[Help wanted!] pretrained model mismatch

[l9761116]

File "demo_folder.py", line 86, in main
my_model = my_model.load_state_dict(torch.load(model_load_path))
File "/home/liuwanzhen2/anaconda3/envs/cydr/lib/python3.7/site-packages/torch/nn/modules/module.py", line 1407, in load_state_dict
self.class.name, "\n\t".join(error_msgs)))
RuntimeError: Error(s) in loading state_dict for cylinder_asym:
size mismatch for cylinder_3d_spconv_seg.downCntx.conv1.weight: copying a param with shape torch.Size([1, 3, 3, 16, 32]) from checkpoint, the shape in current model is torch.Size([32, 1, 3, 3, 16]).
size mismatch for cylinder_3d_spconv_seg.downCntx.conv1_2.weight: copying a param with shape torch.Size([3, 1, 3, 32, 32]) from checkpoint, the shape in current model is torch.Size([32, 3, 1, 3, 32]).
size mismatch for cylinder_3d_spconv_seg.downCntx.conv2.weight: copying a param with shape torch.Size([3, 1, 3, 16, 32]) from checkpoint, the shape in current model is torch.Size([32, 3, 1, 3, 16]).
size mismatch for cylinder_3d_spconv_seg.downCntx.conv3.weight: copying a param with shape torch.Size([1, 3, 3, 32, 32]) from checkpoint, the shape in current model is torch.Size([32, 1, 3, 3, 32]).
size mismatch for cylinder_3d_spconv_seg.resBlock2.conv1.weight: copying a param with shape torch.Size([3, 1, 3, 32, 64]) from checkpoint, the shape in current model is torch.Size([64, 3, 1, 3, 32]).
size mismatch for cylinder_3d_spconv_seg.resBlock2.conv1_2.weight: copying a param with shape torch.Size([1, 3, 3, 64, 64]) from checkpoint, the shape in current model is torch.Size([64, 1, 3, 3, 64]).
size mismatch for cylinder_3d_spconv_seg.resBlock2.conv2.weight: copying a param with shape torch.Size([1, 3, 3, 32, 64]) from checkpoint, the shape in current model is torch.Size([64, 1, 3, 3, 32]).
size mismatch for cylinder_3d_spconv_seg.resBlock2.conv3.weight: copying a param with shape torch.Size([3, 1, 3, 64, 64]) from checkpoint, the shape in current model is torch.Size([64, 3, 1, 3, 64]).
size mismatch for cylinder_3d_spconv_seg.resBlock2.pool.weight: copying a param with shape torch.Size([3, 3, 3, 64, 64]) from checkpoint, the shape in current model is torch.Size([64, 3, 3, 3, 64]).
size mismatch for cylinder_3d_spconv_seg.resBlock3.conv1.weight: copying a param with shape torch.Size([3, 1, 3, 64, 128]) from checkpoint, the shape in current model is torch.Size([128, 3, 1, 3, 64]).
size mismatch for cylinder_3d_spconv_seg.resBlock3.conv1_2.weight: copying a param with shape torch.Size([1, 3, 3, 128, 128]) from checkpoint, the shape in current model is torch.Size([128, 1, 3, 3, 128]).
size mismatch for cylinder_3d_spconv_seg.resBlock3.conv2.weight: copying a param with shape torch.Size([1, 3, 3, 64, 128]) from checkpoint, the shape in current model is torch.Size([128, 1, 3, 3, 64]).
size mismatch for cylinder_3d_spconv_seg.resBlock3.conv3.weight: copying a param with shape torch.Size([3, 1, 3, 128, 128]) from checkpoint, the shape in current model is torch.Size([128, 3, 1, 3, 128]).
size mismatch for cylinder_3d_spconv_seg.resBlock3.pool.weight: copying a param with shape torch.Size([3, 3, 3, 128, 128]) from checkpoint, the shape in current model is torch.Size([128, 3, 3, 3, 128]).
size mismatch for cylinder_3d_spconv_seg.resBlock4.conv1.weight: copying a param with shape torch.Size([3, 1, 3, 128, 256]) from checkpoint, the shape in current model is torch.Size([256, 3, 1, 3, 128]).
size mismatch for cylinder_3d_spconv_seg.resBlock4.conv1_2.weight: copying a param with shape torch.Size([1, 3, 3, 256, 256]) from checkpoint, the shape in current model is torch.Size([256, 1, 3, 3, 256]).
size mismatch for cylinder_3d_spconv_seg.resBlock4.conv2.weight: copying a param with shape torch.Size([1, 3, 3, 128, 256]) from checkpoint, the shape in current model is torch.Size([256, 1, 3, 3, 128]).
size mismatch for cylinder_3d_spconv_seg.resBlock4.conv3.weight: copying a param with shape torch.Size([3, 1, 3, 256, 256]) from checkpoint, the shape in current model is torch.Size([256, 3, 1, 3, 256]).
size mismatch for cylinder_3d_spconv_seg.resBlock4.pool.weight: copying a param with shape torch.Size([3, 3, 3, 256, 256]) from checkpoint, the shape in current model is torch.Size([256, 3, 3, 3, 256]).
size mismatch for cylinder_3d_spconv_seg.resBlock5.conv1.weight: copying a param with shape torch.Size([3, 1, 3, 256, 512]) from checkpoint, the shape in current model is torch.Size([512, 3, 1, 3, 256]).
size mismatch for cylinder_3d_spconv_seg.resBlock5.conv1_2.weight: copying a param with shape torch.Size([1, 3, 3, 512, 512]) from checkpoint, the shape in current model is torch.Size([512, 1, 3, 3, 512]).
size mismatch for cylinder_3d_spconv_seg.resBlock5.conv2.weight: copying a param with shape torch.Size([1, 3, 3, 256, 512]) from checkpoint, the shape in current model is torch.Size([512, 1, 3, 3, 256]).
size mismatch for cylinder_3d_spconv_seg.resBlock5.conv3.weight: copying a param with shape torch.Size([3, 1, 3, 512, 512]) from checkpoint, the shape in current model is torch.Size([512, 3, 1, 3, 512]).
size mismatch for cylinder_3d_spconv_seg.resBlock5.pool.weight: copying a param with shape torch.Size([3, 3, 3, 512, 512]) from checkpoint, the shape in current model is torch.Size([512, 3, 3, 3, 512]).
size mismatch for cylinder_3d_spconv_seg.upBlock0.trans_dilao.weight: copying a param with shape torch.Size([3, 3, 3, 512, 512]) from checkpoint, the shape in current model is torch.Size([512, 3, 3, 3, 512]).
size mismatch for cylinder_3d_spconv_seg.upBlock0.conv1.weight: copying a param with shape torch.Size([1, 3, 3, 512, 512]) from checkpoint, the shape in current model is torch.Size([512, 1, 3, 3, 512]).
size mismatch for cylinder_3d_spconv_seg.upBlock0.conv2.weight: copying a param with shape torch.Size([3, 1, 3, 512, 512]) from checkpoint, the shape in current model is torch.Size([512, 3, 1, 3, 512]).
size mismatch for cylinder_3d_spconv_seg.upBlock0.conv3.weight: copying a param with shape torch.Size([3, 3, 3, 512, 512]) from checkpoint, the shape in current model is torch.Size([512, 3, 3, 3, 512]).
size mismatch for cylinder_3d_spconv_seg.upBlock0.up_subm.weight: copying a param with shape torch.Size([3, 3, 3, 512, 512]) from checkpoint, the shape in current model is torch.Size([512, 3, 3, 3, 512]).
size mismatch for cylinder_3d_spconv_seg.upBlock1.trans_dilao.weight: copying a param with shape torch.Size([3, 3, 3, 512, 256]) from checkpoint, the shape in current model is torch.Size([256, 3, 3, 3, 512]).
size mismatch for cylinder_3d_spconv_seg.upBlock1.conv1.weight: copying a param with shape torch.Size([1, 3, 3, 256, 256]) from checkpoint, the shape in current model is torch.Size([256, 1, 3, 3, 256]).
size mismatch for cylinder_3d_spconv_seg.upBlock1.conv2.weight: copying a param with shape torch.Size([3, 1, 3, 256, 256]) from checkpoint, the shape in current model is torch.Size([256, 3, 1, 3, 256]).
