in_chans=1 working w/ pretrained weights for vision_transformer

timm/models/vision_transformer.py

@@ -37,7 +37,7 @@ def _cfg(url='', **kwargs):
         'num_classes': 1000, 'input_size': (3, 224, 224), 'pool_size': None,
         'crop_pct': .9, 'interpolation': 'bicubic',
         'mean': IMAGENET_DEFAULT_MEAN, 'std': IMAGENET_DEFAULT_STD,
-        'first_conv': '', 'classifier': 'head',
+        'first_conv': 'patch_embed.proj', 'classifier': 'head',
         **kwargs
     }