Merge pull request #201 from jodur/master

Add image scale option

README.md

@@ -58,6 +58,7 @@ Configuration variables:
 - **roi_x_max**: (optional, default 1), range 0-1, must be more than roi_x_min
 - **roi_y_min**: (optional, default 0), range 0-1, must be less than roi_y_max
 - **roi_y_max**: (optional, default 1), range 0-1, must be more than roi_y_min
+-  **scale**: (optional, default 1.0), range 0.1-1.0, scaling factor for the images that are pulled from the camera stream. This results in smaller images files, especially when using  hires-cameras.
 - **source**: Must be a camera.
 - **targets**: The list of target object names and/or `object_type`, default `person`. Optionally a `confidence` can be set for this target, if not the default confidence is used. Note the minimum possible confidence is 10%.
 

custom_components/deepstack_object/image_processing.py

@@ -76,6 +76,7 @@
 CONF_ROI_X_MIN = "roi_x_min"
 CONF_ROI_Y_MAX = "roi_y_max"
 CONF_ROI_X_MAX = "roi_x_max"
+CONF_SCALE = "scale"
 CONF_CUSTOM_MODEL = "custom_model"
 
 DATETIME_FORMAT = "%Y-%m-%d_%H-%M-%S"
@@ -86,6 +87,7 @@
 DEFAULT_ROI_Y_MAX = 1.0
 DEFAULT_ROI_X_MIN = 0.0
 DEFAULT_ROI_X_MAX = 1.0
+DEAULT_SCALE = 1.0
 DEFAULT_ROI = (
     DEFAULT_ROI_Y_MIN,
     DEFAULT_ROI_X_MIN,
@@ -127,6 +129,9 @@
         vol.Optional(CONF_ROI_X_MIN, default=DEFAULT_ROI_X_MIN): cv.small_float,
         vol.Optional(CONF_ROI_Y_MAX, default=DEFAULT_ROI_Y_MAX): cv.small_float,
         vol.Optional(CONF_ROI_X_MAX, default=DEFAULT_ROI_X_MAX): cv.small_float,
+        vol.Optional(CONF_SCALE, default=DEAULT_SCALE): vol.All(
+            vol.Coerce(float, vol.Range(min=0.1, max=1))
+        ),
         vol.Optional(CONF_SAVE_FILE_FOLDER): cv.isdir,
         vol.Optional(CONF_SAVE_TIMESTAMPTED_FILE, default=False): cv.boolean,
         vol.Optional(CONF_SHOW_BOXES, default=True): cv.boolean,
@@ -223,6 +228,7 @@ def setup_platform(hass, config, add_devices, discovery_info=None):
             roi_x_min=config[CONF_ROI_X_MIN],
             roi_y_max=config[CONF_ROI_Y_MAX],
             roi_x_max=config[CONF_ROI_X_MAX],
+            scale=config[CONF_SCALE],
             show_boxes=config[CONF_SHOW_BOXES],
             save_file_folder=save_file_folder,
             save_timestamped_file=config.get(CONF_SAVE_TIMESTAMPTED_FILE),
@@ -234,7 +240,7 @@ def setup_platform(hass, config, add_devices, discovery_info=None):
 
 
 class ObjectClassifyEntity(ImageProcessingEntity):
-    """Perform a face classification."""
+    """Perform a object classification."""
 
     def __init__(
         self,
@@ -249,6 +255,7 @@ def __init__(
         roi_x_min,
         roi_y_max,
         roi_x_max,
+        scale,
         show_boxes,
         save_file_folder,
         save_timestamped_file,
@@ -291,19 +298,34 @@ def __init__(
             "y_max": roi_y_max,
             "x_max": roi_x_max,
         }
-
+        self._scale = scale
         self._show_boxes = show_boxes
         self._last_detection = None
         self._image_width = None
         self._image_height = None
         self._save_file_folder = save_file_folder
         self._save_timestamped_file = save_timestamped_file
+        self._image = None
 
     def process_image(self, image):
         """Process an image."""
-        self._image_width, self._image_height = Image.open(
-            io.BytesIO(bytearray(image))
-        ).size
+        self._image = Image.open(io.BytesIO(bytearray(image)))
+        self._image_width, self._image_height = self._image.size
+
+        # resize image if different then default
+        if self._scale != DEAULT_SCALE:
+            newsize = (self._image_width * self._scale, self._image_width * self._scale)
+            self._image.thumbnail(newsize, Image.ANTIALIAS)
+            self._image_width, self._image_height = self._image.size
+            with io.BytesIO() as output:
+                self._image.save(output, format="JPEG")
+                image = output.getvalue()
+            _LOGGER.debug(
+                (
+                    f"Image scaled with : {self._scale} W={self._image_width} H={self._image_height}"
+                )
+            )
+
         self._state = None
         self._objects = []  # The parsed raw data
         self._targets_found = []
@@ -344,7 +366,8 @@ def process_image(self, image):
 
         if self._save_file_folder and self._state > 0:
             saved_image_path = self.save_image(
-                image, self._targets_found, self._save_file_folder,
+                self._targets_found,
+                self._save_file_folder,
             )
 
         # Fire events
@@ -396,13 +419,13 @@ def device_state_attributes(self) -> Dict:
             attr[CONF_SAVE_TIMESTAMPTED_FILE] = self._save_timestamped_file
         return attr
 
-    def save_image(self, image, targets, directory) -> str:
+    def save_image(self, targets, directory) -> str:
         """Draws the actual bounding box of the detected objects.
 
         Returns: saved_image_path, which is the path to the saved timestamped file if configured, else the default saved image.
         """
         try:
-            img = Image.open(io.BytesIO(bytearray(image))).convert("RGB")
+            img = self._image.convert("RGB")
         except UnidentifiedImageError:
             _LOGGER.warning("Deepstack unable to process image, bad data")
             return
@@ -411,7 +434,12 @@ def save_image(self, image, targets, directory) -> str:
         roi_tuple = tuple(self._roi_dict.values())
         if roi_tuple != DEFAULT_ROI and self._show_boxes:
             draw_box(
-                draw, roi_tuple, img.width, img.height, text="ROI", color=GREEN,
+                draw,
+                roi_tuple,
+                img.width,
+                img.height,
+                text="ROI",
+                color=GREEN,
             )
 
         for obj in targets: