move python inference to Azaion.Inference folder

.gitignore

@@ -11,3 +11,4 @@ venv
 *.c
 *.pyd
 cython_debug*
+dist

Azaion.AI/inference.pyx

@@ -1,130 +0,0 @@
-from ultralytics import YOLO
-import mimetypes
-import cv2
-from ultralytics.engine.results import Boxes
-from remote_command cimport RemoteCommand
-from annotation cimport Detection, Annotation
-from secure_model cimport SecureModelLoader
-from ai_config cimport AIRecognitionConfig
-
-cdef class Inference:
-    def __init__(self, model_bytes, on_annotation):
-        loader = SecureModelLoader()
-        model_path = loader.load_model(model_bytes)
-        self.stop_signal = False
-        self.model = YOLO(<str>model_path)
-        self.on_annotation = on_annotation
-
-    cdef bint is_video(self, str filepath):
-        mime_type, _ = mimetypes.guess_type(<str>filepath)
-        return mime_type and mime_type.startswith("video")
-
-    cdef run_inference(self, RemoteCommand cmd, int batch_size=8):
-        print('run inference..')
-        self.stop_signal = False
-        if self.is_video(cmd.filename):
-            self._process_video(cmd, batch_size)
-        else:
-            self._process_image(cmd)
-
-    cdef _process_video(self, RemoteCommand cmd, int batch_size):
-        frame_count = 0
-        batch_frame = []
-        self._previous_annotation = None
-        v_input = cv2.VideoCapture(<str>cmd.filename)
-        self.ai_config = AIRecognitionConfig.from_msgpack(cmd.data)
-
-        while v_input.isOpened() and not self.stop_signal:
-            ret, frame = v_input.read()
-            ms = v_input.get(cv2.CAP_PROP_POS_MSEC)
-            if not ret or frame is None:
-                break
-
-            frame_count += 1
-            if frame_count % self.ai_config.frame_period_recognition == 0:
-                batch_frame.append((frame, ms))
-
-            if len(batch_frame) == batch_size:
-                frames = list(map(lambda x: x[0], batch_frame))
-                results = self.model.track(frames, persist=True)
-
-                for frame, res in zip(batch_frame, results):
-                    annotation = self.frame_to_annotation(int(frame[1]), frame[0], res.boxes)
-
-                    is_valid = self.is_valid_annotation(<Annotation>annotation)
-                    print(f'Is valid annotation: {is_valid}')
-                    if is_valid:
-                        self._previous_annotation = annotation
-                        self.on_annotation(cmd, annotation)
-                batch_frame.clear()
-
-        v_input.release()
-
-    cdef _process_image(self, RemoteCommand cmd):
-        frame = cv2.imread(<str>cmd.filename)
-        res = self.model.track(frame)
-        annotation = self.frame_to_annotation(0, frame, res[0].boxes)
-        self.on_annotation(cmd, annotation)
-
-    cdef stop(self):
-        self.stop_signal = True
-
-    cdef frame_to_annotation(self, long time, frame, boxes: Boxes):
-        detections = []
-        for box in boxes:
-            b = box.xywhn[0].cpu().numpy()
-            cls = int(box.cls[0].cpu().numpy().item())
-            confidence = box.conf[0].cpu().numpy().item()
-            det = Detection(<double> b[0], <double> b[1], <double> b[2], <double> b[3], cls, confidence)
-            detections.append(det)
-        _, encoded_image = cv2.imencode('.jpg', frame)
-        image_bytes = encoded_image.tobytes()
-        return Annotation(image_bytes, time, detections)
-
-    cdef bint is_valid_annotation(self, Annotation annotation):
-        # No detections, invalid
-        if not annotation.detections:
-            return False
-
-        # First valid annotation, always accept
-        if self._previous_annotation is None:
-            return True
-
-        # Enough time has passed since last annotation
-        if annotation.time >= self._previous_annotation.time + <long>(self.ai_config.frame_recognition_seconds * 1000):
-            return True
-
-        # More objects detected than before
-        if len(annotation.detections) > len(self._previous_annotation.detections):
-            return True
-
-        cdef:
-            Detection current_det, prev_det
-            double dx, dy, distance_sq, min_distance_sq
-            Detection closest_det
-
-        # Check each detection against previous frame
-        for current_det in annotation.detections:
-            min_distance_sq = 1e18  # Initialize with large value
-            closest_det = None
-
-            # Find closest detection in previous frame
-            for prev_det in self._previous_annotation.detections:
-                dx = current_det.x - prev_det.x
-                dy = current_det.y - prev_det.y
-                distance_sq = dx * dx + dy * dy
-
-                if distance_sq < min_distance_sq:
-                    min_distance_sq = distance_sq
-                    closest_det = prev_det
-
-            # Check if beyond tracking distance
-            if min_distance_sq > self.ai_config.tracking_distance_confidence:
-                return True
-
-            # Check probability increase
-            if current_det.confidence >= closest_det.confidence + self.ai_config.tracking_probability_increase:
-                return True
-
-        # No validation criteria met
-        return False

