Could you please write a demo for us?

Hey,

Thank you for this work (and MaskTrackRCNN). I have generated masks as (height, width) PyTorch-BoolTensors and would like to evaluate them using the CodaLab evaluation page. Each segmentation seems to have the structure {'size': (height, width), 'counts': rle} where rle is an encoding of the mask. How do I convert my masks? I have tried:

import third_party.cocoapi.PythonAPI.pycocotools.mask as cocomask

# mask is (height, width) torch.BoolTensor
mask = np.array(mask.cpu().numpy()[:, :, np.newaxis], order='F')
rle = cocomask.encode(mask)[0]
rle['counts'] = rle['counts'].decode()
return rle

Best regards,
Joakim

Hi, thanks to your work, but I encountered some problems.
I download version2019 video instance segmentation dataset from this link:https://competitions.codalab.org/competitions/20128#participate-get_data.
But the labels of json style are inconsistent with the format given on this web page.
For example, I open the valid.json with vscode and search ' "segmenation" ' keyword while there are no any results.
Then I tried to load data with PythonAPI, I modified the path parameter and run pycocoDemo.ipynb to load valid.json which was downloaded from https://drive.google.com/drive/folders/1VBeVXSf-HfrhyBurtu1xfGn6zQ-ALpjZ?usp=sharing, but I got this error:

print(len(coco.dataset['categories']))
print(coco.dataset['categories'][0]['id'])
print(len(coco.imgs))
catIds = coco.getCatIds(catNms=['person','dog','skateboard']);
print(type(catIds[0]))
imgIds = coco.getImgIds(catIds=catIds );
print(imgIds)
imgIds = coco.getImgIds(imgIds = [324158])
img = coco.loadImgs(imgIds[np.random.randint(0,len(imgIds))])[0]

40
1
0
<class 'int'>
[]
---------------------------------------------------------------------------
KeyError                                  Traceback (most recent call last)
<ipython-input-25-021d63060e40> in <module>
      8 print(imgIds)
      9 imgIds = coco.getImgIds(imgIds = [324158])
---> 10 img = coco.loadImgs(imgIds[np.random.randint(0,len(imgIds))])[0]

/data/EventSegmentation/MaskTrackRCNN/cocoapi/PythonAPI/pycocotools/coco.py in loadImgs(self, ids)
    229             return [self.imgs[id] for id in ids]
    230         elif type(ids) == int:
--> 231             return [self.imgs[ids]]
    232 
    233     def showAnns(self, anns):

KeyError: 324158

Is there anything wrong with my operation? or the format is really different with the style on that web page?

Hi,

Thank you for your work! I was trying to submit my results to the CodaLab server but came across the following error. Could you guide me as to why this may be occurring?

Traceback (most recent call last):
  File "/worker/worker.py", line 330, in run
    if input_rel_path not in bundles:
TypeError: argument of type 'NoneType' is not iterable

As a sanity check, I also tried uploading the valid_submission_sample.zip to the server for evaluation, but I got the same error in that case as well. Is something wrong with the server evaluation?

I generated the result.json file but don't know how to use the evaluation toolkit

Could you please write a demo for us?

Hi,

I was using loadRes in ytvos.py at Line 216, and I came across the following error:
NameError: name 'unicode' is not defined

I believe it has something to with Python3 replacing unicode with str.

Is pycocotools not built for Python3? Could this be changed?