[BUG] train2017 is not found in config
AnhMinhTran opened this issue · 2 comments
Describe the bug
Train2017 is not found in config using COCO dataset when loading dataset_train, dataset_eval for training
To Reproduce
Steps to reproduce the behavior:
- Download COCO dataset
- Extract val, train, and annotation folder
Expected behavior
Script start training COCO dataset
Screenshots
If applicable, add screenshots to help explain your problem.
File structure
Error
Desktop (please complete the following information):
- OS: Ubuntu 18.04
- This repository version: Download from GIT
- PyTorch version w/ CUDA/cuDNN: 1.7.1
Additional context
Thank you so much for looking at the problem
The folders are named as you have them, but the split in the code is called train not train2017 as per
- https://github.com/rwightman/efficientdet-pytorch/blob/master/effdet/data/dataset_config.py#L21 ...
- https://github.com/rwightman/efficientdet-pytorch/blob/master/effdet/data/dataset_factory.py#L15
... so I'm not sure how/why it's looking for a split called train2017 , did you modify code or pass diff split args? the 2017 is passed via the dataset arg, the split names are consistent across dataset types
I found the issue, it was one of my args pass, somehow it found it way to the cofig. Closing the issue now