|
@@ -22,10 +22,13 @@ try:
|
|
|
except Exception:
|
|
|
print("[warning] EVA-CLIP 没有预处理配置,使用默认参数构造 ImageProcessor")
|
|
|
from transformers import CLIPImageProcessor
|
|
|
- image_processor = CLIPImageProcessor(size={"shortest_edge": 224}, resample=3,
|
|
|
- crop_size={"height": 224, "width": 224},
|
|
|
- image_mean=[0.48145466, 0.4578275, 0.40821073],
|
|
|
- image_std=[0.26862954, 0.26130258, 0.27577711])
|
|
|
+ image_processor = CLIPImageProcessor(
|
|
|
+ size={"shortest_edge": 448}, # 最短边 resize 到 448
|
|
|
+ resample=3,
|
|
|
+ crop_size={"height": 448, "width": 448}, # 中心裁剪 448×448
|
|
|
+ image_mean=[0.48145466, 0.4578275, 0.40821073],
|
|
|
+ image_std=[0.26862954, 0.26130258, 0.27577711]
|
|
|
+ )
|
|
|
|
|
|
# 如果后续要做 text embedding,可以加 tokenizer
|
|
|
tokenizer = AutoTokenizer.from_pretrained(MODEL_NAME)
|