shengyi4

> You can create a blip2_retrieval model by modifying [blip2_qformer](https://github.com/salesforce/LAVIS/blob/main/lavis/models/blip2_models/blip2_qformer.py) to take into account `samples["image_id"]` when computing ITC and ITM, as done in [blip_retrieval](https://github.com/salesforce/LAVIS/blob/main/lavis/models/blip_models/blip_retrieval.py). > > Then, you can create...

Thanks for your prompt reply! I trained the retrieval model on my own dataset, but here is the evaluate.txt output: {"txt_r1": 0.3333333333333333, "txt_r5": 0.6666666666666666, "txt_r10": 2.6666666666666665, "txt_r_mean": 1.222222222222222, "img_r1": 0.25170801869830994,...