Method overlay

walrus/methods/action/workflowexecutor.py:439–543

(self)

    # Go to next

    def overlay(self):

        self.temp = tempfile.mkdtemp()

        inference = self.previous_action_event.brain_inference

        # Assumes label_dict as label_file_id

        label_dict = inference.ai.ml.label_dict

        # Defaults to storing our db id : model sequential number ie 121 : 1
        # Use this to convert model output to get actual labels
        inverted_label_file_dict = {v: k for k, v in label_dict.items()}

        # Class image() object
        image = inference.file.image

        if not inference.boxes['boxes']:
            return

        ## Now get the image and draw the overlay

        # Raw original image
        data_tools = Data_tools().data_tools

        image_string = data_tools.download_bytes(image.url_signed_blob_path)

        image_np = imread(BytesIO(image_string))

        # Convert to pill setup for drawing operations
        pil_image = PIL_Image.fromarray(image_np)

        # Get overlay image
        if self.action.overlay_kind == "image":
            overlay_string = data_tools.download_bytes(image.url_signed_blob_path)

            # Convert to numpy array
            overlay_blob_np = imread(BytesIO(overlay_string))

            pil_overlay_image = PIL_Image.fromarray(overlay_blob_np)

        # TODO optional "normal" box drawing

        # Caution, this expects boxes to be in a list
        # ie [ [0, 0, 0 ,0] ] even if only 1 box!!!!

        for i, box in enumerate(inference.boxes['boxes']):

            label = inference.classes['classes'][i]
            score = inference.scores['scores'][i]

            label_file_id = inverted_label_file_dict[label]

            # Pass on files that aren't equal to chosen label
            if int(label_file_id) != (self.action.overlay_label_file_id):
                continue

            # Denormalize cordinates