VisibleDeprecationWarning when doing transfer learning

[amaanda]

hi! first of all, thanks for sharing your work.

i'm doing transfer learning with the resnet50 model provided for yolact++ on an ec2 instance. I followed the steps suggested in https://github.com/dbolya/yolact/issues/36 but I keep getting this deprecation warning every 100 iterations, just wanted to check if it's something i should be concerned with.

/home/ubuntu/odd/yolact/utils/augmentations.py:309: VisibleDeprecationWarning: Creating an ndarray from ragged nested sequences (which is a list-or-tuple of lists-or-tuples-or ndarrays with different lengths or shapes) is deprecated. If you meant to do this, you must specify 'dtype=object' when creating the ndarray mode = random.choice(self.sample_options)

i am doing transfer learning to only 1 class, this is the command i'm running:

python3 train.py --config=yolact_plus_resnet50_my_dataset_config --resume=weight/yolact_plus_resnet50_54_800000.pth --start_iter=0 --save_interval=10000 --log_folder=/home/ubuntu/odd/yolact_s3/yolact/log --batch_size=16 --save_folder=/home/ubuntu/odd/yolact_s3/yolact/weight/

thanks in advance!

[davodogster]

@amaanda I'm also getting it. Training YOLACT on local GPU. I'm Using the Resnet101 backbone. Not loading an old model

[oulfik]

You can disable the warnings by adding these lines to augmentations.py:

import warnings
warnings.filterwarnings('ignore')