Memory management.

[lambdacerro]

Hello,

I've started currently using Human for a project and so far is an awesome library :)

However i'm having an issue when dealing with memory leaks.

I can't get my app to stop leaking tensors even after disposing them, right now im doing it like this.

var tensor = tf.node.decodeImage(imageData);
        var output = await this.human.detect(tensor, {
            videoOptimized: false
        });

        tf.dispose(tensor);

However after every call some tensors are left behind and after some time the kernel kills the app due to high memory usage.

If I remove the detect call, the tensor count stays stable.

I dont know if there's something else I should do in order to avoid leaking tensors.

Thank you in advance!

[vladmandic]

can you open an issue for this and post a link to full reproduction as i cannot detect any tensor leaks?

i just tried with

const human = new Human();
await human.load();
await human.warmup();
const buffer = fs.readFileSync(inputFile);
console.log(human.tf.engine().memory().numTensors);
let tensor;
for (let i = 0; i < 100; i++) {
  tensor = human.tf.node.decodeImage(buffer);
  await human.detect(tensor);
  human.tf.dispose(tensor);
}
console.log(human.tf.engine().memory().numTensors);

[vladmandic]

(resolved via issue #232)