Azaion.AI/token

@@ -1 +0,0 @@
-eyJhbGciOiJIUzI1NiIsInR5cCI6IkpXVCJ9.eyJuYW1laWQiOiJkOTBhMzZjYS1lMjM3LTRmYmQtOWM3Yy0xMjcwNDBhYzg1NTYiLCJ1bmlxdWVfbmFtZSI6ImFkbWluQGF6YWlvbi5jb20iLCJyb2xlIjoiQXBpQWRtaW4iLCJuYmYiOjE3MzgzNjUwMjksImV4cCI6MTczODM3OTQyOSwiaWF0IjoxNzM4MzY1MDI5LCJpc3MiOiJBemFpb25BcGkiLCJhdWQiOiJBbm5vdGF0b3JzL09yYW5nZVBpL0FkbWlucyJ9.5teWb-gnhRngV337u_0OyUQ-o2-plN7shrvvKUsckPw

Azaion.Annotator/Azaion.Annotator.csproj

@@ -27,7 +27,6 @@
       <PackageReference Include="SkiaSharp" Version="2.88.9" />
       <PackageReference Include="VideoLAN.LibVLC.Windows" Version="3.0.21" />
       <PackageReference Include="WindowsAPICodePack" Version="7.0.4" />
-      <PackageReference Include="YoloV8.Gpu" Version="5.0.4" />
     </ItemGroup>
 
     <ItemGroup>

Azaion.Common/DTO/Config/AIRecognitionConfig.cs

@@ -5,11 +5,13 @@ namespace Azaion.Common.DTO.Config;
 [MessagePackObject]
 public class AIRecognitionConfig
 {
-    [Key("FrameRecognitionSeconds")] public double FrameRecognitionSeconds { get; set; }
+    [Key(nameof(FramePeriodRecognition))] public int FramePeriodRecognition { get; set; }
+    [Key(nameof(FrameRecognitionSeconds))] public double FrameRecognitionSeconds { get; set; }
+    [Key(nameof(ProbabilityThreshold))] public double ProbabilityThreshold { get; set; }
 
-    [Key("TrackingDistanceConfidence")] public double TrackingDistanceConfidence { get; set; }
-    [Key("TrackingProbabilityIncrease")] public double TrackingProbabilityIncrease { get; set; }
-    [Key("TrackingIntersectionThreshold")] public double TrackingIntersectionThreshold { get; set; }
-    [Key("FramePeriodRecognition")] public int FramePeriodRecognition { get; set; }
-    [Key("Data")] public byte[] Data { get; set; }
+    [Key(nameof(TrackingDistanceConfidence))] public double TrackingDistanceConfidence { get; set; }
+    [Key(nameof(TrackingProbabilityIncrease))] public double TrackingProbabilityIncrease { get; set; }
+    [Key(nameof(TrackingIntersectionThreshold))] public double TrackingIntersectionThreshold { get; set; }
+
+    [Key(nameof(Data))] public byte[] Data { get; set; }
 }

