Issue with Dataset Format Conversion for the COCO Dataset formet of RES/REC Task

[Hearum]

Hi！ Thank you for your work on this project; it's a great and awosome resource for the community working on ComputerVision.
I am currently attempting to prepare the RES&RES dataset as described in the README, specifically the part where it mentions running the script for converting the dataset format to COCO format using:
python3 conversion/convert_mix_ref.py
The README suggests this script will convert the original JSON files to COCO format and merge them into one dataset (refcoco-mixed). I have downloaded the dataset files from https://drive.google.com/drive/folders/1IXnSieVr5CHF2pVJpj0DlwC6R3SbfolU

However, I've encountered an issue where the format of the downloaded dataset files does not seem to match the expected format by convert_mix_ref.py script. It appears there might be a discrepancy between the format of the dataset provided and the expectations of the script, leading to difficulties in executing the conversion properly. It seems need 'instances_train.json', but download 'instances.json', and loss necessary key of the dict.
Could you please provide some guidance on how to resolve this issue? I am looking for information. Whether there are updated versions of the dataset or the script that I should use. Or any additional steps or modifications to the script that might be necessary to accommodate the current format of the dataset files.