fix

image2text.py

@@ -5,7 +5,7 @@
 from .wd_v3_model import WdV3Model
 from PIL import Image
 import numpy as np
-from .utils import remove_specific_patterns
+from .utils import remove_specific_patterns,tensor2pil
 
 GLOBAL_WdV3Model = None
 
@@ -81,13 +81,13 @@ def get_value(self, model, image, query, custom_query, print_log):
         # Iterate over each batch of images
         for img in image:
             # Convert Tensor to image
-            img = Image.fromarray(
-                np.clip(255.0 * img.cpu().numpy(), 0, 255).astype(np.uint8)
-            )
+
+            img = tensor2pil(img)
+            img = img.convert("RGB")
             # Additional processing for specific models
             if model.name == "internlm":
                 query = f"<ImageHere>{query}"
-
+            
             result = model.answer_question(img, query)
             if print_log:
                 print(result)

text2prompt.py

@@ -1,5 +1,6 @@
 from .qwen_model import QwenModel
 
+
 class LoadText2PromptModel:
     def __init__(self):
         pass

utils.py

@@ -11,6 +11,10 @@ def tensor2pil(image):
     return Image.fromarray(np.clip(255. * image.cpu().numpy().squeeze(), 0, 255).astype(np.uint8))
 
 
+
+
+
+
 def remove_specific_patterns(text):
     # 构建一个匹配特定词汇前面有数字的正则表达式
     pattern = r"\d+(girl|boy|man|person|woman|men|women|girls|boys|mans),*\b"