J3698

def main():
    clear_exports()
    export_test()
    optimize_model("test")
    convert_blob("test")
    pipeline = setup_pipeline()
    run_pipeline(pipeline)

J3698

class Test(nn.Module):
    def __init__(self):
        super().__init__()

    def forward(self, x): 
        return x / 2

J3698

def export_test():
    image = torch.ones((1, 3, 256, 256), requires_grad = False)
    model = Test()

    torch.onnx.export(model, image, "exports/test.onnx",
                      export_params = True,
                      opset_version = 11, 
                      do_constant_folding = True,
                      input_names = ['input'],
                      output_names = ['output']

J3698

def optimize_model(name):
    optimizer_path = "/opt/intel/openvino_2020.1.023/" + \
                     "deployment_tools/model_optimizer/mo.py"
    subprocess.run(["python3", optimizer_path,
                    "--input_model", f"./exports/{name}.onnx",
                    "--data_type", "FP16"])
    os.rename(name + ".bin", os.path.join("./exports", name + ".bin"))
    os.rename(name + ".xml", os.path.join("./exports", name + ".xml"))
    os.rename(name + ".mapping", os.path.join("./exports", name + ".mapping"))

J3698

import blobconverter

def convert_blob(name):
    xmlfile = os.path.join("./exports", name) + ".xml"
    binfile = os.path.join("./exports", name) + ".bin"

    blob_path = blobconverter.from_openvino(
        xml = xmlfile,
        bin = binfile,
        data_type = "FP16",

J3698

def setup_pipeline():
    pipeline = depthai.Pipeline()

    cam_rgb = pipeline.createColorCamera()
    cam_rgb.setPreviewSize(256, 256)
    cam_rgb.setInterleaved(False)

    detection_nn = pipeline.createNeuralNetwork()
    detection_nn.setBlobPath("./exports/test_openvino_2021.4_5shave.blob")
    cam_rgb.preview.link(detection_nn.input)

J3698

def run_pipeline(pipeline):
    with depthai.Device(pipeline) as device:
        q_nn = device.getOutputQueue("nn", maxSize=4, blocking=False)
        while True:
            in_nn = q_nn.tryGet()
            if in_nn is not None:
                # get data
                output = in_nn.getAllLayerNames()[-1]
                data = np.array(in_nn.getLayerFp16(output))

J3698

from flask import Flask

app = Flask(__name__)

@app.route("/classify")
def classify_symbol():
    return {"top5": ["A", "B", "C", "D", "E"]}

J3698

function addClassifyRequestInterval() {
    setInterval(function() {
        if (!shouldUpdate) {
            return
        }
        shouldUpdate = false;

        var xhttp = new XMLHttpRequest();
        xhttp.onreadystatechange = function() {
            if (xhttp.readyState == 4 && xhttp.status == 200) {

J3698

function addReTypesetHandler() {
    var script = document.createElement("script");
    script.innerHTML = ` 
        last = "1";
        setInterval(function() {
            var tag = document.getElementsByClassName("invisible")[0];
            if (tag.innerHTML != last) {
                MathJax.Hub.Queue(["Typeset", MathJax.Hub]);
                last = tag.innerHTML;
            }