[PyTorch] simplepose_resnet18_coco model weights loading error

[lrzpellegrini]

The simplepose_resnet18_coco pretrained weights cannot be loaded using pytorchcv.

How to reproduce

1. Create and activate a new environment

conda create -n avl_simplepose python=3.9
conda activate avl_simplepose

2. Install pytorchcv and torch packages (pip). Note: the exact torch install command may vary. I used the one from the official site for CUDA 11.6

pip install pytorchcv
pip3 install torch torchvision torchaudio --extra-index-url https://download.pytorch.org/whl/cu116

3. Make sure no .torch folder exists in the home directory (may not be needed to reproduce the issue).
4. Run the following commands (+log):

Python 3.9.13 | packaged by conda-forge | (main, May 27 2022, 16:58:50) 
[GCC 10.3.0] on linux
Type "help", "copyright", "credits" or "license" for more information.
>>> import torch
>>> from pytorchcv.model_provider import get_model as ptcv_get_model
>>> 
>>> ptcv_get_model("simplepose_resnet18_coco", pretrained=True)
Downloading /home/lorenzo/.torch/models/simplepose_resnet18_coco-6631-7c3656b3.pth.zip from https://github.com/osmr/imgclsmob/releases/download/v0.0.455/simplepose_resnet18_coco-6631-7c3656b3.pth.zip...

Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "/home/lorenzo/miniconda3/envs/prova_avl_simplepose/lib/python3.9/site-packages/pytorchcv/model_provider.py", line 1233, in get_model
    net = _models[name](**kwargs)
  File "/home/lorenzo/miniconda3/envs/prova_avl_simplepose/lib/python3.9/site-packages/pytorchcv/models/simplepose_coco.py", line 155, in simplepose_resnet18_coco
    return get_simplepose(backbone=backbone, backbone_out_channels=512, keypoints=keypoints,
  File "/home/lorenzo/miniconda3/envs/prova_avl_simplepose/lib/python3.9/site-packages/pytorchcv/models/simplepose_coco.py", line 129, in get_simplepose
    download_model(
  File "/home/lorenzo/miniconda3/envs/prova_avl_simplepose/lib/python3.9/site-packages/pytorchcv/models/model_store.py", line 827, in download_model
    load_model(
  File "/home/lorenzo/miniconda3/envs/prova_avl_simplepose/lib/python3.9/site-packages/pytorchcv/models/model_store.py", line 804, in load_model
    net.load_state_dict(pretrained_state)
  File "/home/lorenzo/miniconda3/envs/prova_avl_simplepose/lib/python3.9/site-packages/torch/nn/modules/module.py", line 1667, in load_state_dict
    raise RuntimeError('Error(s) in loading state_dict for {}:\n\t{}'.format(
RuntimeError: Error(s) in loading state_dict for SimplePose:
        Missing key(s) in state_dict: "backbone.0.conv.conv.weight", "backbone.0.conv.bn.weight", "backbone.0.conv.bn.bias", "backbone.0.conv.bn.running_mean", "backbone.0.conv.bn.running_var", "backbone.1.unit1.body.conv1.conv.weight", "backbone.1.unit1.body.conv1.bn.weight", "backbone.1.unit1.body.conv1.bn.bias", "backbone.1.unit1.body.conv1.bn.running_mean", "backbone.1.unit1.body.conv1.bn.running_var", "backbone.1.unit1.body.conv2.conv.weight", "backbone.1.unit1.body.conv2.bn.weight", "backbone.1.unit1.body.conv2.bn.bias", "backbone.1.unit1.body.conv2.bn.running_mean", "backbone.1.unit1.body.conv2.bn.running_var", "backbone.1.unit2.body.conv1.conv.weight", "backbone.1.unit2.body.conv1.bn.weight", "backbone.1.unit2.body.conv1.bn.bias", "backbone.1.unit2.body.conv1.bn.running_mean", "backbone.1.unit2.body.conv1.bn.running_var", "backbone.1.unit2.body.conv2.conv.weight", "backbone.1.unit2.body.conv2.bn.weight", "backbone.1.unit2.body.conv2.bn.bias", "backbone.1.unit2.body.conv2.bn.running_mean", "backbone.1.unit2.body.conv2.bn.running_var", "backbone.2.unit1.body.conv1.conv.weight", "backbone.2.unit1.body.conv1.bn.weight", "backbone.2.unit1.body.conv1.bn.bias", "backbone.2.unit1.body.conv1.bn.running_mean", "backbone.2.unit1.body.conv1.bn.running_var", "backbone.2.unit1.body.conv2.conv.weight", "backbone.2.unit1.body.conv2.bn.weight", "backbone.2.unit1.body.conv2.bn.bias", "backbone.2.unit1.body.conv2.bn.running_mean", "backbone.2.unit1.body.conv2.bn.running_var", "backbone.2.unit1.identity_conv.conv.weight", "backbone.2.unit1.identity_conv.bn.weight", "backbone.2.unit1.identity_conv.bn.bias", "backbone.2.unit1.identity_conv.bn.running_mean", "backbone.2.unit1.identity_conv.bn.running_var", "backbone.2.unit2.body.conv1.conv.weight", "backbone.2.unit2.body.conv1.bn.weight", "backbone.2.unit2.body.conv1.bn.bias", "backbone.2.unit2.body.conv1.bn.running_mean", "backbone.2.unit2.body.conv1.bn.running_var", "backbone.2.unit2.body.conv2.conv.weight", "backbone.2.unit2.body.conv2.bn.weight", "backbone.2.unit2.body.conv2.bn.bias", "backbone.2.unit2.body.conv2.bn.running_mean", "backbone.2.unit2.body.conv2.bn.running_var", "backbone.3.unit1.body.conv1.conv.weight", "backbone.3.unit1.body.conv1.bn.weight", "backbone.3.unit1.body.conv1.bn.bias", "backbone.3.unit1.body.conv1.bn.running_mean", "backbone.3.unit1.body.conv1.bn.running_var", "backbone.3.unit1.body.conv2.conv.weight", "backbone.3.unit1.body.conv2.bn.weight", "backbone.3.unit1.body.conv2.bn.bias", "backbone.3.unit1.body.conv2.bn.running_mean", "backbone.3.unit1.body.conv2.bn.running_var", "backbone.3.unit1.identity_conv.conv.weight", "backbone.3.unit1.identity_conv.bn.weight", "backbone.3.unit1.identity_conv.bn.bias", "backbone.3.unit1.identity_conv.bn.running_mean", "backbone.3.unit1.identity_conv.bn.running_var", "backbone.3.unit2.body.conv1.conv.weight", "backbone.3.unit2.body.conv1.bn.weight", "backbone.3.unit2.body.conv1.bn.bias", "backbone.3.unit2.body.conv1.bn.running_mean", "backbone.3.unit2.body.conv1.bn.running_var", "backbone.3.unit2.body.conv2.conv.weight", "backbone.3.unit2.body.conv2.bn.weight", "backbone.3.unit2.body.conv2.bn.bias", "backbone.3.unit2.body.conv2.bn.running_mean", "backbone.3.unit2.body.conv2.bn.running_var", "backbone.4.unit1.body.conv1.conv.weight", "backbone.4.unit1.body.conv1.bn.weight", "backbone.4.unit1.body.conv1.bn.bias", "backbone.4.unit1.body.conv1.bn.running_mean", "backbone.4.unit1.body.conv1.bn.running_var", "backbone.4.unit1.body.conv2.conv.weight", "backbone.4.unit1.body.conv2.bn.weight", "backbone.4.unit1.body.conv2.bn.bias", "backbone.4.unit1.body.conv2.bn.running_mean", "backbone.4.unit1.body.conv2.bn.running_var", "backbone.4.unit1.identity_conv.conv.weight", "backbone.4.unit1.identity_conv.bn.weight", "backbone.4.unit1.identity_conv.bn.bias", "backbone.4.unit1.identity_conv.bn.running_mean", "backbone.4.unit1.identity_conv.bn.running_var", "backbone.4.unit2.body.conv1.conv.weight", "backbone.4.unit2.body.conv1.bn.weight", "backbone.4.unit2.body.conv1.bn.bias", "backbone.4.unit2.body.conv1.bn.running_mean", "backbone.4.unit2.body.conv1.bn.running_var", "backbone.4.unit2.body.conv2.conv.weight", "backbone.4.unit2.body.conv2.bn.weight", "backbone.4.unit2.body.conv2.bn.bias", "backbone.4.unit2.body.conv2.bn.running_mean", "backbone.4.unit2.body.conv2.bn.running_var".