RTMDet-ins annotation.py ValueError: full_shape must be provided

[eppane]

Hello,

encountered following error with the latest release (0.11.18), perhaps has something to do with the new mask processing improvements:

---------------------------------------------------------------------------
File ~/miniconda3/envs/xxx/lib/python3.11/site-packages/sahi/predict.py:100, in get_prediction(image, detection_model, shift_amount, full_shape, postprocess, verbose)
     98 time_start = time.time()
     99 # works only with 1 batch
--> 100 detection_model.convert_original_predictions(
    101     shift_amount=shift_amount,
    102     full_shape=full_shape,
    103 )
    104 object_prediction_list: List[ObjectPrediction] = detection_model.object_prediction_list
    106 # postprocess matching predictions

File ~/miniconda3/envs/xxx/lib/python3.11/site-packages/sahi/models/base.py:168, in DetectionModel.convert_original_predictions(self, shift_amount, full_shape)
    154 def convert_original_predictions(
    155     self,
    156     shift_amount: Optional[List[int]] = [0, 0],
    157     full_shape: Optional[List[int]] = None,
    158 ):
    159     """
    160     Converts original predictions of the detection model to a list of
    161     prediction.ObjectPrediction object. Should be called after perform_inference().
   (...)
    166             Size of the full image after shifting, should be in the form of [height, width]
    167     """
--> 168     self._create_object_prediction_list_from_original_predictions(
    169         shift_amount_list=shift_amount,
    170         full_shape_list=full_shape,
    171     )
    172     if self.category_remapping:
    173         self._apply_category_remapping()

File ~/miniconda3/envs/xxx/lib/python3.11/site-packages/sahi/models/mmdet.py:302, in MmdetDetectionModel._create_object_prediction_list_from_original_predictions(self, shift_amount_list, full_shape_list)
    299         logger.warning(f"ignoring invalid prediction with bbox: {bbox}")
    300         continue
--> 302     object_prediction = ObjectPrediction(
    303         bbox=bbox,
    304         category_id=category_id,
    305         score=score,
    306         segmentation=segmentation,
    307         category_name=category_name,
    308         shift_amount=shift_amount,
    309         full_shape=full_shape,
    310     )
    311     object_prediction_list.append(object_prediction)
    312 object_prediction_list_per_image.append(object_prediction_list)

File ~/miniconda3/envs/xxx/lib/python3.11/site-packages/sahi/prediction.py:79, in ObjectPrediction.__init__(self, bbox, category_id, category_name, segmentation, score, shift_amount, full_shape)
     53 """
     54 Creates ObjectPrediction from bbox, score, category_id, category_name, segmentation.
     55 
   (...)
     76         the form of [height, width]
     77 """
     78 self.score = PredictionScore(score)
---> 79 super().__init__(
     80     bbox=bbox,
     81     category_id=category_id,
     82     segmentation=segmentation,
     83     category_name=category_name,
     84     shift_amount=shift_amount,
     85     full_shape=full_shape,
     86 )

File ~/miniconda3/envs/xxx/lib/python3.11/site-packages/sahi/annotation.py:516, in ObjectAnnotation.__init__(self, bbox, segmentation, category_id, category_name, shift_amount, full_shape)
    514     raise ValueError("you must provide a bbox or segmentation")
    515 if segmentation is not None:
--> 516     self.mask = Mask(
    517         segmentation=segmentation,
    518         shift_amount=shift_amount,
    519         full_shape=full_shape,
    520     )
    521     bbox_from_segmentation = get_bbox_from_coco_segmentation(segmentation)
    522     # https://github.com/obss/sahi/issues/235

File ~/miniconda3/envs/xxx/lib/python3.11/site-packages/sahi/annotation.py:181, in Mask.__init__(self, segmentation, full_shape, shift_amount)
    179 # confirm full_shape is given
    180 if full_shape is None:
--> 181     raise ValueError("full_shape must be provided")
    183 self.shift_x = shift_amount[0]
    184 self.shift_y = shift_amount[1]

ValueError: full_shape must be provided

Any suggestions? 🙏

[Sultan91]

i am getting this error only recently, probably version problem?

[fcakyon]

@eppane @Sultan91 can you please try with the latest release? this bug should be fixed: https://github.com/obss/sahi/releases/tag/0.11.19

[eppane]

@fcakyon it seems to be working! While the error is quite self-explanatory, it is not entirely clear where the full_shape needs to be provided. The notebook for mmdetection inference does not use it for getting masks, but e.g. notebook for yolov8 does at the segmentation part. I think the mmdet notebook is outdated. When predicting with e.g. rtmdet-ins model, one can use something like:

res = get_prediction(input_img, model, full_shape=input_img.shape[:2])