Clarification on the use of local_query during inference

[qcy22]

Hi, thanks for your great work and for sharing the code!

In the paper, you mention:

"we first introduce a set of learnable queries 𝑄 to extract neighborhood-aware features from each modality."

However, while reading the inference code (modeling_earthmind_chat.py), I noticed that the local_query variable seems to be defined but not explicitly used during inference.

Could you please clarify whether local_query is supposed to play a role during inference? Or is it only used during training ?

Thanks in advance for your time!

[shuyansy]

hi, sorry for late reply. we use local_query both in training and inference time. The current code only contains singe image inference so it does not contain local_query operation which functions when you input rgb and sar data together. I will organized the multi-sensor inference code in one week and let you know.

[lydj-zhuzhu]

Thank you very much for your outstanding work and the latest code update!
We noticed your previous reply mentioning that you would sort out the multi-sensor reasoning code within a week, and we have checked the latest updated code warehouse today. However, we found that the code (including train/test scripts, eval scripts and modeling files) still does not include any logic related to multi-sensor processing (e.g., RGB+SAR multi-modal feature extraction/fusion, local_query usage in multi-sensor scenarios, multi-sensor training pipeline).
We are very interested in reproducing the multi-sensor related experiments in your paper, so we would like to kindly ask:
When will the complete multi-sensor code (including both reasoning and training pipelines) be publicly released?
Is there a preliminary schedule for the release of multi-sensor dataset processing scripts (e.g., data loading/modal alignment for RGB+SAR)?
Thank you again for your time and efforts, and we look forward to your reply!