| 106 | # return sim_samples_to_img, sim_samples_to_text |
| 107 | import re |
| 108 | class ImageDirEvaluator(CLIPEvaluator): |
| 109 | def __init__(self, device, clip_model='ViT-B/32') -> None: |
| 110 | super().__init__(device, clip_model) |
| 111 | |
| 112 | def evaluate(self, gen_samples, src_images, target_text): |
| 113 | |
| 114 | sim_samples_to_img = self.img_to_img_similarity(src_images, gen_samples) |
| 115 | # sim_samples_to_text = self.txt_to_img_similarity(target_text.replace("*", ""), gen_samples) |
| 116 | clean_text = re.sub(r"<new\d+>", " ", target_text).strip() |
| 117 | sim_samples_to_text = self.txt_to_img_similarity(clean_text, gen_samples) |
| 118 | # sim_samples_to_text = self.txt_to_img_similarity(target_text.replace("<new1>", ""), gen_samples) |
| 119 | |
| 120 | return sim_samples_to_img, sim_samples_to_text |
| 121 | |
| 122 | import argparse |
| 123 | from ldmm.data.personalized import PersonalizedBase |