Azaion.CommonSecurity/SecurityConstants.cs

@@ -23,4 +23,6 @@ public class SecurityConstants
     public const int ZMQ_PORT = 5127;
 
     #endregion SocketClient
+
+    public static string AzaionInferencePath = "azaion-inference.exe";
 }

Azaion.CommonSecurity/Services/PythonResourceLoader.cs

@@ -46,21 +46,19 @@ public class PythonResourceLoader : IResourceLoader, IAuthProvider
 
     private void StartPython( ApiConfig apiConfig, ApiCredentials credentials)
     {
-        //var inferenceExe = LoadPythonFile().GetAwaiter().GetResult();
-        string outputProcess = "";
-        string errorProcess = "";
-
-        var path = "azaion-inference.exe";
-        var arguments = $"-e {credentials.Email} -p {credentials.Password} -f {apiConfig.ResourcesFolder}";
         using var process = new Process();
-        process.StartInfo.FileName = path;
-        process.StartInfo.Arguments = arguments;
-        process.StartInfo.UseShellExecute = false;
-        process.StartInfo.RedirectStandardOutput = true;
-        process.StartInfo.RedirectStandardError = true;
-        //process.StartInfo.CreateNoWindow = true;
-        process.OutputDataReceived += (sender, e) => { if (e.Data != null) Console.WriteLine(e.Data); };
-        process.ErrorDataReceived += (sender, e) => { if (e.Data != null) Console.WriteLine(e.Data); };
+        process.StartInfo = new ProcessStartInfo
+        {
+            FileName = SecurityConstants.AzaionInferencePath,
+            Arguments = $"-e {credentials.Email} -p {credentials.Password} -f {apiConfig.ResourcesFolder}",
+            UseShellExecute = false,
+            RedirectStandardOutput = true,
+            RedirectStandardError = true,
+            //CreateNoWindow = true
+        };
+
+        process.OutputDataReceived += (_, e) => { if (e.Data != null) Console.WriteLine(e.Data); };
+        process.ErrorDataReceived += (_, e) => { if (e.Data != null) Console.WriteLine(e.Data); };
         process.Start();
     }
 

Azaion.Inference/README.md

@@ -45,11 +45,6 @@ This is crucial for the build because build needs Python.h header and other file
 
 ```
     python -m pip install --upgrade pip
-    pip install --upgrade huggingface_hub
-    pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu121
-    pip install ultralytics
-    
-    pip uninstall -y opencv-python
     pip install opencv-python cython msgpack cryptography rstream pika zmq pyjwt pyinstaller tensorboard
 ```
 In case of  fbgemm.dll  error (Windows specific):

Azaion.Inference/ai_config.pxd

@@ -1,9 +1,12 @@
 cdef class AIRecognitionConfig:
     cdef public double frame_recognition_seconds
+    cdef public int frame_period_recognition
+    cdef public double probability_threshold
+
     cdef public double tracking_distance_confidence
     cdef public double tracking_probability_increase
     cdef public double tracking_intersection_threshold
-    cdef public int frame_period_recognition
+
     cdef public bytes file_data
 
     @staticmethod

Azaion.Inference/ai_config.pyx

@@ -2,18 +2,24 @@ from msgpack import unpackb
 
 cdef class AIRecognitionConfig:
     def __init__(self,
+                 frame_period_recognition,
                  frame_recognition_seconds,
+                 probability_threshold,
+
                  tracking_distance_confidence,
                  tracking_probability_increase,
                  tracking_intersection_threshold,
-                 frame_period_recognition,
+
                  file_data
                  ):
+        self.frame_period_recognition = frame_period_recognition
         self.frame_recognition_seconds = frame_recognition_seconds
+        self.probability_threshold = probability_threshold
+
         self.tracking_distance_confidence = tracking_distance_confidence
         self.tracking_probability_increase = tracking_probability_increase
         self.tracking_intersection_threshold = tracking_intersection_threshold
-        self.frame_period_recognition = frame_period_recognition
+
         self.file_data = file_data
 
