Comments bug in FluxPriorReduxPipeline

[chenxiao111222]

diffusers/src/diffusers/pipelines/flux/pipeline_flux_prior_redux.py

Lines 388 to 392 in fdcbbdf

	image (`torch.Tensor`, `PIL.Image.Image`, `np.ndarray`, `List[torch.Tensor]`, `List[PIL.Image.Image]`, or `List[np.ndarray]`):
	`Image`, numpy array or tensor representing an image batch to be used as the starting point. For both
	numpy array and pytorch tensor, the expected value range is between `[0, 1]` If it's a tensor or a list
	or tensors, the expected shape should be `(B, C, H, W)` or `(C, H, W)`. If it is a numpy array or a
	list of arrays, the expected shape should be `(B, H, W, C)` or `(H, W, C)`

For both numpy array and pytorch tensor, the expected value range should be between [0, 255], not [0, 1].

[yiyixuxu]

we expect [0,1] because our pipelines also output [0,1] for if the output_type is np or pt

[chenxiao111222]

we expect [0,1] because our pipelines also output [0,1] for if the output_type is np or pt

But when I send a ndarray/tensor image with range 0~1 to redux pipeline, the result is a pure black image. Maybe SiglipImageProcessor should be modified to adapt the input.

[github-actions]

This issue has been automatically marked as stale because it has not had recent activity. If you think this still needs to be addressed please comment on this thread.

Please note that issues that do not follow the contributing guidelines are likely to be ignored.

[a-r-r-o-w]

not stale

[chenxiao111222]

Here is a piece of code to reproduce the problem.

from PIL import Image
import torch
from torchvision.transforms.functional import to_tensor
from diffusers.utils import load_image
from diffusers.pipelines.flux.pipeline_flux_prior_redux import FluxPriorReduxPipeline


redux_id = "black-forest-labs/FLUX.1-Redux-dev"
pipeline_redux = FluxPriorReduxPipeline.from_pretrained(
    redux_id, 
    torch_dtype=torch.bfloat16
)
pipeline_redux.to(device='cuda')


test_pil_image = load_image("https://huggingface.co/datasets/YiYiXu/testing-images/resolve/main/style_ziggy/img5.png")
test_image = to_tensor(test_pil_image).unsqueeze(0)


images = pipeline_redux.feature_extractor.preprocess(images=test_image, do_resize=True, return_tensors="pt", do_convert_rgb=True)
print(images.pixel_values.min().item(), images.pixel_values.max().item())  # -0.9998462200164795 -0.9921568632125854
images = pipeline_redux.feature_extractor.preprocess(images=test_image * 255, do_resize=True, return_tensors="pt", do_convert_rgb=True)
print(images.pixel_values.min().item(), images.pixel_values.max().item())  # -0.9607843160629272 1.0

When I input a tensor with values ranging between 0 and 1 as per the comments, it is obvious that output of feature_extractor.preprocess is incorrect.

At the same time, a prompt appears out of nowhere:

It looks like you are trying to rescale already rescaled images. If the input images have pixel values between 0 and 1, set `do_rescale=False` to avoid rescaling them again.

When I adjust the value range of the tensor from 0 to 255, the results meet the expectations.

@yiyixuxu @a-r-r-o-w