Where in the code is randomized intensities?

[bansh123]

I couldn't find it in the code.
It seems like just setting t_0=0.5.

[brandontrabucco]

Hello,

The code for randomizing the intensity strength is in two parts:

da-fusion/train_classifier.py

Lines 113 to 117 in dbcdc9c

	for (aug, guidance_scale,
	strength, mask, inverted) in zip(
	aug, guidance_scale,
	strength, mask, inverted
	)

This first part defines multiple augmentations with different intensities to sample from. The following second part samples one of the intensities:

da-fusion/semantic_aug/augmentations/compose.py

Lines 49 to 56 in dbcdc9c

	def forward(self, image: Image.Image, label: int,
	metadata: dict) -> Tuple[Image.Image, int]:
	idx = np.random.choice(len(self.probs), p=self.probs)
	image, label = self.augs[idx](image, label, metadata)
	return image, label

By setting multiple values for --strength when calling train_classifier.py, such as as done here:

da-fusion/scripts/stacking/launch_textual_inversion=1.0-0.75-0.5-0.25_coco.sh

Lines 22 to 26 in dbcdc9c

	--guidance-scale 7.5 7.5 7.5 7.5 \
	--strength 1.0 0.75 0.5 0.25 \
	--mask 0 0 0 0 \
	--inverted 0 0 0 0 \
	--probs 0.25 0.25 0.25 0.25 \

the intensity will be sampled uniformly from a list given via the command line

[bansh123]

Thanks!
I have one more question.
Can you provide instruction to obtain tokens from textual inversion for reproducing the results in figure 5? ( Or download link for learned tokens )