     def __str__(self):
@@ -24,9 +30,13 @@ cdef class AIRecognitionConfig:
     cdef from_msgpack(bytes data):
         unpacked = unpackb(data, strict_map_key=False)
         return AIRecognitionConfig(
+            unpacked.get("FramePeriodRecognition", 0),
             unpacked.get("FrameRecognitionSeconds", 0.0),
+            unpacked.get("ProbabilityThreshold", 0.0),
+
             unpacked.get("TrackingDistanceConfidence", 0.0),
             unpacked.get("TrackingProbabilityIncrease", 0.0),
             unpacked.get("TrackingIntersectionThreshold", 0.0),
-            unpacked.get("FramePeriodRecognition", 0),
+
+
             unpacked.get("Data", b''))

Azaion.Inference/annotation.pyx

@@ -13,10 +13,10 @@ cdef class Detection:
         return f'{self.cls}: {self.x:.2f} {self.y:.2f} {self.w:.2f} {self.h:.2f}, prob: {(self.confidence*100):.1f}%'
 
 cdef class Annotation:
-    def __init__(self, bytes image_bytes, long time, list[Detection] detections):
-        self.image = image_bytes
+    def __init__(self, long time, list[Detection] detections):
         self.time = time
         self.detections = detections if detections is not None else []
+        self.image = b''
 
     cdef bytes serialize(self):
         return msgpack.packb({

Azaion.Inference/build.cmd

@@ -1,12 +1,10 @@
-pyinstaller --onefile ^
+pyinstaller --onefile ^
 --collect-all jwt ^
 --collect-all requests ^
 --collect-all psutil ^
 --collect-all cryptography ^
 --collect-all msgpack ^
 --collect-all expecttest ^
---collect-all torch ^
---collect-all ultralytics ^
 --collect-all zmq ^
 --hidden-import user ^
 --hidden-import security ^
@@ -19,4 +17,6 @@
 --hidden-import ai_config ^
 --hidden-import inference ^
 --hidden-import remote_command_handler ^
+--hidden-import cv2 ^
+--hidden-import onnxruntime ^
 start.py

Azaion.Inference/constants.pyx

@@ -7,6 +7,6 @@ cdef str ANNOTATIONS_QUEUE = "azaion-annotations"
 cdef str API_URL = "https://api.azaion.com"  # Base URL for the external API
 cdef str TOKEN_FILE = "token"
 cdef str QUEUE_CONFIG_FILENAME = "secured-config.json"
-cdef str AI_MODEL_FILE = "azaion.pt"
+cdef str AI_MODEL_FILE = "azaion.onnx"
 
 cdef bytes DONE_SIGNAL = b"DONE"

Azaion.Inference/inference.pxd

@@ -3,17 +3,25 @@ from annotation cimport Annotation
 from ai_config cimport AIRecognitionConfig
 
 cdef class Inference:
-    cdef object model
+    cdef object session
     cdef object on_annotation
     cdef Annotation _previous_annotation
     cdef AIRecognitionConfig ai_config
     cdef bint stop_signal
 
+    cdef str model_input
+    cdef int model_width
+    cdef int model_height
+
     cdef bint is_video(self, str filepath)
     cdef run_inference(self, RemoteCommand cmd, int batch_size=?)
     cdef _process_video(self, RemoteCommand cmd, int batch_size)
     cdef _process_image(self, RemoteCommand cmd)
     cdef stop(self)
 
-    cdef frame_to_annotation(self, long time, frame, boxes: object)
+    cdef preprocess(self, frame)
+    cdef postprocess(self, output, int img_width, int img_height)
+
+
+    cdef detect_frame(self, frame, long time)
     cdef bint is_valid_annotation(self, Annotation annotation)

