Set presion of torch to high for consistency with training and faster inference.

Update the model codes, including the previous inconsistencies.

Update the weights to the one for general use with massive training.

Fix app.py (#2)

Make the comparison effect of the results more obvious (#3)

ZeroGPU and ImageSlider (#1)

Update the approximated time cost from 500s to 40s. The long time cost may be caused by the unnecessary part of our_ref in inference.

Remove redundant part of our_ref in inference.

Adapt the resize to cv2 version to keep consistent with that used in training.

Update the well-trained weights.

Add the link to my github repo.

Add the resolution value for examples.

Add option for resolution in text box to do worse but faster prediction.

Initialization on my BiRefNet online demo.

Initialization on my BiRefNet online demo.