search

arcadelab / FastSAM3D

Code for "FastSAM3D: An Efficient Segment Anything Model for 3D Volumetric Medical Images"
https://arxiv.org/abs/2403.09827
Apache License 2.0
68 stars 4 forks source link

Size mismatch between checkpoint and existing models. #7

Closed williamtrayn0r closed 1 month ago

williamtrayn0r commented 1 month ago

Hello,

When trying to run ./infer.sh on my data, I use the FastSAM3D.pth checkpoint provided. However, I get an error message when trying to load the state dictionary of the checkpoint into the default model.

I've attached the message below. I'm unsure if I am running infer.sh incorrectly but I wanted to ask in case it was an easy fix.

(FastSAM3D) [william.traynor@DT002135 FastSAM3D-v1]$ ./infer.sh
set seed as 2023
get 1 datasets
Dataset size: 2
Dataset size: 2
device: cuda
Traceback (most recent call last):
  File "/home/william.traynor@MRECANON/FastSAM3D-v1/validation.py", line 450, in <module>
    sam_model_tune.load_state_dict(state_dict)
  File "/home/william.traynor@MRECANON/.conda/envs/FastSAM3D/lib/python3.9/site-packages/torch/nn/modules/module.py", line 2215, in load_state_dict
    raise RuntimeError('Error(s) in loading state_dict for {}:\n\t{}'.format(
RuntimeError: Error(s) in loading state_dict for Sam3D:
        Missing key(s) in state_dict: "image_encoder.blocks.0.norm1.weight", "image_encoder.blocks.0.norm1.bias", "image_encoder.blocks.0.attn.rel_pos_d", "image_encoder.blocks.0.attn.rel_pos_h", "image_encoder.blocks.0.attn.rel_pos_w", "image_encoder.blocks.0.attn.qkv.weight", "image_encoder.blocks.0.attn.qkv.bias", "image_encoder.blocks.0.attn.proj.weight", "image_encoder.blocks.0.attn.proj.bias", "image_encoder.blocks.1.norm1.weight", "image_encoder.blocks.1.norm1.bias", "image_encoder.blocks.1.attn.rel_pos_d", "image_encoder.blocks.1.attn.rel_pos_h", "image_encoder.blocks.1.attn.rel_pos_w", "image_encoder.blocks.1.attn.qkv.weight", "image_encoder.blocks.1.attn.qkv.bias", "image_encoder.blocks.1.attn.proj.weight", "image_encoder.blocks.1.attn.proj.bias", "image_encoder.blocks.6.norm1.weight", "image_encoder.blocks.6.norm1.bias", "image_encoder.blocks.6.attn.rel_pos_d", "image_encoder.blocks.6.attn.rel_pos_h", "image_encoder.blocks.6.attn.rel_pos_w", "image_encoder.blocks.6.attn.qkv.weight", "image_encoder.blocks.6.attn.qkv.bias", "image_encoder.blocks.6.attn.proj.weight", "image_encoder.blocks.6.attn.proj.bias", "image_encoder.blocks.6.norm2.weight", "image_encoder.blocks.6.norm2.bias", "image_encoder.blocks.6.mlp.lin1.weight", "image_encoder.blocks.6.mlp.lin1.bias", "image_encoder.blocks.6.mlp.lin2.weight", "image_encoder.blocks.6.mlp.lin2.bias", "image_encoder.blocks.7.norm1.weight", "image_encoder.blocks.7.norm1.bias", "image_encoder.blocks.7.attn.rel_pos_d", "image_encoder.blocks.7.attn.rel_pos_h", "image_encoder.blocks.7.attn.rel_pos_w", "image_encoder.blocks.7.attn.qkv.weight", "image_encoder.blocks.7.attn.qkv.bias", "image_encoder.blocks.7.attn.proj.weight", "image_encoder.blocks.7.attn.proj.bias", "image_encoder.blocks.7.norm2.weight", "image_encoder.blocks.7.norm2.bias", "image_encoder.blocks.7.mlp.lin1.weight", "image_encoder.blocks.7.mlp.lin1.bias", "image_encoder.blocks.7.mlp.lin2.weight", "image_encoder.blocks.7.mlp.lin2.bias", "image_encoder.blocks.8.norm1.weight", "image_encoder.blocks.8.norm1.bias", "image_encoder.blocks.8.attn.rel_pos_d", "image_encoder.blocks.8.attn.rel_pos_h", "image_encoder.blocks.8.attn.rel_pos_w", "image_encoder.blocks.8.attn.qkv.weight", "image_encoder.blocks.8.attn.qkv.bias", "image_encoder.blocks.8.attn.proj.weight", "image_encoder.blocks.8.attn.proj.bias", "image_encoder.blocks.8.norm2.weight", "image_encoder.blocks.8.norm2.bias", "image_encoder.blocks.8.mlp.lin1.weight", "image_encoder.blocks.8.mlp.lin1.bias", "image_encoder.blocks.8.mlp.lin2.weight", "image_encoder.blocks.8.mlp.lin2.bias", "image_encoder.blocks.9.norm1.weight", "image_encoder.blocks.9.norm1.bias", "image_encoder.blocks.9