Azaion.Inference/inference.pyx

@@ -0,0 +1,188 @@
+import mimetypes
+import time
+
+import cv2
+import numpy as np
+import onnxruntime as onnx
+
+from remote_command cimport RemoteCommand
+from annotation cimport Detection, Annotation
+from ai_config cimport AIRecognitionConfig
+
+cdef class Inference:
+    def __init__(self, model_bytes, on_annotation):
+        self.stop_signal = False
+        self.session = onnx.InferenceSession(
+            model_bytes, providers=["CUDAExecutionProvider", "CPUExecutionProvider"]
+        )
+        self.on_annotation = on_annotation
+        self.ai_config = AIRecognitionConfig(4, 2, 0.25,  0.15, 15, 0.8, b'')
+        model_inputs = self.session.get_inputs()
+        self.model_input = model_inputs[0].name
+        input_shape = model_inputs[0].shape
+        self.model_width = input_shape[2]
+        self.model_height = input_shape[3]
+        print(f'AI detection model input: {self.model_input} ({self.model_width}, {self.model_height})')
+        model_meta = self.session.get_modelmeta()
+        print("Metadata:", model_meta.custom_metadata_map)
+
+    cdef preprocess(self, frame):
+        img = cv2.cvtColor(frame, cv2.COLOR_BGR2RGB)
+        img = cv2.resize(img, (self.model_width, self.model_height))
+        image_data = np.array(img) / 255.0
+        image_data = np.transpose(image_data, (2, 0, 1))  # Channel first
+        image_data = np.expand_dims(image_data, axis=0).astype(np.float32)
+        return image_data
+
+    cdef postprocess(self, output, int img_width, int img_height):
+        outputs = np.transpose(np.squeeze(output[0]))
+        rows = outputs.shape[0]
+
+        boxes = []
+        scores = []
+        class_ids = []
+
+        x_factor = img_width / self.model_width
+        y_factor = img_height / self.model_height
+
+        for i in range(rows):
+            classes_scores = outputs[i][4:]
+            max_score = np.amax(classes_scores)
+
+            if max_score >= self.ai_config.probability_threshold:
+                class_id = np.argmax(classes_scores)
+                x, y, w, h = outputs[i][0], outputs[i][1], outputs[i][2], outputs[i][3]
+
+                left = int((x - w / 2) * x_factor)
+                top = int((y - h / 2) * y_factor)
+                width = int(w * x_factor)
+                height = int(h * y_factor)
+
+                class_ids.append(class_id)
+                scores.append(max_score)
+                boxes.append([left, top, width, height])
+        indices = cv2.dnn.NMSBoxes(boxes, scores, self.ai_config.probability_threshold, 0.45)
+        detections = []
+        for i in indices:
+            x, y, w, h = boxes[i]
+            detections.append(Detection(x, y, w, h, class_ids[i], scores[i]))
+        return detections
+
+    cdef bint is_video(self, str filepath):
+        mime_type, _ = mimetypes.guess_type(<str>filepath)
+        return mime_type and mime_type.startswith("video")
+
+    cdef run_inference(self, RemoteCommand cmd, int batch_size=8):
+        print('run inference..')
+        self.ai_config = AIRecognitionConfig.from_msgpack(cmd.data)
+        self.stop_signal = False
+        if self.is_video(cmd.filename):
+            self._process_video(cmd, batch_size)
+        else:
+            self._process_image(cmd)
+
+    cdef _process_video(self, RemoteCommand cmd, int batch_size):
+        frame_count = 0
+        batch_frame = []
+        self._previous_annotation = None
+        self.start_video_time = time.time()
+
+        v_input = cv2.VideoCapture(<str>cmd.filename)
+        while v_input.isOpened():
+            ret, frame = v_input.read()
+            if not ret or frame is None:
+                break
+
+            frame_count += 1
