Maskrcnn-benchmark: Mask_R-CNN_demo assert error

[antran89]

🐛 Bug

Error when run the webcam or notebook demo

AssertionError                            Traceback (most recent call last)
<ipython-input-16-ba06eae97bcc> in <module>()
      1 # compute predictions
----> 2 predictions = coco_demo.run_on_opencv_image(image)
      3 imshow(predictions)

/Users/an.tran/Desktop/map-workspace/code/maskrcnn-benchmark/demo/predictor.pyc in run_on_opencv_image(self, image)
    167                 the BoxList via `prediction.fields()`
    168         """
--> 169         predictions = self.compute_prediction(image)
    170         top_predictions = self.select_top_predictions(predictions)
    171 

/Users/an.tran/Desktop/map-workspace/code/maskrcnn-benchmark/demo/predictor.pyc in compute_prediction(self, original_image)
    212             # in the image, as defined by the bounding boxes
    213             masks = prediction.get_field("mask")
--> 214             masks = self.masker(masks, prediction)
    215             prediction.add_field("mask", masks)
    216         return prediction

/Users/an.tran/Desktop/map-workspace/code/maskrcnn-benchmark/maskrcnn_benchmark/modeling/roi_heads/mask_head/inference.pyc in __call__(self, masks, boxes)
    183 
    184         # Make some sanity check
--> 185         assert len(boxes) == len(masks), "Masks and boxes should have the same length."
    186 
    187         # TODO:  Is this JIT compatible?

AssertionError: Masks and boxes should have the same length.

To Reproduce

Steps to reproduce the behavior:

1. On Mac
2. After installation, run the Python notebook Mask_R-CNN_demo.ipynb
   1.

Expected behavior

Environment

Please copy and paste the output from the
environment collection script from PyTorch
(or fill out the checklist below manually).

You can get the script and run it with:

wget https://raw.githubusercontent.com/pytorch/pytorch/master/torch/utils/collect_env.py
# For security purposes, please check the contents of collect_env.py before running it.
python collect_env.py

Collecting environment information...
PyTorch version: 1.0.0.dev20181119
Is debug build: No
CUDA used to build PyTorch: None

OS: Mac OSX 10.13.6
GCC version: Could not collect
CMake version: Could not collect

Python version: 2.7
Is CUDA available: No
CUDA runtime version: No CUDA
GPU models and configuration: No CUDA
Nvidia driver version: No CUDA
cuDNN version: No CUDA

Versions of relevant libraries:
[pip] 18.0
[conda] Could not collect

Additional context

[keineahnung2345]

Same problems here!

[jinfagang]

@keineahnung2345 Seems the most decent master branch has the logic error.

But I cloned last night work fine...................... No idea what changed here. But I can also reproduce this logic error.
@fmassa

[jinfagang]

@keineahnung2345 Currently, you can change it back to this logic:

class Masker(object):
    """
    Projects a set of masks in an image on the locations
    specified by the bounding boxes
    """

    def __init__(self, threshold=0.5, padding=1):
        self.threshold = threshold
        self.padding = padding

    def forward_single_image(self, masks, boxes):
        boxes = boxes.convert("xyxy")
        im_w, im_h = boxes.size
        res = [
            paste_mask_in_image(mask[0], box, im_h, im_w, self.threshold, self.padding)
            for mask, box in zip(masks, boxes.bbox)
        ]
        if len(res) > 0:
            res = torch.stack(res, dim=0)[:, None]
        else:
            res = masks.new_empty((0, 1, masks.shape[-2], masks.shape[-1]))
        return res

    def __call__(self, masks, boxes):
        # TODO do this properly
        if isinstance(boxes, BoxList):
            boxes = [boxes]
        assert len(boxes) == 1, "Only single image batch supported"
        result = self.forward_single_image(masks, boxes[0])
        return result

[fmassa]

I merged a PR yesterday that is the culprit. I'll fix this in the next 2h. Sorry for the trouble!

[antran89]

@fmassa Hope to hear your responses soon! :-) Just try to dig into problems by myself. Could not find a clue.

[keineahnung2345]

@antran89 @fmassa @jinfagang Hello, I tried to fix this problem and open a PR in my own repo: https://github.com/keineahnung2345/maskrcnn-benchmark/pull/1/files.
But this will make my kernel restart. I guess it's because results = torch.stack(results, dim=0) takes too much space.
I hope this can help someone who is also trying to fix this issue.

[fmassa]

I've fixed this issue in #187
Let me know if you still face the same issue, and thanks for reporting it!

[antran89]

@fmassa it solved!

[mattans]

I encountered it when I cloned and installed https://github.com/Maosef/maskrcnn-benchmark today, as I need the fix of this PR: https://github.com/facebookresearch/maskrcnn-benchmark/pull/271
Supposely @Maosef 's master is based on a recent enough master from maskrcnn-benchmark so this shouldn't have happened.
Anyway it is fixed by manually editing predictor.py according to this: https://github.com/facebookresearch/maskrcnn-benchmark/pull/187/commits/7cf1d982527d8fb5ded4bd1b4eae77517bb122ef

[fmassa]

@mattans you can see that Maosef branch is 22 commits behind master, and it doesn't have the fix in https://github.com/facebookresearch/maskrcnn-benchmark/commit/7cf1d982527d8fb5ded4bd1b4eae77517bb122ef