.attn.rel_pos_d", "image_encoder.blocks.9.attn.rel_pos_h", "image_encoder.blocks.9.attn.rel_pos_w", "image_encoder.blocks.9.attn.qkv.weight", "image_encoder.blocks.9.attn.qkv.bias", "image_encoder.blocks.9.attn.proj.weight", "image_encoder.blocks.9.attn.proj.bias", "image_encoder.blocks.9.norm2.weight", "image_encoder.blocks.9.norm2.bias", "image_encoder.blocks.9.mlp.lin1.weight", "image_encoder.blocks.9.mlp.lin1.bias", "image_encoder.blocks.9.mlp.lin2.weight", "image_encoder.blocks.9.mlp.lin2.bias", "image_encoder.blocks.10.norm1.weight", "image_encoder.blocks.10.norm1.bias", "image_encoder.blocks.10.attn.rel_pos_d", "image_encoder.blocks.10.attn.rel_pos_h", "image_encoder.blocks.10.attn.rel_pos_w", "image_encoder.blocks.10.attn.qkv.weight", "image_encoder.blocks.10.attn.qkv.bias", "image_encoder.blocks.10.attn.proj.weight", "image_encoder.blocks.10.attn.proj.bias", "image_encoder.blocks.10.norm2.weight", "image_encoder.blocks.10.norm2.bias", "image_encoder.blocks.10.mlp.lin1.weight", "image_encoder.blocks.10.mlp.lin1.bias", "image_encoder.blocks.10.mlp.lin2.weight", "image_encoder.blocks.10.mlp.lin2.bias", "image_encoder.blocks.11.norm1.weight", "image_encoder.blocks.11.norm1.bias", "image_encoder.blocks.11.attn.rel_pos_d", "image_encoder.blocks.11.attn.rel_pos_h", "image_encoder.blocks.11.attn.rel_pos_w", "image_encoder.blocks.11.attn.qkv.weight", "image_encoder.blocks.11.attn.qkv.bias", "image_encoder.blocks.11.attn.proj.weight", "image_encoder.blocks.11.attn.proj.bias", "image_encoder.blocks.11.norm2.weight", "image_encoder.blocks.11.norm2.bias", "image_encoder.blocks.11.mlp.lin1.weight", "image_encoder.blocks.11.mlp.lin1.bias", "image_encoder.blocks.11.mlp.lin2.weight", "image_encoder.blocks.11.mlp.lin2.bias". 
        size mismatch for image_encoder.blocks.2.attn.rel_pos_d: copying a param with shape torch.Size([15, 128]) from checkpoint, the shape in current model is torch.Size([15, 64]).
        size mismatch for image_encoder.blocks.2.attn.rel_pos_h: copying a param with shape torch.Size([15, 128]) from checkpoint, the shape in current model is torch.Size([15, 64]).
        size mismatch for image_encoder.blocks.2.attn.rel_pos_w: copying a param with shape torch.Size([15, 128]) from checkpoint, the shape in current model is torch.Size([15, 64]).
        size mismatch for image_encoder.blocks.3.attn.rel_pos_d: copying a param with shape torch.Size([15, 128]) from checkpoint, the shape in current model is torch.Size([27, 64]).
        size mismatch for image_encoder.blocks.3.attn.rel_pos_h: copying a param with shape torch.Size([15, 128]) from checkpoint, the shape in current model is torch.Size([27, 64]).
        size mismatch for image_encoder.blocks.3.attn.rel_pos_w: copying a param with shape torch.Size([15, 128]) from checkpoint, the shape in current model is torch.Size([27, 64]).
        size mismatch for image_encoder.blocks.4.attn.rel_pos_d: copying a param with shape torch.Size([15, 128]) from checkpoint, the shape in current model is torch.Size([27, 64]).
        size mismatch for image_encoder.blocks.4.attn.rel_pos_h: copying a param with shape torch.Size([15, 128]) from checkpoint, the shape in current model is torch.Size([27, 64]).
        size mismatch for image_encoder.blocks.4.attn.rel_pos_w: copying a param with shape torch.Size([15, 128]) from checkpoint, the shape in current model is torch.Size([27, 64]).
        size mismatch for image_encoder.blocks.5.attn.rel_pos_d: copying a param with shape torch.Size([15, 128]) from checkpoint, the shape in current model is torch.Size([15, 64]).
        size mismatch for image_encoder.blocks.5.attn.rel_pos_h: copying a param with shape torch.Size([15, 128]) from checkpoint, the shape in current model is torch.Size([15, 64]).
        size mismatch for image_encoder.blocks.5.attn.rel_pos_w: copying a param with shape torch.Size([15, 128]) from checkpoint, the shape in current model is torch.Size([15, 64]).
has context menu