Hi, in my computer science project, I'm using CLIP's text encoder and vision encoder to extract image-text features:
image_features = self.encode_image(image)
text_features = self.encode_text(text)
If I directly compute loss using these image_features and text_features with a custom loss function (without invoking CLIP's forward function), would this affect the interpretability of the project?
Hi, in my computer science project, I'm using CLIP's text encoder and vision encoder to extract image-text features:
image_features = self.encode_image(image)
text_features = self.encode_text(text)
If I directly compute loss using these image_features and text_features with a custom loss function (without invoking CLIP's forward function), would this affect the interpretability of the project?