feat: Add model and BERT model download functionality

- Implemented logic to download the main model and BERT model if not present locally
- Added remote URLs and MD5 checksums for model verification

annolid/detector/countgd/models/__init__.py

@@ -7,4 +7,4 @@
 from .GroundingDINO import build_groundingdino
 
 def build_model(args):
-    return build(args)
+    return build_groundingdino(args)

annolid/detector/countgd/predict.py

@@ -1,6 +1,7 @@
 import random
 import torch
 import os
+import gdown
 from PIL import Image
 import numpy as np
 import argparse
@@ -37,6 +38,24 @@ def __init__(self, model_path: str = "checkpoint_best_regular.pth",
         self.here = os.path.dirname(os.path.abspath(__file__))
         config_path = os.path.join(self.here, config_path)
         model_path = os.path.join(self.here, model_path)
+        self._REMOTE_MODEL_URL = "https://github.com/healthonrails/annolid/releases/download/v1.2.0/checkpoint_best_regular.pth"
+        self._MD5 = "1492bfdd161ac1de471d0aafb32b174d"
+        if not os.path.exists(model_path):
+            gdown.cached_download(self._REMOTE_MODEL_URL,
+                                  model_path,
+                                  md5=self._MD5
+                                  )
+
+        self._REMOTE_BERT_MODEL_URL = "https://github.com/healthonrails/annolid/releases/download/v1.2.0/model.safetensors"
+        self._BERT_MD5 = "cd18ceb6b110c04a8033ce01de41b0b7"
+        self._BERT_MODEL_PATH = os.path.join(
+            self.here, "checkpoints/bert-base-uncased/model.safetensors")
+        if not os.path.exists(self._BERT_MODEL_PATH):
+            gdown.cached_download(self._REMOTE_BERT_MODEL_URL,
+                                  self._BERT_MODEL_PATH,
+                                  md5=self._BERT_MD5
+                                  )
+
         self.model = self._load_model(model_path, config_path, self.device)
         self.transform = self._build_transforms()