We understand that the following condition has been newly added since the start of Phase 2:
Egocentric image evaluation now uses low resolution (960×1280); standard image resolution is unchanged.
However, when reviewing the code, it appears that all images are being resized. Could you please confirm if this is the case?
Also, regarding the batch_generate_response function of the agent, is it correct to assume that the input images are a mix of both (960×1280) and original resolutions?