sarena: explicit LPIPS model path via LMUDataRoot, use cached tokenizer path

Author: TianhaoLiang2000

vlmeval/dataset/utils/SArena/LPIPS.py

@@ -1,18 +1,21 @@
+import os
 import torch
 import lpips
 
 from tqdm import tqdm
 from torch.utils.data import DataLoader
 from torchvision.transforms import ToTensor, Normalize
 from .base_metric import BaseMetric
+from vlmeval.smp.file import LMUDataRoot
 
 
 class LPIPSCalculator(BaseMetric):
     def __init__(self):
         super().__init__()
         self.class_name = self.__class__.__name__
         self.device = "cuda" if torch.cuda.is_available() else "cpu"
-        self.model = lpips.LPIPS(net='vgg').to(self.device)
+        model_path = os.path.join(LMUDataRoot(), 'aux_models', 'vgg.pth')
+        self.model = lpips.LPIPS(net='vgg', model_path=model_path).to(self.device)
         self.metric = self.LPIPS
         self.to_tensor = ToTensor()
         self.normalize = Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225])

vlmeval/dataset/utils/SArena/token_length.py

@@ -7,12 +7,18 @@
 from .base_metric import BaseMetric
 from .average_meter import AverageMeter
 
+from vlmeval.smp.misc import get_cache_path
+
 
 class TokenLengthCalculator(BaseMetric):
+
     def __init__(self, tokenizer_path: str):
         super().__init__()
         self.class_name = self.__class__.__name__
         self.device = "cuda" if torch.cuda.is_available() else "cpu"
+        if get_cache_path(tokenizer_path, repo_type='models'):
+            # AutoTokenizer won't use cached path.
+            tokenizer_path = get_cache_path(tokenizer_path, repo_type='models')
         self.tokenizer = AutoTokenizer.from_pretrained(tokenizer_path)
         self.metric = self.count_token_length