RefCOCO/+/g json files

[nankepan]

Thanks for your great work!
Can you provide the RefCOCO/+/g json files? I din't find them in this repository.

[wjn922]

Hi, you can refer to this file to generate the json files by yourself.

We may release the generated json files in a few days since we are busy on other project right now. Thanks for your understanding.

[nankepan]

Thank you for your response!

[wjn922]

@nankepan We have uploaded the pre-processed refcoco/+/g files in OneDrive.

[itruonghai]

@wjn922 Could you release the joint training script?

[wjn922]

@itruonghai The script is like this, but note that we use 32 V100 GPUs for the joint-training:

python3 -m torch.distributed.launch --nproc_per_node=8  --use_env \
main_joint.py  --dataset_file joint --binary --with_box_refine \
--batch_size 1 --num_frames 5 \
--epochs 12 --lr_drop 8 10 \
--freeze_text_encoder \
[backbone]

[itruonghai]

@wjn922 Could you also release the training log? This could help me reproduce the result efficiently because I do not have such computational resources (32 V100 GPUs)

[wjn922]

Sure. Please see log.txt.

[itruonghai]

@wjn922 Could you share all the log files, including the joint training across models, as well as the finetuning stages? Thanks for your time.