search

jiwoon-ahn / irn

Weakly Supervised Learning of Instance Segmentation with Inter-pixel Relations, CVPR 2019 (Oral)
MIT License
519 stars 100 forks source link

Training on own dataset #20

Open discretecoder opened 4 years ago

discretecoder commented 4 years ago

thanks for the great work, I wish to train the network on Berkeley deep drive dataset where we have 2d bounding boxes in json files. what would be the steps? is there a data converter available? I am trying to get the dataset in pascal voc 2012 format

thanks

SuzannaLin commented 3 years ago

I am trying to do the same. Have you found a way to adjust the code?

SuzannaLin commented 3 years ago

What I did was convert the ground truth labels to .xml format, maintaining only the unique labels (not even including bounding box locations). Then, in the create cls_labels.py script, I changed the labels, but maintained the amount of 20 (just having empty labels). That worked for me.