Query regarding image resolution for SeeCoder during training

Hello,
I am currently delving into your intriguing SeeCoder, and have a question that I could not find an answer to in the paper. Could you please clarify the resolution of the batch input images used during the training process? 