I did not understand tags. I think "cc" should be in "if condition". please clarify the use of each "attr", "sc", "t2i", 'ctcc', 'cc'.

[LokeshBorawar]

In pass_transformer_joint.py file and when training is false at line 409

if ('attr' in self.net_config.test_task_type or 'sc' in self.net_config.test_task_type or 't2i' in self.net_config.test_task_type or 'cross' in self.net_config.test_task_type) and 'ctcc' not in self.net_config.test_task_type:

I think "cc" should be in "if condition", because if self.net_config.test_task_type has not "cc" code will jump to "text_embeds = self.visual_encoder_m(instruction)" but for this instruction should be an image.

please also clarify me the use of each "attr", "sc", "t2i", 'ctcc', 'cc'.

[DejaYang]

I do think so. And I think datalise in ltcc dataset is wrong about "white_shirt.jpg", it should be the real clothes image path.

[Marches7]

From my point of view:
ctcc: clothes template clothing changing
cc: clothing changing
t2i: text to instrcution
cross: cross modality (Visible-Infrad)
sc: traditional (preprocessor_sc.py has a variable attr_item='do not change clothes', that strikes me).
attr: Language-instructed