0.6.0

features:

• implemented batch processing of images for FaceLandmarkNet and FaceRecognitionNet

fixes:

• pad face landmark net input to square and center to prevent stretching of input image #33
• fixed some memory leaks

breaking changes:

• NetInput class now holds array of images as tf.Tensor3D instead of HTMLCanvasElement -> use NetInput.inputs instead of NetInput.canvases from now on to retrieve the inputs

0.5.0

fixes:

• wait for input media to finish loading before passing them into the net, to avoid passing in empty image tensors
• fixed issue with padToSquare #29

other:

• bumped tfjs-core version

breaking changes:

• use await faceapi.toNetInput(input) instead of new faceapi.NetInput(input), to ensure all media inputs are loaded
• forward methods of FaceDetectionNet and FaceRecognitionNet now return Promises

0.4.0

• added faceapi.allFaces(img) to retrieve face locations, landmarks and descriptor for all faces of an input image
• most of the net methods can now be called on the faceapi object, without having to manage net instances

0.3.1

• compile tsc to commonjs
• seperate tsc build from compiled bundle

0.3.0

model weights have been quantized, to reduce the model size by ~75%:

• face detection model: 21.7 MB -> 5.4 MB
• face recognition model: 28.7 MB -> 7.0 MB
• face landmark model: 21.9 MB -> 6.2 MB

plus model weights are sharded in chunks of 4 MB to allow them to be cached in the browser

0.2.0

implemented face alignment from 68 point face landmarks, which allows to compute a much more accurate face descriptor

0.1.0

68 point face landmark detection

0.0.0

First release