Thanks for your reply,.
In fact, I have tried following your steps, but found that just changing the root directory path does not seem to work, because dist_fn_euclidean_imgbb_resnet50_capbb_bigru_embed_size_2048_batch_64_lr_0.0002_step_size_15_captype_sent_img_meltlayer_1_np_False is missing, How could I get it?
Training from --load_model_path 0 will get the same result.
Originally posted by @cuiaiyu in #1 (comment)