Maskrcnn-benchmark: Maybe this is some problem associated with pycocotools

Created on 27 Dec 2018 · 9Comments · Source: facebookresearch/maskrcnn-benchmark

🐛 Bug

I am not sure if this is some problem of my own configurations, but when I run an instance segmentation problem, I received an error message.

Running per image evaluation...
Evaluate annotation type *segm*
Traceback (most recent call last):
  File "tools/train_net.py", line 171, in <module>
    main()
  File "tools/train_net.py", line 167, in main
    test(cfg, model, args.distributed)
  File "tools/train_net.py", line 104, in test
    output_folder=output_folder,
  File "/data/zhangzy/projects/play/maskrcnn-benchmark/maskrcnn_benchmark/engine/inference.py"
, line 104, in inference
    **extra_args)
  File "/data/zhangzy/projects/play/maskrcnn-benchmark/maskrcnn_benchmark/data/datasets/evalua
tion/__init__.py", line 22, in evaluate
    return coco_evaluation(**args)
  File "/data/zhangzy/projects/play/maskrcnn-benchmark/maskrcnn_benchmark/data/datasets/evalua
tion/coco/__init__.py", line 20, in coco_evaluation
    expected_results_sigma_tol=expected_results_sigma_tol,
  File "/data/zhangzy/projects/play/maskrcnn-benchmark/maskrcnn_benchmark/data/datasets/evalua
tion/coco/coco_eval.py", line 57, in do_coco_evaluation
    dataset.coco, coco_results[iou_type], file_path, iou_type
  File "/data/zhangzy/projects/play/maskrcnn-benchmark/maskrcnn_benchmark/data/datasets/evalua
tion/coco/coco_eval.py", line 287, in evaluate_predictions_on_coco
    coco_eval.evaluate()
  File "/home/zhangzy/.local/lib/python3.5/site-packages/pycocotools-2.0-py3.5-linux-x86_64.eg
g/pycocotools/cocoeval.py", line 141, in evaluate
    self._prepare()
  File "/home/zhangzy/.local/lib/python3.5/site-packages/pycocotools-2.0-py3.5-linux-x86_64.eg
g/pycocotools/cocoeval.py", line 105, in _prepare
    _toMask(gts, self.cocoGt)
  File "/home/zhangzy/.local/lib/python3.5/site-packages/pycocotools-2.0-py3.5-linux-x86_64.eg
g/pycocotools/cocoeval.py", line 93, in _toMask
    rle = coco.annToRLE(ann)
  File "/home/zhangzy/.local/lib/python3.5/site-packages/pycocotools-2.0-py3.5-linux-x86_64.eg
g/pycocotools/coco.py", line 416, in annToRLE
    rles = maskUtils.frPyObjects(segm, h, w)
  File "pycocotools/_mask.pyx", line 293, in pycocotools._mask.frPyObjects
TypeError: Argument 'bb' has incorrect type (expected numpy.ndarray, got list)

To Reproduce

Just make a small dataset with the format of coco dataset, and train on it.

Environment

Collecting environment information...
PyTorch version: 1.0.0a0+37627a1
Is debug build: No
CUDA used to build PyTorch: 9.0.176

OS: Ubuntu 16.04.4 LTS
GCC version: (Ubuntu 5.4.0-6ubuntu1~16.04.10) 5.4.0 20160609
CMake version: version 3.12.0

Python version: 3.5
Is CUDA available: Yes
CUDA runtime version: 9.0.176
GPU models and configuration: 
GPU 0: GeForce GTX 1080 Ti
GPU 1: GeForce GTX 1080 Ti
GPU 2: GeForce GTX 1080 Ti
GPU 3: GeForce GTX 1080 Ti
GPU 4: GeForce GTX 1080 Ti
GPU 5: GeForce GTX 1080 Ti
GPU 6: GeForce GTX 1080 Ti
GPU 7: GeForce GTX 1080 Ti

Nvidia driver version: 384.69
cuDNN version: Probably one of the following:
/usr/local/cuda-9.0/lib64/libcudnn.so.7.0.5
/usr/local/cuda-9.0/lib64/libcudnn_static.a

Versions of relevant libraries:
[pip3] numpy (1.14.5)
[pip3] torch (1.0.0a0+37627a1)
[pip3] torchvision (0.2.1)
[conda] Could not collect

Source

CoinCheung

Most helpful comment

Sorry, this trouble is brought by our annotation coworkers. There is a annotation that the segmentation field is [793, 718, 792, 716] and the box field is [792, 716, 1, 2]. This segmentation misleads the program to execute this line, since the segmentation has only two points. I will check our data more carefully.

CoinCheung on 29 Dec 2018

👍2

All 9 comments

It looks like your model is not outputting the mask predictions in the right format. Did you add the Masker in your code? Also, what else have you modified?

fmassa on 27 Dec 2018

I got the same error message o(╥﹏╥)o

JavisPeng on 28 Dec 2018

I did not change the primary code. I cloned a new set of code from the repository and then add my own dataset paths to maskrcnn_benchmark/config/paths_catalog.py. After training and evaluating and before it printing the results to the screen, the error appears.

CoinCheung on 28 Dec 2018

https://github.com/cocodataset/cocoapi/issues/139 says that the question caused by "This happens if mask length for an annotation = 4", however, I filtered out this situation in my data process code, this error still exists. o(╥﹏╥)o

JavisPeng on 28 Dec 2018

So your dataset is in the COCO format, right?
My guess would be that one of the annotations has a problem in the evaluation dataset maybe?

fmassa on 28 Dec 2018

what is the problem likely to be please? I created my dataset into coco format and it works well when I used it to train the model.

CoinCheung on 29 Dec 2018