Closed junming-yang closed 2 months ago
In our training and evaluation, we pad the image with siglip.image_processor.image_mean (0.5) * 255 = 127. And it doesn't work for grayscale image.
After changing it to 'white', the model can work but it may result in some performance loss (not sure.
How to correctly handle the inference of such black-and-white images?
image = Image.open('math.jpg').convert('RGB')
pillow == 10.2.0 transformers == 4.40.0 你好,在bunny-llama3图像处理中对于一些黑白照片,存在该报错问题,我将expand2square函数中的background_color由grey (127, 127, 127)修改为'white'后无报错,请问这样修改是否可以