.3.0
Major notes:
- Updated tensorflow.js dependency to v1.1.2 via @WenheLI and @joeyklee
- includes update to @magenta/sketch to v0.2.0 via @joeyklee
- FeatureExtractor takes number of labels as input to
.classification()
allowing users to specify how many labels they want to reclassify: #376 via @joeyklee , @yining1023, and @shiffman
Models added:
- UNET - learn more: https://towardsdatascience.com/u-net-b229b32b4a71 by @zaid
- CVAE - conditional variational auto encoder - learn more: https://wiseodd.github.io/techblog/2016/12/17/conditional-vae/ by @WenheLI
- Sentiment - learn more: https://www.tensorflow.org/tutorials/keras/basic_text_classification via @itay
- DCGAN - Deep Convolutional Generative Adversarial Networks - https://github.com/carpedm20/DCGAN-tensorflow
Examples added:
- UNET - https://github.com/ml5js/ml5-examples/tree/development/p5js/UNET/UNET_webcam by @joeyklee & @shiffman
- CVAE - https://github.com/ml5js/ml5-examples/tree/development/p5js/CVAE by @wenhe li, @joeyklee, & @shiffman
- Sentiment - https://github.com/ml5js/ml5-examples/tree/development/p5js/Sentiment @itay & @joeyklee & @shiffman
- DCGAN - https://github.com/ml5js/ml5-examples/tree/development/p5js/DCGAN @viztopia & @joeyklee
Handy things:
- Adding custom names to saved feature extractor - see: #353
- Added p5Utils - now we have a suite of functions for tighter integration of ml5 to handle p5 images: #358 @wenhe li
- CharRNN tests now much faster: #349 by @brent
- Begin to add
jsdoc
style comments in code for better documentation: #378 via @WenheLI and @joeyklee - @magenta/sketch updated its tensorflow dependency to 1.0.2 making it compatible with our update
Fixes:
- PoseNet
.singlePose()
and.multiPose()
now return consistent results #381 - Adds check for
mozCaptureStream
in YOLO: #375
Thanks all for your hard work + contributions! Have a nice weekend!
V0.2.3
v0.2.2
v0.2.2
Release: ml5.js v0.2.2 🎉
API updates:
- PoseNet now supports accessing parts of the body by name: #293
- ml5 now supports p5.js's
preload
function for specified models: #313 - ImageClassifier and FeatureExtractor returns
label
andconfidence
: #292 - YOLO is now returns
label
andconfidence
vs.className
andclassProb
: #316
Development:
- Added tests to CharRNN - #307
- Added functionality to add a manual test folder while experimenting with ml5: #295
- We're sunsetting the use of
master
as a branch name convention and specifyingdevelopment
orrelease
for changes: #319 - added documentation on CONTRIBUTING: #315
- solves linting support issues for vsCode: #324
Fixes:
0.0.5
0.2.1
0.1.1
0.0.8
Merge pull request #173 from meiamsome/testing/word2vec Word2Vec tests