These weights were obtained for 4-fold cross validation using TACO-10 class map.
Each file with the name: <dataset>_<map_number_of_target_classes>_<fold number>.zip contains the following:

• The actual weights: mask_rcnn_taco_<epoch number>.h5. Copy this to detector/models/logs
• The respective dataset splits: annotations_0_<subset>.json. Copy these 3 files to your dataset directory.

Setup used:
tensorflow 1.9.0
keras 2.16

Note: Later versions can be incompatible. Check #17