+            if frame_count % self.ai_config.frame_period_recognition == 0:
+                ms = int(v_input.get(cv2.CAP_PROP_POS_MSEC))
+                annotation = self.detect_frame(frame, ms)
+                if annotation is not None:
+                    self._previous_annotation = annotation
+                    self.on_annotation(annotation)
+
+
+    cdef detect_frame(self, frame, long time):
+        cdef Annotation annotation
+        img_height, img_width = frame.shape[:2]
+
+        start_time = time.time()
+        img_data = self.preprocess(frame)
+        preprocess_time = time.time()
+        outputs = self.session.run(None, {self.model_input: img_data})
+        inference_time = time.time()
+        detections = self.postprocess(outputs, img_width, img_height)
+        postprocess_time = time.time()
+        print(f'video time, ms: {time / 1000:.3f}. total time, s : {postprocess_time - self.start_video_time:.3f} '
+              f'preprocess time: {preprocess_time - start_time:.3f}, inference time: {inference_time - preprocess_time:.3f},'
+              f' postprocess time: {postprocess_time - inference_time:.3f}, total time: {postprocess_time - start_time:.3f}')
+        if len(detections) > 0:
+            annotation = Annotation(frame, time, detections)
+            if self.is_valid_annotation(annotation):
+                _, image = cv2.imencode('.jpg', frame)
+                annotation.image = image.tobytes()
+                return annotation
+        return None
+
+
+    cdef _process_image(self, RemoteCommand cmd):
+        self._previous_annotation = None
+        frame = cv2.imread(<str>cmd.filename)
+        annotation = self.detect_frame(frame, 0)
+        if annotation is None:
+            _, image = cv2.imencode('.jpg', frame)
+            annotation = Annotation(frame, time, [])
+            annotation.image = image.tobytes()
+        self.on_annotation(cmd, annotation)
+
+
+    cdef stop(self):
+        self.stop_signal = True
+
+
+    cdef bint is_valid_annotation(self, Annotation annotation):
+        # No detections, invalid
+        if not annotation.detections:
+            return False
+
+        # First valid annotation, always accept
+        if self._previous_annotation is None:
+            return True
+
+        # Enough time has passed since last annotation
+        if annotation.time >= self._previous_annotation.time + <long>(self.ai_config.frame_recognition_seconds * 1000):
+            return True
+
+        # More objects detected than before
+        if len(annotation.detections) > len(self._previous_annotation.detections):
+            return True
+
+        cdef:
+            Detection current_det, prev_det
+            double dx, dy, distance_sq, min_distance_sq
+            Detection closest_det
+
+        # Check each detection against previous frame
+        for current_det in annotation.detections:
+            min_distance_sq = 1e18  # Initialize with large value
+            closest_det = None
+
+            # Find the closest detection in previous frame
+            for prev_det in self._previous_annotation.detections:
+                dx = current_det.x - prev_det.x
+                dy = current_det.y - prev_det.y
+                distance_sq = dx * dx + dy * dy
+
+                if distance_sq < min_distance_sq:
+                    min_distance_sq = distance_sq
+                    closest_det = prev_det
+
+            # Check if beyond tracking distance
+            if min_distance_sq > self.ai_config.tracking_distance_confidence:
+                return True
+
+            # Check probability increase
+            if current_det.confidence >= closest_det.confidence + self.ai_config.tracking_probability_increase:
+                return True
+
+        return False

