opencv: beta

plugins/opencv/package.json

@@ -36,5 +36,5 @@
    "devDependencies": {
       "@scrypted/sdk": "file:../../sdk"
    },
-   "version": "0.0.69"
+   "version": "0.0.70"
 }

plugins/opencv/src/requirements.txt

@@ -9,3 +9,4 @@ imutils>=0.5.0
 av>=10.0.0; sys_platform != 'linux' or platform_machine == 'x86_64' or platform_machine == 'aarch64'
 # not available on armhf
 opencv-python; sys_platform != 'linux' or platform_machine == 'x86_64' or platform_machine == 'aarch64'
+

plugins/tensorflow-lite/src/predict/__init__.py

@@ -13,7 +13,6 @@
 
 from detect import DetectionSession, DetectPlugin
 
-from .sort_oh import tracker
 import numpy as np
 import traceback
 
@@ -24,14 +23,12 @@
 
 class PredictSession(DetectionSession):
     image: Image.Image
-    tracker: sort_oh.tracker.Sort_OH
 
     def __init__(self, start_time: float) -> None:
         super().__init__()
         self.image = None
         self.processed = 0
         self.start_time = start_time
-        self.tracker = None
 
 def parse_label_contents(contents: str):
     lines = contents.splitlines()
@@ -121,7 +118,6 @@ def __init__(self, PLUGIN_MIME_TYPE: str, nativeId: str | None = None):
         self.toMimeType = scrypted_sdk.ScryptedMimeTypes.MediaObject.value
 
         self.crop = False
-        self.trackers: Mapping[str, tracker.Sort_OH] = {}
 
         # periodic restart because there seems to be leaks in tflite or coral API.
         loop = asyncio.get_event_loop()
@@ -210,23 +206,7 @@ def getModelSettings(self, settings: Any = None) -> list[Setting]:
             ],
         }
 
-        trackerWindow: Setting = {
-            'title': 'Tracker Window',
-            'subgroup': 'Advanced',
-            'description': 'Internal Setting. Do not change.',
-            'key': 'trackerWindow',
-            'value': 3,
-            'type': 'number',
-        }
-        trackerCertainty: Setting = {
-            'title': 'Tracker Certainty',
-            'subgroup': 'Advanced',
-            'description': 'Internal Setting. Do not change.',
-            'key': 'trackerCertainty',
-            'value': .2,
-            'type': 'number',
-        }
-        return [allowList, trackerWindow, trackerCertainty]
+        return [allowList]
 
     def create_detection_result(self, objs: List[Prediction], size, allowList, convert_to_src_size=None) -> ObjectsDetected:
         detections: List[ObjectDetectionResult] = []
@@ -400,24 +380,6 @@ async def run_detection_image(self, detection_session: PredictSession, image: Im
         (w, h) = self.get_input_size() or image.size
         (iw, ih) = image.size
 
-        if detection_session and not detection_session.tracker:
-            t = self.trackers.get(detection_session.id)
-            if not t:
-                t = tracker.Sort_OH(scene=np.array([iw, ih]))
-                trackerCertainty = settings.get('trackerCertainty')
-                if not isinstance(trackerCertainty, int):
-                    trackerCertainty = .2
-                t.conf_three_frame_certainty = trackerCertainty * 3
-                trackerWindow = settings.get('trackerWindow')
-                if not isinstance(trackerWindow, int):
-                    trackerWindow = 3
-                t.conf_unmatched_history_size = trackerWindow
-                self.trackers[detection_session.id] = t
-            detection_session.tracker = t
-            # conf_trgt = 0.35
-            # conf_objt = 0.75
-            # detection_session.tracker.conf_trgt = conf_trgt
-            # detection_session.tracker.conf_objt = conf_objt
 
         # this a single pass or the second pass. detect once and return results.
         if multipass_crop:
@@ -543,61 +505,11 @@ def is_same_detection_middle(d1: ObjectDetectionResult, d2: ObjectDetectionResul
             ret = ret1
             ret['detections'] = dedupe_detections(ret1['detections'] + ret2['detections'], is_same_detection=is_same_detection_middle)
 
