About the implement of Image retrieval

[aleeyang]

what a great job in ICCV23! I have one question about the experiment on Image retrieval. The BOT ReId model you used inputs RGB images, which have three channels. However, Q_{event} G_{an_event} has only one channel. Did you train the BOT model on their training set, or did you use a pretrained model? If a pretrained model was used, how did you bridge the gap between RGB and grayscale images? Thank you for your time!