Missing annotations_cleaned_v2.json and unclear data preprocessing instructions

[bkxmw]

Hello, thank you for your great work! I encountered an issue when trying to run the eval_qwen2_vl_3d.py script. The script requires a file named annotations_cleaned_v2.json, which appears to be missing from the repository or the official dataset package.
Furthermore, after downloading the complete dataset, I am unsure about the correct data preprocessing steps. The code references a directory called keyframes_resized, but I could not find clear instructions or scripts to generate this resized keyframes data from the raw videos.

Could you please provide:

• The missing annotations_cleaned_v2.json file, or
• Instructions on how to generate this file, or
• Detailed guidance on the data preprocessing steps ?

Thank you for your time and assistance!