loading pretrained-model show error

[ZhaoWenjun123]

I downloaded your pretrained model, but I got an error while loading
root@dl-0835182853-pod-jupyter-67cc9db95c-4gq8q:~/monocon_modifyV2-master/experiments/example# python ../../tools/train_val.py --config kitti_example.yaml --e 2022-05-20 20:48:28,228 INFO ################### Evaluation Only ################## 2022-05-20 20:48:34,673 INFO ==> Loading from checkpoint '/root/monocon_modifyV2-master/checkpoints/checkpoint_epoch_140.pth' Traceback (most recent call last): File "../../tools/train_val.py", line 81, in <module> main() File "../../tools/train_val.py", line 50, in main tester.test() File "/root/monocon_modifyV2-master/lib/helpers/tester_helper.py", line 36, in test logger=self.logger) File "/root/monocon_modifyV2-master/lib/helpers/save_helper.py", line 37, in load_checkpoint model.load_state_dict(checkpoint['model_state']) File "/root/.local/lib/python3.7/site-packages/torch/nn/modules/module.py", line 1052, in load_state_dict self.__class__.__name__, "\n\t".join(error_msgs))) RuntimeError: Error(s) in loading state_dict for MonoCon: Missing key(s) in state_dict: "center2kpt_offset.0.weight", "center2kpt_offset.0.bias", "center2kpt_offset.2.weight", "center2kpt_offset.2.bias", "kpt_heatmap.0.weight", "kpt_heatmap.0.bias", "kpt_heatmap.2.weight", "kpt_heatmap.2.bias", "kpt_heatmap_offset.0.weight", "kpt_heatmap_offset.0.bias", "kpt_heatmap_offset.2.weight", "kpt_heatmap_offset.2.bias". size mismatch for heatmap.0.weight: copying a param with shape torch.Size([256, 64, 3, 3]) from checkpoint, the shape in current model is torch.Size([64, 64, 3, 3]). size mismatch for heatmap.0.bias: copying a param with shape torch.Size([256]) from checkpoint, the shape in current model is torch.Size([64]). size mismatch for heatmap.2.weight: copying a param with shape torch.Size([3, 256, 1, 1]) from checkpoint, the shape in current model is torch.Size([3, 64, 1, 1]). size mismatch for offset_2d.0.weight: copying a param with shape torch.Size([256, 64, 3, 3]) from checkpoint, the shape in current model is torch.Size([64, 64, 3, 3]). size mismatch for offset_2d.0.bias: copying a param with shape torch.Size([256]) from checkpoint, the shape in current model is torch.Size([64]). size mismatch for offset_2d.2.weight: copying a param with shape torch.Size([2, 256, 1, 1]) from checkpoint, the shape in current model is torch.Size([2, 64, 1, 1]). size mismatch for size_2d.0.weight: copying a param with shape torch.Size([256, 64, 3, 3]) from checkpoint, the shape in current model is torch.Size([64, 64, 3, 3]). size mismatch for size_2d.0.bias: copying a param with shape torch.Size([256]) from checkpoint, the shape in current model is torch.Size([64]). size mismatch for size_2d.2.weight: copying a param with shape torch.Size([2, 256, 1, 1]) from checkpoint, the shape in current model is torch.Size([2, 64, 1, 1]). size mismatch for depth.0.weight: copying a param with shape torch.Size([256, 64, 3, 3]) from checkpoint, the shape in current model is torch.Size([64, 64, 3, 3]). size mismatch for depth.0.bias: copying a param with shape torch.Size([256]) from checkpoint, the shape in current model is torch.Size([64]). size mismatch for depth.2.weight: copying a param with shape torch.Size([2, 256, 1, 1]) from checkpoint, the shape in current model is torch.Size([2, 64, 1, 1]). size mismatch for offset_3d.0.weight: copying a param with shape torch.Size([256, 64, 3, 3]) from checkpoint, the shape in current model is torch.Size([64, 64, 3, 3]). size mismatch for offset_3d.0.bias: copying a param with shape torch.Size([256]) from checkpoint, the shape in current model is torch.Size([64]). size mismatch for offset_3d.2.weight: copying a param with shape torch.Size([2, 256, 1, 1]) from checkpoint, the shape in current model is torch.Size([2, 64, 1, 1]). size mismatch for size_3d.0.weight: copying a param with shape torch.Size([256, 64, 3, 3]) from checkpoint, the shape in current model is torch.Size([64, 64, 3, 3]). size mismatch for size_3d.0.bias: copying a param with shape torch.Size([256]) from checkpoint, the shape in current model is torch.Size([64]). size mismatch for size_3d.2.weight: copying a param with shape torch.Size([3, 256, 1, 1]) from checkpoint, the shape in current model is torch.Size([3, 64, 1, 1]). size mismatch for heading.0.weight: copying a param with shape torch.Size([256, 64, 3, 3]) from checkpoint, the shape in current model is torch.Size([64, 64, 3, 3]). size mismatch for heading.0.bias: copying a param with shape torch.Size([256]) from checkpoint, the shape in current model is torch.Size([64]). size mismatch for heading.2.weight: copying a param with shape torch.Size([24, 256, 1, 1]) from checkpoint, the shape in current model is torch.Size([24, 64, 1, 1]).

[ZhaoWenjun123]

This is the performance after my training, far from the performance in the author's paper
Car [email protected], 0.70, 0.70:
bbox AP:96.3709, 92.1202, 84.8560
bev AP:22.6310, 18.1839, 15.9116
3d AP:15.8479, 12.4094, 11.1673
aos AP:95.91, 91.43, 83.64

[Senwang98]

@ZhaoWenjun123
hello, have you get the reason?
I found that the head's channel in this repo is 64, while the monodle's channel is 256.

I got the following results.

Car [email protected], 0.70, 0.70:
bbox AP:98.7093, 92.2566, 84.9337
bev  AP:24.9916, 19.0743, 16.5050
3d   AP:17.8765, 13.6974, 11.7378

The checkpoint provided by this repo's author has the following performance:

2023-01-29 22:57:20,035   INFO  Car [email protected], 0.70, 0.70:
bbox AP:90.1249, 88.3711, 79.8854
bev  AP:31.2481, 24.7352, 23.4494
3d   AP:23.7383, 20.7088, 17.9975
aos  AP:89.09, 87.18, 78.03
Car [email protected], 0.70, 0.70:
bbox AP:95.9643, 91.8798, 84.7518
bev  AP:25.8329, 20.7988, 18.1310
3d   AP:18.2511, 14.5574, 12.9762
aos  AP:94.81, 90.54, 82.53
Car [email protected], 0.50, 0.50:
bbox AP:90.1249, 88.3711, 79.8854
bev  AP:61.6220, 50.2440, 44.6793
3d   AP:57.7742, 44.3630, 42.4209
aos  AP:89.09, 87.18, 78.03
Car [email protected], 0.50, 0.50:
bbox AP:95.9643, 91.8798, 84.7518
bev  AP:61.4155, 47.3504, 41.9609
3d   AP:56.0406, 42.8349, 38.6653
aos  AP:94.81, 90.54, 82.53