concepts.vision.fm_match.dino.get_feature.compute_dino_feature#
- compute_dino_feature(source_img, target_imgs=None, *, model_size='base', use_dino_v2=True, stride=None, edge_pad=False, pca=False, pca_dim=256, reusable_extractor=None)[source]#
- return: (result, resized_imgs, downsampled_imgs), where result is a tensor of shape (N, pca_dim, num_patches, num_patches),
resized_imgs is a list of PIL image_scene resized to the input size of the dino model, and downsampled_imgs is a list of PIL image_scene resized to the output size of the dino model.