| 124 | self.normalize_transform = transforms.Normalize(mean=image_mean, std=image_std, inplace=True) |
| 125 | |
| 126 | def __call__(self, img, img_num=1): |
| 127 | |
| 128 | image = convert_to_rgb(img) |
| 129 | image = to_numpy_array(image) |
| 130 | image = resize( |
| 131 | image, |
| 132 | size=(448, 448), |
| 133 | resample=3 |
| 134 | ) |
| 135 | # image = self.rescale(image=image, scale= 0.00392156862745098, input_data_format=input_data_format) |
| 136 | # image = self.normalize( |
| 137 | # image=image, |
| 138 | # mean=image_mean, |
| 139 | # std=image_std, |
| 140 | # input_data_format=input_data_format, |
| 141 | # ) |
| 142 | # return image |
| 143 | # img = img.permute(2, 0, 1) |
| 144 | # print('image numpy', image.shape) |
| 145 | img = self.to_tensor_transform(image) |
| 146 | # img = img.permute(2, 0, 1) |
| 147 | # print('img', img.shape) |
| 148 | img = self.normalize_transform(img) |
| 149 | return img |
| 150 | |
| 151 | class QwenVL2ImageTransform: |
| 152 | def __init__( |