Method get_cond_vec

models/clipseg.py:202–223

(self, conditional, batch_size)

        return [promt.format(w) for promt, w in zip(prompts, words)]

    def get_cond_vec(self, conditional, batch_size):
        # compute conditional from a single string
        if conditional is not None and type(conditional) == str:
            cond = self.compute_conditional(conditional)
            cond = cond.repeat(batch_size, 1)

        # compute conditional from string list/tuple
        elif conditional is not None and type(conditional) in {list, tuple} and type(conditional[0]) == str:
            assert len(conditional) == batch_size
            cond = self.compute_conditional(conditional)

        # use conditional directly
        elif conditional is not None and type(conditional) == torch.Tensor and conditional.ndim == 2:
            cond = conditional

        # compute conditional from image
        elif conditional is not None and type(conditional) == torch.Tensor:
            with torch.no_grad():
                cond, _, _ = self.visual_forward(conditional)
        else:
            raise ValueError('invalid conditional')
        return cond   

    def compute_conditional(self, conditional):
        import clip