size mismatch for cylinder_3d_spconv_seg.upBlock1.conv3.weight: copying a param with shape torch.Size([3, 3, 3, 256, 256]) from checkpoint, the shape in current model is torch.Size([256, 3, 3, 3, 256]).
size mismatch for cylinder_3d_spconv_seg.upBlock1.up_subm.weight: copying a param with shape torch.Size([3, 3, 3, 256, 256]) from checkpoint, the shape in current model is torch.Size([256, 3, 3, 3, 256]).
size mismatch for cylinder_3d_spconv_seg.upBlock2.trans_dilao.weight: copying a param with shape torch.Size([3, 3, 3, 256, 128]) from checkpoint, the shape in current model is torch.Size([128, 3, 3, 3, 256]).
size mismatch for cylinder_3d_spconv_seg.upBlock2.conv1.weight: copying a param with shape torch.Size([1, 3, 3, 128, 128]) from checkpoint, the shape in current model is torch.Size([128, 1, 3, 3, 128]).
size mismatch for cylinder_3d_spconv_seg.upBlock2.conv2.weight: copying a param with shape torch.Size([3, 1, 3, 128, 128]) from checkpoint, the shape in current model is torch.Size([128, 3, 1, 3, 128]).
size mismatch for cylinder_3d_spconv_seg.upBlock2.conv3.weight: copying a param with shape torch.Size([3, 3, 3, 128, 128]) from checkpoint, the shape in current model is torch.Size([128, 3, 3, 3, 128]).
size mismatch for cylinder_3d_spconv_seg.upBlock2.up_subm.weight: copying a param with shape torch.Size([3, 3, 3, 128, 128]) from checkpoint, the shape in current model is torch.Size([128, 3, 3, 3, 128]).
size mismatch for cylinder_3d_spconv_seg.upBlock3.trans_dilao.weight: copying a param with shape torch.Size([3, 3, 3, 128, 64]) from checkpoint, the shape in current model is torch.Size([64, 3, 3, 3, 128]).
size mismatch for cylinder_3d_spconv_seg.upBlock3.conv1.weight: copying a param with shape torch.Size([1, 3, 3, 64, 64]) from checkpoint, the shape in current model is torch.Size([64, 1, 3, 3, 64]).
size mismatch for cylinder_3d_spconv_seg.upBlock3.conv2.weight: copying a param with shape torch.Size([3, 1, 3, 64, 64]) from checkpoint, the shape in current model is torch.Size([64, 3, 1, 3, 64]).
size mismatch for cylinder_3d_spconv_seg.upBlock3.conv3.weight: copying a param with shape torch.Size([3, 3, 3, 64, 64]) from checkpoint, the shape in current model is torch.Size([64, 3, 3, 3, 64]).
size mismatch for cylinder_3d_spconv_seg.upBlock3.up_subm.weight: copying a param with shape torch.Size([3, 3, 3, 64, 64]) from checkpoint, the shape in current model is torch.Size([64, 3, 3, 3, 64]).
size mismatch for cylinder_3d_spconv_seg.ReconNet.conv1.weight: copying a param with shape torch.Size([3, 1, 1, 64, 64]) from checkpoint, the shape in current model is torch.Size([64, 3, 1, 1, 64]).
size mismatch for cylinder_3d_spconv_seg.ReconNet.conv1_2.weight: copying a param with shape torch.Size([1, 3, 1, 64, 64]) from checkpoint, the shape in current model is torch.Size([64, 1, 3, 1, 64]).
size mismatch for cylinder_3d_spconv_seg.ReconNet.conv1_3.weight: copying a param with shape torch.Size([1, 1, 3, 64, 64]) from checkpoint, the shape in current model is torch.Size([64, 1, 1, 3, 64]).
size mismatch for cylinder_3d_spconv_seg.logits.weight: copying a param with shape torch.Size([3, 3, 3, 128, 20]) from checkpoint, the shape in current model is torch.Size([20, 3, 3, 3, 128]).

[xinge008]

Different version of spconv makes the difference. It should be spconv=1.2.1 for Cylinder3D in this repo.

[L-Reichardt]

Spconv 1.x -> 2.x introduced breaking changes of the algorithm. Use old Spconv, or rewritten version here for spconv-2.x (~80% faster).