Azaion.Inference/requirements.txt

@@ -0,0 +1,5 @@
+setuptools
+Cython
+opencv-python
+numpy
+onnxruntime-gpu

Azaion.Inference/setup.py

@@ -1,5 +1,6 @@
 from setuptools import setup, Extension
 from Cython.Build import cythonize
+import numpy as np
 
 extensions = [
     Extension('constants', ['constants.pyx']),
@@ -13,7 +14,6 @@ extensions = [
     Extension('secure_model', ['secure_model.pyx']),
     Extension('ai_config', ['ai_config.pyx']),
     Extension('inference', ['inference.pyx']),
-
     Extension('main', ['main.pyx']),
 ]
 

Azaion.Inference/start.spec

@@ -0,0 +1,57 @@
+# -*- mode: python ; coding: utf-8 -*-
+from PyInstaller.utils.hooks import collect_all
+
+datas = []
+binaries = []
+hiddenimports = ['user', 'security', 'secure_model', 'api_client', 'hardware_service', 'constants', 'annotation', 'remote_command', 'ai_config', 'inference', 'remote_command_handler', 'cv2', 'onnxruntime']
+tmp_ret = collect_all('jwt')
+datas += tmp_ret[0]; binaries += tmp_ret[1]; hiddenimports += tmp_ret[2]
+tmp_ret = collect_all('requests')
+datas += tmp_ret[0]; binaries += tmp_ret[1]; hiddenimports += tmp_ret[2]
+tmp_ret = collect_all('psutil')
+datas += tmp_ret[0]; binaries += tmp_ret[1]; hiddenimports += tmp_ret[2]
+tmp_ret = collect_all('cryptography')
+datas += tmp_ret[0]; binaries += tmp_ret[1]; hiddenimports += tmp_ret[2]
+tmp_ret = collect_all('msgpack')
+datas += tmp_ret[0]; binaries += tmp_ret[1]; hiddenimports += tmp_ret[2]
+tmp_ret = collect_all('expecttest')
+datas += tmp_ret[0]; binaries += tmp_ret[1]; hiddenimports += tmp_ret[2]
+tmp_ret = collect_all('zmq')
+datas += tmp_ret[0]; binaries += tmp_ret[1]; hiddenimports += tmp_ret[2]
+
+
+a = Analysis(
+    ['start.py'],
+    pathex=[],
+    binaries=binaries,
+    datas=datas,
+    hiddenimports=hiddenimports,
+    hookspath=[],
+    hooksconfig={},
+    runtime_hooks=[],
+    excludes=[],
+    noarchive=False,
+    optimize=0,
+)
+pyz = PYZ(a.pure)
+
+exe = EXE(
+    pyz,
+    a.scripts,
+    a.binaries,
+    a.datas,
+    [],
+    name='start',
+    debug=False,
+    bootloader_ignore_signals=False,
+    strip=False,
+    upx=True,
+    upx_exclude=[],
+    runtime_tmpdir=None,
+    console=True,
+    disable_windowed_traceback=False,
+    argv_emulation=False,
+    target_arch=None,
+    codesign_identity=None,
+    entitlements_file=None,
+)

Azaion.Inference/token

@@ -0,0 +1 @@
+eyJhbGciOiJIUzI1NiIsInR5cCI6IkpXVCJ9.eyJuYW1laWQiOiJkOTBhMzZjYS1lMjM3LTRmYmQtOWM3Yy0xMjcwNDBhYzg1NTYiLCJ1bmlxdWVfbmFtZSI6ImFkbWluQGF6YWlvbi5jb20iLCJyb2xlIjoiQXBpQWRtaW4iLCJuYmYiOjE3Mzg4Mjk0NTMsImV4cCI6MTczODg0Mzg1MywiaWF0IjoxNzM4ODI5NDUzLCJpc3MiOiJBemFpb25BcGkiLCJhdWQiOiJBbm5vdGF0b3JzL09yYW5nZVBpL0FkbWlucyJ9.t6ImX8KkH5IQ4zNNY5IbXESSI6uia4iuzyMhodvM7AA

Azaion.Suite/config.json

@@ -36,11 +36,13 @@
     "RightPanelWidth": 230.0
   },
   "AIRecognitionConfig": {
+    "FramePeriodRecognition": 4,
     "FrameRecognitionSeconds": 2.0,
+    "ProbabilityThreshold": 0.25,
+
     "TrackingDistanceConfidence": 0.15,
     "TrackingProbabilityIncrease": 15.0,
-    "TrackingIntersectionThreshold": 0.8,
-    "FramePeriodRecognition": 4
+    "TrackingIntersectionThreshold": 0.8
   },
   "ThumbnailConfig": { "Size": "240,135", "Border": 10 }
 }