Pre-training file and network dimension mismatch

[200203Z]

When I load the pre-training file (pretrained_ckpts/vo/act_forward.pth) to the model @BaselineRegistry.register_vo_model(name="vo_cnn_rgb_d_dd_top_down") Class VisualOdometryCNNDiscretizedDepthTopDownView(VisualOdometryCNNBase) reports an error message about dimension mismatch: size mismatch for visual_encoder.compression.0.weight: copying a param with shape torch.Size([31, 256, 3, 3]) from checkpoint, the shape in current model is torch.Size([32, 256, 3, 3]). size mismatch for visual_encoder.compression.1.weight: copying a param with shape torch.Size([31]) from checkpoint, the shape in current model is torch.Size([32]). size mismatch for visual_encoder.compression.1.bias: copying a param with shape torch.Size([31]) from checkpoint, the shape in current model is torch.Size([32]). size mismatch for visual_fc.2.weight: copying a param with shape torch.Size([512, 2046]) from checkpoint, the shape in current model is torch.Size([512, 2048]) Why does the mismatch problem occur, is the network model loading error, and which network should correspond to this pre-training file correctly?