search

thsant / wgisd

Embrapa Wine Grape Instance Segmentation Dataset - Embrapa WGISD
Other
123 stars 39 forks source link

grape bunch grouping #8

Closed andreaceruti closed 2 years ago

andreaceruti commented 3 years ago

Looking at the dataset with your notebook WGISD.ipynb I have noted that sometimes the same grape bunch mask is splitted in 2 different parts. For example I have inserted in train_masked.txt the image "CDY_2015" in order to visualize it and the result is the one in the screenshot you see. You can see on the left that instead of one single mask we have 2 masks (purple and green adjacent masks) and also the grape bunch below is separated in 2 parts.

So I was wondering if this is a design choice since the image was used for testing and not for training or if It regards something that I am not aware of.

thsant commented 2 years ago

Dear @andreaceruti, if you look for the peduncles, you will see the green and purple bunches came from different peduncles, so they are different brunches. However, annotation is error-prone and WGISD is not free of errors, most of them caused by single view of complex 3-D objects with overlap. But this is not the case in the example above: a close look will reveal that the green and the purple masks are really from different bunches.