Open mohitchiku opened 3 years ago
you can refer to the issues in matterport Maskrcnn repo and can add custom callback https://github.com/matterport/Mask_RCNN/pull/1024#issuecomment-468019810
That would only calculate these metrics for the cell-detection part of the network. To calculate row/col adjacency F1 score, the model is passing tab_results
to coco evaluation which doesn't calculate row/col adjacency from what I know.
I might be missing something but how is these metrics calculated for the Table Structure Recognition tasks to fill up Table 4 in the paper?
how to get mAP, mAR, F1_score using utils.compute_ap(..) https://github.com/sachinraja13/TabStructNet/blob/master/mrcnn/utils.py#L717