requiremets.txt might help with potential user issues

[anhelina007]

Hi! I just started using you amazing tool for a university robotics project.

We as a team have already spend numerous days on trying to figure out the error the constantly get with any kind of json we are using

    return _mask.decode(rleObjs)
  File "pycocotools/_mask.pyx", line 145, in pycocotools._mask.decode
  File "pycocotools/_mask.pyx", line 127, in pycocotools._mask._frString
TypeError: list indices must be integers or slices, not str

which happens in the following line of code:

masks = [maskUtils.decode(ann['segmentation']) for ann in dataset_dict['annotations']].

We have also tried CanaTree100 dataset that you provided.

I would really appreciate your help since we want to upgrade your work and train it on jungle data.

[pie-leroy]

Hello,

I had a similar error some time ago, installing detectron2 leads to many errors that are related to version incompatibilities between several libraries.

To save time I created a notebook on google collab that allows to run everything (check the use of gpu in google collab).

If you have any questions, don't hesitate!

https://colab.research.google.com/drive/1lRF7GhxwFwVgJBtBLy5TYLXD_vnRDHDv?usp=sharing

[anhelina007]

Thank you a lot! Could you please make it public?

[pie-leroy]

is is good now ?
https://colab.research.google.com/drive/1lRF7GhxwFwVgJBtBLy5TYLXD_vnRDHDv?usp=sharing

[VGrondin]

Hi! I just started using you amazing tool for a university robotics project.

We as a team have already spend numerous days on trying to figure out the error the constantly get with any kind of json we are using

    return _mask.decode(rleObjs)
  File "pycocotools/_mask.pyx", line 145, in pycocotools._mask.decode
  File "pycocotools/_mask.pyx", line 127, in pycocotools._mask._frString
TypeError: list indices must be integers or slices, not str

which happens in the following line of code:

masks = [maskUtils.decode(ann['segmentation']) for ann in dataset_dict['annotations']].

We have also tried CanaTree100 dataset that you provided.

I would really appreciate your help since we want to upgrade your work and train it on jungle data.

Hi, I remember getting this error a few times when switching between the synthetic/real dataset. IIRC this issue is caused by the segmentation mask encoding method. Depending on which encoding (RLE or bitmask) was used for the segmentation mask annotations, this error arises. In your case, it would seem that your annotations are already in RLE format since the
return _mask.decode(rleObjs) is an error.
masks = [maskUtils.decode(ann['segmentation']) for ann in dataset_dict['annotations']]
This is because the maskUtils.decode function decodes a binary mask into a RLE format (https://github.com/cocodataset/cocoapi/blob/master/PythonAPI/pycocotools/mask.py), but your input is already in RLE format. You might be able to simply replace it with
masks = [ann['segmentation'] for ann in dataset_dict['annotations']]

Hope it helps!