Issues with weights for the model, missing keys and wrong size.

[margokhokhlova]

Hello! I am trying to use the pre-trained model to the inference. Unfortunately, I am failing to load the provided weights. I use this model: backend_name="resgcn-n39-r8" with parameters: model_args = { "A": torch.tensor(graph.A, dtype=torch.float32, requires_grad=False), "num_class": embedding_layer_size, "num_input": 3 if multi_branch else 1, "num_channel": 5 if multi_input else 3, "parts": graph.parts, } And weights: gaitgraph_resgcn-n39-r8_coco_seq_60.pth

However, the keys and sizes between the model and the saved dictionary do not match, so I cannot load from the checkpoint, and many keys are missing: load_state_dict raise RuntimeError('Error(s) in loading state_dict for {}:\n\t{}'.format( RuntimeError: Error(s) in loading state_dict for ResGCN: Missing key(s) in state_dict: "input_branches.1.A", "input_branches.1.bn.weight", "input_branches.1.bn.bias", "input_branches.1.bn.running_mean", "input_branches.1.bn.running_var", "input_branches.1.layers.0.edge", "input_branches.1.layers.0.scn.conv.gcn.weight", "input_branches.1.layers.0.scn.conv.gcn.bias", "input_branches.1.layers.0.scn.bn.weight", "input_branches.1.layers.0.scn.bn.bias", "input_branches.1.layers.0.scn.bn.running_mean", "input_branches.1.layers.0.scn.bn.running_var", "input_branches.1.layers.0.tcn.conv.weight", "input_branches.1.layers.0.tcn.conv.bias", "input_branches.1.layers.0.tcn.bn.weight", "input_branches.1.layers.0.tcn.bn.bias", "input_branches.1.layers.0.tcn.bn.running_mean", "input_branches.1.layers.0.tcn.bn.running_var", "input_branches.1.layers.1.edge", "input_branches.1.layers.1.scn.conv_down.weight", "input_branches.1.layers.1.scn.conv_down.bias", "input_branches.1.layers.1.scn.bn_down.weight", "input_branches.1.layers.1.scn.bn_down.bias", "input_branches.1.layers.1.scn.bn_down.running_mean", "input_branches.1.layers.1.scn.bn_down.running_var", "input_branches.1.layers.1.scn.conv.gcn.weight", "input_branches.1.layers.1.scn.conv.gcn.bias", "input_branches.1.layers.1.scn.bn.weight", "input_branches.1.layers.1.scn.bn.bias", "input_branches.1.layers.1.scn.bn.running_mean", "input_branches.1.layers.1.scn.bn.running_var", "input_branches.1.layers.1.scn.conv_up.weight", "input_branches.1.layers.1.scn.conv_up.bias", "input_branches.1.layers.1.scn.bn_up.weight", "input_branches.1.layers.1.scn.bn_up.bias", "input_branches.1.layers.1.scn.bn_up.running_mean", "input_branches.1.layers.1.scn.bn_up.running_var", "input_branches.1.layers.1.tcn.conv_down.weight", "input_branches.1.layers.1.tcn.conv_down.bias", "input_branches.1.layers.1.tcn.bn_down.weight", "input_branches.1.layers.1.tcn.bn_down.bias", "input_branches.1.layers.1.tcn.bn_down.running_mean", "input_branches.1.layers.1.tcn.bn_down.running_var", "input_branches.1.layers.1.tcn.conv.weight", "input_branches.1.layers.1.tcn.conv.bias", "input_branches.1.layers.1.tcn.bn.weight", "input_branches.1.layers.1.tcn.bn.bias", "input_branches.1.layers.1.tcn.bn.running_mean", "input_branches.1.layers.1.tcn.bn.running_var", "input_branches.1.layers.1.tcn.conv_up.weight", "input_branches.1.layers.1.tcn.conv_up.bias", "input_branches.1.layers.1.tcn.bn_up.weight", "input_branches.1.layers.1.tcn.bn_up.bias", "input_branches.1.layers.1.tcn.bn_up.running_mean", "input_branches.1.layers.1.tcn.bn_up.running_var", "input_branches.1.layers.2.edge", "input_branches.1.layers.2.scn.residual.0.weight", "input_branches.1.layers.2.scn.residual.0.bias", "input_branches.1.layers.2.scn.residual.1.weight", "input_branches.1.layers.2.scn.residual.1.bias", "input_branches.1.layers.2.scn.residual.1.running_mean", "input_branches.1.layers.2.scn.residual.1.running_var", "input_branches.1.layers.2.scn.conv_down.weight", "input_branches.1.layers.2.scn.conv_down.bias", "input_branches.1.layers.2.scn.bn_down.weight", "input_branches.1.layers.2.scn.bn_down.bias", "input_branches.1.layers.2.scn.bn_down.running_mean", "input_branches.1.layers.2.scn.bn_down.running_var", "input_branches.1.layers.2.scn.conv.gcn.weight", "input_branches.1.layers.2.scn.conv.gcn.bias", "input_branches.1.layers.2.scn.bn.weight", "input_branches.1.layers.2.scn.bn.bias", "input_branches.1.layers.2.scn.bn.running_mean", "input_branches.1.layers.2.scn.bn.running_var", "input_branches.1.layers.2.scn.conv_up.weight", "input_branches.1.layers.2.scn.conv_up.bias", "input_branches.1.layers.2.scn.bn_up.weight", "input_branches.1.layers.2.scn.bn_up.bias", "input_branches.1.layers.2.scn.bn_up.running_mean", "input_branches.1.layers.2.scn.bn_up.running_var", "input_branches.1.layers.2.tcn.conv_down.weight", "input_branches.1.layers.2.tcn.conv_down.bias", "input_branches.1.layers.2.tcn.bn_down.weight", "input_branches.1.layers.2.tcn.bn_down.bias", "input_branches.1.layers.2.tcn.bn_down.running_mean", "input_branches.1.layers.2.tcn.bn_down.running_var", "input_branches.1.layers.2.tcn.conv.weight", "input_branches.1.layers.2.tcn.conv.bias", "input_branches.1.layers.2.tcn.bn.weight", "input_branches.1.layers.2.tcn.bn.bias", "input_branches.1.layers.2.tcn.bn.running_mean", "input_branches.1.layers.2.tcn.bn.running_var", "input_branches.1.layers.2.tcn.conv_up.weight", "input_branches.1.layers.2.tcn.conv_up.bias", "input_branches.1.layers.2.tcn.bn_up.weight", "input_branches.1.layers.2.tcn.bn_up.bias", "input_branches.1.layers.2.tcn.bn_up.running_mean", "input_branches.1.layers.2.tcn.bn_up.running_var", "input_branches.2.A", "input_branches.2.bn.weight", "input_branches.2.bn.bias", "input_branches.2.bn.running_mean", "input_branches.2.bn.running_var", "input_branches.2.layers.0.edge", "input_branches.2.layers.0.scn.conv.gcn.weight", "input_branches.2.layers.0.scn.conv.gcn.bias", "input_branches.2.layers.0.scn.bn.weight", "input_branches.2.layers.0.scn.bn.bias", "input_branches.2.layers.0.scn.bn.running_mean", "input_branches.2.layers.0.scn.bn.running_var", "input_branches.2.layers.0.tcn.conv.weight", "input_branches.2.layers.0.tcn.conv.bias", "input_branches.2.layers.0.tcn.bn.weight", "input_branches.2.layers.0.tcn.bn.bias", "input_branches.2.layers.0.tcn.bn.running_mean", "input_branches.2.layers.0.tcn.bn.running_var", "input_branches.2.layers.1.edge", "input_branches.2.layers.1.scn.conv_down.weight", "input_branches.2.layers.1.scn.conv_down.bias", "input_branches.2.layers.1.scn.bn_down.weight", "input_branches.2.layers.1.scn.bn_down.bias", "input_branches.2.layers.1.scn.bn_down.running_mean", "input_branches.2.layers.1.scn.bn_down.running_var", "input_branches.2.layers.1.scn.conv.gcn.weight", "input_branches.2.layers.1.scn.conv.gcn.bias", "input_branches.2.layers.1.scn.bn.weight", "input_branches.2.layers.1.scn.bn.bias", "input_branches.2.layers.1.scn.bn.running_mean", "input_branches.2.layers.1.scn.bn.running_var", "input_branches.2.layers.1.scn.conv_up.weight", "input_branches.2.layers.1.scn.conv_up.bias", "input_branches.2.layers.1.scn.bn_up.weight", "input_branches.2.layers.1.scn.bn_up.bias", "input_branches.2.layers.1.scn.bn_up.running_mean", "input_branches.2.layers.1.scn.bn_up.running_var", "input_branches.2.layers.1.tcn.conv_down.weight", "input_branches.2.layers.1.tcn.conv_down.bias", "input_branches.2.layers.1.tcn.bn_down.weight", "input_branches.2.layers.1.tcn.bn_down.bias", "input_branches.2.layers.1.tcn.bn_down.running_mean", "input_branches.2.layers.1.tcn.bn_down.running_var", "input_branches.2.layers.1.tcn.conv.weight", "input_branches.2.layers.1.tcn.conv.bias", "input_branches.2.layers.1.tcn.bn.weight", "input_branches.2.layers.1