支持本地的huggingface

.gitignore

@@ -1,3 +1,4 @@
+test_weight/
 # IDE
 .idea/
 .vscode/

demo/inference_on_a_image.py

@@ -27,6 +27,7 @@ def plot_boxes_to_image(image_pil, tgt):
     for box, label in zip(boxes, labels):
         # from 0..1 to 0..W, 0..H
         box = box * torch.Tensor([W, H, W, H])
+        print("label:",label)
         # from xywh to xyxy
         box[:2] -= box[2:] / 2
         box[2:] += box[:2]

groundingdino/util/get_tokenlizer.py

@@ -1,3 +1,4 @@
+import os
 from transformers import AutoTokenizer, BertModel, BertTokenizer, RobertaModel, RobertaTokenizerFast
 
 
@@ -23,4 +24,7 @@ def get_pretrained_language_model(text_encoder_type):
         return BertModel.from_pretrained(text_encoder_type)
     if text_encoder_type == "roberta-base":
         return RobertaModel.from_pretrained(text_encoder_type)
+
+    if os.path.isdir(text_encoder_type):
+        return BertModel.from_pretrained(text_encoder_type)
     raise ValueError("Unknown text_encoder_type {}".format(text_encoder_type))

groundingdino/version.py

@@ -1 +1 @@
-__version__ = "0.1.0"
+__version__ = '0.1.0'