The number of cells in 'structure' file and 'chunk' file does not match.

[xuewenyuan]

For many annotations, I find the number of cells in 'structure' file and 'chunk' file does not match. The cell ids also do not matched each other.
E.g.,

File	# cells
train/chunk/0907.1815v1.2.chunk	221
train/structure/0907.1815v1.2.json	252

And the maximum cell id in 'rel' file is 218.

[maheshp1]

That's because they have added empty cells in the json file whereas in the chunk files only cells with content are present. I am facing the same problem too, while extracting the structure of the cells from the chunk and json files simultaneously. The json files contain the cell-id but the chunk doesn't. So, if anyone has any solution to match both the files, that would be good.

[xuewenyuan]

That's because they have added empty cells in the json file whereas in the chunk files only cells with content are present. I am facing the same problem too, while extracting the structure of the cells from the chunk and json files simultaneously. The json files contain the cell-id but the chunk doesn't. So, if anyone has any solution to match both the files, that would be good.

Thank you for your comment. It seems that most works only use this dataset as training set for cell detection while testing on the ICDAR13-Table dataset.