search

lihong2303 / AGM

[ICCV2023] The repo for "Boosting Multi-modal Model Performance with Adaptive Gradient Modulation".
MIT License
22 stars 2 forks source link

The code of calculating the strength of a modality seems incorrect. #11

Open fanyunfeng-bit opened 5 months ago

fanyunfeng-bit commented 5 months ago

image In paper, the strength of a modality is that its score is subtracted from the scores of the other modalities, but in the code it seems to be in reverse order. 图片1

lihong2303 commented 5 months ago

Thank you for your detailed observations concerning our work. The discrepancy you pointed out can be ascribed to the computation of the CE, within which there exists an additional subtraction.