fixed landmark postprocessing + resize input canvases instead of tensors to net input size, which is much more performant

fixed landmark postprocessing + resize input canvases instead of tensors to net input size, which is much more performant

move tslib from devDeps to deps

Export ssdMobilenetv1 as alias of locateFaces -  no build

consume shared codebase from tfjs-tiny-yolov2 and tfjs-image-recognition-base

Tiny yolov2 seperable conv2d

weight loading of quantized tiny-yolov2-seperable-conv2d + use separable conv model by default + testcases

fixed image size overflowing in extractFaceTensors and extractFaces

fixed image size overflowing in extractFaceTensors and extractFaces when rectangle overlaps image borders + testcases

Adding support for external model uris