Custom_Train_MaskRCNN

Training on custom dataset with (multi/unique class) of a Mask RCNN

Requirements (no specific version requirements)

  python3
  pycocotools
  matplotlib
  mrcnn
  tqdm
  numpy
  pylab
  skimage

Note: installation for mrcnn will be explained in the medium article linked in the repo.

Structure

• dataset: folder where you put the train and val folders (read inside to know what to put)
• logs: folder where we store the intermediate/checkpoints and final weights after training
• weights: weights for the model, we fetch the weights from here for the test script
• detect_segment_test.py: test script for the segmentation, displays mask on top of input image, usage given by --h argument
• train.py: main script for this section, read medium article to know what to modify

Usage

First training usage, more options showed in the train.py script as comment:

   python3 train.py train --dataset=./dataset --weights=coco