.tcn.bn.bias", "input_branches.2.layers.1.tcn.bn.running_mean", "input_branches.2.layers.1.tcn.bn.running_var", "input_branches.2.layers.1.tcn.conv_up.weight", "input_branches.2.layers.1.tcn.conv_up.bias", "input_branches.2.layers.1.tcn.bn_up.weight", "input_branches.2.layers.1.tcn.bn_up.bias", "input_branches.2.layers.1.tcn.bn_up.running_mean", "input_branches.2.layers.1.tcn.bn_up.running_var", "input_branches.2.layers.2.edge", "input_branches.2.layers.2.scn.residual.0.weight", "input_branches.2.layers.2.scn.residual.0.bias", "input_branches.2.layers.2.scn.residual.1.weight", "input_branches.2.layers.2.scn.residual.1.bias", "input_branches.2.layers.2.scn.residual.1.running_mean", "input_branches.2.layers.2.scn.residual.1.running_var", "input_branches.2.layers.2.scn.conv_down.weight", "input_branches.2.layers.2.scn.conv_down.bias", "input_branches.2.layers.2.scn.bn_down.weight", "input_branches.2.layers.2.scn.bn_down.bias", "input_branches.2.layers.2.scn.bn_down.running_mean", "input_branches.2.layers.2.scn.bn_down.running_var", "input_branches.2.layers.2.scn.conv.gcn.weight", "input_branches.2.layers.2.scn.conv.gcn.bias", "input_branches.2.layers.2.scn.bn.weight", "input_branches.2.layers.2.scn.bn.bias", "input_branches.2.layers.2.scn.bn.running_mean", "input_branches.2.layers.2.scn.bn.running_var", "input_branches.2.layers.2.scn.conv_up.weight", "input_branches.2.layers.2.scn.conv_up.bias", "input_branches.2.layers.2.scn.bn_up.weight", "input_branches.2.layers.2.scn.bn_up.bias", "input_branches.2.layers.2.scn.bn_up.running_mean", "input_branches.2.layers.2.scn.bn_up.running_var", "input_branches.2.layers.2.tcn.conv_down.weight", "input_branches.2.layers.2.tcn.conv_down.bias", "input_branches.2.layers.2.tcn.bn_down.weight", "input_branches.2.layers.2.tcn.bn_down.bias", "input_branches.2.layers.2.tcn.bn_down.running_mean", "input_branches.2.layers.2.tcn.bn_down.running_var", "input_branches.2.layers.2.tcn.conv.weight", "input_branches.2.layers.2.tcn.conv.bias", "input_branches.2.layers.2.tcn.bn.weight", "input_branches.2.layers.2.tcn.bn.bias", "input_branches.2.layers.2.tcn.bn.running_mean", "input_branches.2.layers.2.tcn.bn.running_var", "input_branches.2.layers.2.tcn.conv_up.weight", "input_branches.2.layers.2.tcn.conv_up.bias", "input_branches.2.layers.2.tcn.bn_up.weight", "input_branches.2.layers.2.tcn.bn_up.bias", "input_branches.2.layers.2.tcn.bn_up.running_mean", "input_branches.2.layers.2.tcn.bn_up.running_var". size mismatch for input_branches.0.bn.weight: copying a param with shape torch.Size([3]) from checkpoint, the shape in current model is torch.Size([5]). size mismatch for input_branches.0.bn.bias: copying a param with shape torch.Size([3]) from checkpoint, the shape in current model is torch.Size([5]). size mismatch for input_branches.0.bn.running_mean: copying a param with shape torch.Size([3]) from checkpoint, the shape in current model is torch.Size([5]). size mismatch for input_branches.0.bn.running_var: copying a param with shape torch.Size([3]) from checkpoint, the shape in current model is torch.Size([5]). size mismatch for input_branches.0.layers.0.scn.conv.gcn.weight: copying a param with shape torch.Size([192, 3, 1, 1]) from checkpoint, the shape in current model is torch.Size([192, 5, 1, 1]). size mismatch for main_stream.0.scn.residual.0.weight: copying a param with shape torch.Size([128, 32, 1, 1]) from checkpoint, the shape in current model is torch.Size([128, 96, 1, 1]). size mismatch for main_stream.0.scn.conv_down.weight: copying a param with shape torch.Size([16, 32, 1, 1]) from checkpoint, the shape in current model is torch.Size([16, 96, 1, 1]).

Please, can you advice on where can I find the weights & model which correspond to each other? Thank you!