No module named 'blip_utils'

[Lotsoliu]

Hi, I'm new to this task, it seems not work without blip_utils package, but I don't know where to get it.

Traceback (most recent call last): File "./model/train.py", line 7, in from utils import get_args, set_manualSeed, image_transform, WinoLoss, CLIPLoss, MarginLoss

from blip_utils import get_rank, all_gather_batch

ModuleNotFoundError: No module named 'blip_utils'

[Lotsoliu]

[BigHyf]

Thanks for your attention, we don't use CLIPLoss in utils.py, so we don't need to use from blip_utils import get_rank, all_gather_batch. In the future, I will carefully organize the code and remove the irrelevant parts of the code.

[Lotsoliu]

thanks a lot!

[wdi-nancy]

Thanks for your attention, we don't use CLIPLoss in utils.py, so we don't need to use from blip_utils import get_rank, all_gather_batch. In the future, I will carefully organize the code and remove the irrelevant parts of the code.

请问作者为什么不用CLIPLoss了呢？因为论文里提到：in order to maintain the general representation ability of the model, we combine the original mini-batch image-text contrastive learning loss and the proposed loss for joint training. 删掉CLIPLoss之后，如何维持general representation ability？以及有什么实验去证明这个能力，因为从表格看貌似只有在vg上做relation和attribution的下游任务结果。