Some questions about my model

[Linaom1214]

In my model, there is always a fake target in the upper left of the image, and I tested this for all sequences

Here is my code

class IRDetectionModel(DetectionModel):
    def load_model(self):
        self.model = BaseEngine(self.model_path)
        
    def perform_inference(self, image: np.ndarray, image_size: int = 256):
        image = cv2.resize(image, (image_size, image_size))
        image = np.transpose(image,(2,0,1))
        image = image / 255.
        image = np.array(image,order='C',dtype=np.float32)
        image = np.ravel(image)
        prediction_result = self.model.infer(image)
        prediction_result = np.reshape(prediction_result,(-1,5))
        prediction_result = np.expand_dims(prediction_result, 0)
        self._original_predictions = prediction_result
    
    def _create_object_prediction_list_from_original_predictions(self, 
        shift_amount_list: Optional[List[int]] = [[0, 0]], 
        full_shape_list: Optional[List[int]] = None):

        original_predictions =  self._original_predictions
        shift_amount_list = fix_shift_amount_list(shift_amount_list)
        full_shape_list = fix_full_shape_list(full_shape_list)
        # handle all predictions
        object_prediction_list_per_image = []
        for image_ind, image_predictions in enumerate(original_predictions):
            shift_amount = shift_amount_list[image_ind]
            full_shape = None if full_shape_list is None else full_shape_list[image_ind]

            object_prediction_list =[]
            for predictions in image_predictions:
                x1 = int(predictions[0]*256)
                y1 = int(predictions[1]*256)
                x2 = int(predictions[2]*256)
                y2 = int(predictions[3]*256)
                
                bbox = [x1, y1, x2, y2]
                if bbox[0] > bbox[2] or bbox[1] > bbox[3] or bbox[0] < 0 or bbox[1] < 0 or bbox[2] <0 or bbox[3] < 0:
                    continue
                if full_shape is not None and (
                    bbox[1] > full_shape[0] or
                    bbox[3] > full_shape[0] or 
                    bbox[0] > full_shape[1] or 
                    bbox[2] > full_shape[1]):
                    continue
                score = predictions[4]
                if score < 0.8:
                    continue

                if (x2 - x1) > 10 or (y2 - y1) > 10:
                    continue
                category_id = 0

                category_name = 'target'

                object_prediction = ObjectPrediction(
                 bbox = bbox,
                 score = score, 
                 category_id = category_id, 
                 category_name = category_name,
                 shift_amount = shift_amount,
                 full_shape = full_shape,
                 bool_mask=None)
                object_prediction_list.append(object_prediction)
            object_prediction_list_per_image.append(object_prediction_list)
        self._object_prediction_list_per_image = object_prediction_list_per_image

        
            
def predict(img, model, sw, sh, ohr, owwr, img_size):
    result = get_sliced_prediction(
        img, model, slice_width=sw, 
        slice_height=sh, 
        overlap_height_ratio=ohr,
        overlap_width_ratio=owwr,
        image_size = img_size,
        verbose= 0,
        postprocess_match_threshold= 0.5,
        perform_standard_pred= True
        )
    bboxes = []
    scores = []
    result_len = result.to_coco_annotations()

    for pred in result_len:
        bboxes.append(pred['bbox'])
        scores.append(pred['score'])
    return bboxes, scores


def plot(image, bboxes, scores):
    for box, score in zip(bboxes, scores):
        if score > 0.9:
            x1 = int(box[0])
            y1 = int(box[1])
            x2 = int(box[2])
            y2 = int(box[3])
            cv2.rectangle(image, (x1, y1), (x1+x2, y1+ y2), (255,0,0), 2)

is some wrong ?

[fcakyon]

@Linaom1214 have you tried calling get_sliced_prediction with perform_standard_pred=False? Does the issue persist that way?

[Linaom1214]

thanks very much . Isolve my issue