-
-
Notifications
You must be signed in to change notification settings - Fork 1.8k
Open
@isaaccorley
Description
The following code using a plain ViT-B/16 as a backbone for a U-Net/DeepLabv3+/SegFormer/UPerNet returns the following trace:
model = smp.create_model("unet", "tu-vit_base_patch16_224", encoder_weights=None)
Traceback (most recent call last): File "<input>", line 1, in <module> model = smp.create_model("unet", "tu-vit_base_patch16_224", encoder_weights=None, in_chann els=5, classes=2) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ ^^^^^^^^^^^^^^^^^^ File "/Users/isaaccorley/miniconda3/envs/torchgeo/lib/python3.11/site-packages/segmentation_ models_pytorch/__init__.py", line 63, in create_model return model_class( ^^^^^^^^^^^^ File "/Users/isaaccorley/miniconda3/envs/torchgeo/lib/python3.11/site-packages/segmentation_ models_pytorch/base/hub_mixin.py", line 153, in wrapper return func(self, *args, **kwargs) ^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "/Users/isaaccorley/miniconda3/envs/torchgeo/lib/python3.11/site-packages/segmentation_ models_pytorch/decoders/unet/model.py", line 132, in __init__ self.encoder = get_encoder( ^^^^^^^^^^^^ File "/Users/isaaccorley/miniconda3/envs/torchgeo/lib/python3.11/site-packages/segmentation_ models_pytorch/encoders/__init__.py", line 87, in get_encoder encoder = TimmUniversalEncoder( ^^^^^^^^^^^^^^^^^^^^^ File "/Users/isaaccorley/miniconda3/envs/torchgeo/lib/python3.11/site-packages/segmentation_ models_pytorch/encoders/timm_universal.py", line 121, in __init__ raise ValueError("Unsupported model downsampling pattern.") ValueError: Unsupported model downsampling pattern. >>> model = smp.create_model("unet", "tu-vit_base_patch16_224", encoder_weights=None) Traceback (most recent call last): File "<input>", line 1, in <module> model = smp.create_model("unet", "tu-vit_base_patch16_224", encoder_weights=None) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "/Users/isaaccorley/miniconda3/envs/torchgeo/lib/python3.11/site-packages/segmentation_ models_pytorch/__init__.py", line 63, in create_model return model_class( ^^^^^^^^^^^^ File "/Users/isaaccorley/miniconda3/envs/torchgeo/lib/python3.11/site-packages/segmentation_ models_pytorch/base/hub_mixin.py", line 153, in wrapper return func(self, *args, **kwargs) ^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "/Users/isaaccorley/miniconda3/envs/torchgeo/lib/python3.11/site-packages/segmentation_ models_pytorch/decoders/unet/model.py", line 132, in __init__ self.encoder = get_encoder( ^^^^^^^^^^^^ File "/Users/isaaccorley/miniconda3/envs/torchgeo/lib/python3.11/site-packages/segmentation_ models_pytorch/encoders/__init__.py", line 87, in get_encoder encoder = TimmUniversalEncoder( ^^^^^^^^^^^^^^^^^^^^^ File "/Users/isaaccorley/miniconda3/envs/torchgeo/lib/python3.11/site-packages/segmentation_ models_pytorch/encoders/timm_universal.py", line 121, in __init__ raise ValueError("Unsupported model downsampling pattern.") ValueError: Unsupported model downsampling pattern.
However this was made to work in TorchSeg https://github.com/isaaccorley/torchseg.
Any solution here?
Metadata
Metadata
Assignees
Labels
No labels