Closed crxporter closed 3 years ago
It has to be a buffered image. the API only accepts a buffered image
tf.node.decodeImage(buffer);
and see https://js.tensorflow.org/api/latest/#buffer
I have the user use node red's buffer instead of doing it myself like the example code below.
const buffer = fs.readFileSync(img);
as it gives the user more control to fetch the image from a url or severer and then buffer it using node-red.
I hope this helps with your question
Cool! Thanks for the extra detail. I'll carry on and try to find the quickest way to pass a buffer into the node.
also I'm a month or 2 away from releaseing a way better version of this node based on Human https://github.com/vladmandic/human
this should be way more accurate and a tad faster. see ---> https://github.com/vladmandic/face-api/discussions/26#discussioncomment-229336
I've been just so dang busy chasing the dollar of late to finish off the code
Cool! I'll definitely be watching.
I only loaded this node the first time a couple hours ago. I can get it to recognize faces on my pi 4 but I wasn't successful in having it recognize who's face it is yet. I loaded a folder full of my face but didn't get anything.
Maybe soon I'll be able to unlock the front door just by standing at the doorbell!
so to detect your face load in 5 images of your face into a folder with your name @ location of nodes/node-red-contrib-facial-recognition/example/labeled_face/your_name_here
then it should ID you
Cool I'll try it tomorrow
Would it be possible to have the input payload be a file or web path (my security camera images are located on LAN) rather than having to pass a buffer to the node? Loading and sending the buffer takes 2 or 3 times as long as this node does to process faces... If the input payload could just be a path to an image it might speed things up a lot!
Thanks for a cool node!