AttributeError: 'Namespace' object has no attribute 'train_datasets'

ChenRocks / UNITER

Research code for ECCV 2020 paper "UNITER: UNiversal Image-TExt Representation Learning"

https://arxiv.org/abs/1909.11740

777 stars 109 forks source link

AttributeError: 'Namespace' object has no attribute 'train_datasets' #92

Open ForawardStar opened 2 years ago

ForawardStar commented 2 years ago

When I launched the 2nd stage training using :

horovodrun -np 4 python pretrain_vcr.py --config config/pretrain-vcr-base-4gpu.json \
    --output_dir $PRETRAIN_VCR_EXP

An error occured, which is:

Traceback (most recent call last):

File "pretrain_vcr.py", line 560, in <module>

main(args)

File "pretrain_vcr.py", line 205, in main

all_dbs = [db for datasets in [opts.train_datasets, opts.val_datasets]

AttributeError: 'Namespace' object has no attribute 'train_datasets'

How to fix this problem? Thanks.

linjieli222 commented 2 years ago

train_datasets is indeed in opts (which is loaded from config)

https://github.com/ChenRocks/UNITER/blob/1dbfa62bb8ddb1f2b11d20775ce324c5e45a3ab4/config/pretrain-vcr-base-4gpu.json#L30

ForawardStar commented 2 years ago

Thanks for your reply! It is helpful to me. Here I want to launch the first stage pre-training on COCO, Visual Genome, CC, and SBU. Can your provide the linking for downloading CC, and SBU? If their sizes are too large, can you simply provide the .db file of text input, or the scripts for producing such .db file?