-        if detection_session:
-            self.track(detection_session, ret)
-
         if not len(ret['detections']):
             return ret, RawImage(image)
 
         return ret, RawImage(image)
 
-    def track(self, detection_session: PredictSession, ret: ObjectsDetected):
-        detections = ret['detections']
-        sort_input = []
-        for d in ret['detections']:
-            r: ObjectDetectionResult = d
-            l, t, w, h = r['boundingBox']
-            sort_input.append([l, t, l + w, t + h, r['score']])
-        trackers, unmatched_trckr, unmatched_gts = detection_session.tracker.update(np.array(sort_input), [])
-        for td in trackers:
-            x0, y0, x1, y1, trackID = td[0].item(), td[1].item(
-            ), td[2].item(), td[3].item(), td[4].item()
-            slop = 0
-            obj: ObjectDetectionResult = None
-            ta = (x1 - x0) * (y1 - y0)
-            box = Rectangle(x0, y0, x1, y1)
-            for d in detections:
-                if d.get('id'):
-                    continue
-                ob: ObjectDetectionResult = d
-                dx0, dy0, dw, dh = ob['boundingBox']
-                dx1 = dx0 + dw
-                dy1 = dy0 + dh
-                da = dw * dh
-                area = intersect_area(Rectangle(dx0, dy0, dx1, dy1), box)
-                if not area:
-                    continue
-                # intersect area always gonna be smaller than
-                # the detection or tracker area.
-                # greater numbers, ie approaching 2, is better.
-                dslop = area / ta + area / da
-                if (dslop > slop):
-                    slop = dslop
-                    obj = ob
-            if obj:
-                obj['id'] = str(trackID)
-            # this may happen if tracker predicts something is still in the scene
-            # but was not detected
-            # else:
-            #     print('unresolved tracker')
-        # for d in detections:
-        #     if not d.get('id'):
-        #         # this happens if the tracker is not confident in a new detection yet due
-        #         # to low score or has not been found in enough frames
-        #         if d['className'] == 'person':
-        #             print('untracked %s: %s' % (d['className'], d['score']))
-
-
     async def run_detection_crop(self, detection_session: DetectionSession, sample: RawImage, settings: Any, src_size, convert_to_src_size, bounding_box: Tuple[float, float, float, float]) -> ObjectsDetected:
         (ret, _) = await self.run_detection_image(detection_session, sample.image, settings, src_size, convert_to_src_size, bounding_box)
         return ret

plugins/tensorflow-lite/src/requirements.txt

@@ -7,10 +7,4 @@ Pillow>=5.4.1; sys_platform != 'linux' or platform_machine != 'x86_64'
 pillow-simd; sys_platform == 'linux' and platform_machine == 'x86_64'
 pycoral~=2.0
 PyGObject>=3.30.4; sys_platform != 'win32'
-# libav doesnt work on arm7
-av>=10.0.0; sys_platform != 'linux' or platform_machine == 'x86_64' or platform_machine == 'aarch64'
 tflite-runtime==2.5.0.post1
-
-# sort_oh
-scipy
-filterpy

plugins/tensorflow-lite/src/tflite/__init__.py

@@ -21,6 +21,7 @@
 import concurrent.futures
 import queue
 import asyncio
+from time import time
 
 def parse_label_contents(contents: str):
     lines = contents.splitlines()
@@ -116,6 +117,7 @@ async def detect_once(self, input: Image.Image, settings: Any, src_size, cvss):
         def predict():
             interpreter = self.interpreters.get()
             try:
+                print('predict s %s'  % time())
                 common.set_input(
                     interpreter, input)
                 scale = (1, 1)
@@ -131,6 +133,7 @@ def predict():
                 raise e
             finally:
                 self.interpreters.put(interpreter)
+                print('predict e %s'  % time())
 
         objs = await asyncio.get_event_loop().run_in_executor(self.executor, predict)