Multiple Input Problems in Non Classified Models

[sususama]

As described in the title, my network is a generative network with two inputs. When I run SmoothGradCAMpp, it will report an error and missing another necessary parameter.

[sususama]

I fused multiple inputs into one, but still reported this error

 File "/workspace/128_MoreLoss_0531/main.py", line 572, in train
    with SmoothGradCAMpp(G_G, input_shape=[3, 128, 128]) as cam_extractor:
  File "/opt/conda/lib/python3.8/site-packages/torchcam/methods/gradient.py", line 239, in __init__
    super().__init__(model, target_layer, input_shape, **kwargs)
  File "/opt/conda/lib/python3.8/site-packages/torchcam/methods/gradient.py", line 34, in __init__
    super().__init__(model, target_layer, input_shape, **kwargs)
  File "/opt/conda/lib/python3.8/site-packages/torchcam/methods/core.py", line 53, in __init__
    target_name = locate_candidate_layer(model, input_shape)
  File "/opt/conda/lib/python3.8/site-packages/torchcam/methods/_utils.py", line 43, in locate_candidate_layer
    _ = mod(torch.zeros((1, *input_shape), device=next(mod.parameters()).data.device))
  File "/opt/conda/lib/python3.8/site-packages/torch/nn/modules/module.py", line 1071, in _call_impl
    result = forward_call(*input, **kwargs)
  File "/workspace/128_MoreLoss_0531/model.py", line 783, in forward
    vggm_img = self.model_mirror(vgg_img)
  File "/opt/conda/lib/python3.8/site-packages/torch/nn/modules/module.py", line 1076, in _call_impl
    hook_result = hook(self, input, result)
  File "/opt/conda/lib/python3.8/site-packages/torchcam/methods/_utils.py", line 34, in _record_output_shape
    output_shapes.append((name, output.shape))
AttributeError: 'list' object has no attribute 'shape'

[frgfm]

Hi @sususama 👋

Thanks for reporting this but without a runnable snippet I cannot reproduce that bug and really help you (I can only try to guess). Would you mind sharing a minimal snippet to reproduce this?

Now circling back to my guess, you have a custom architecture that takes things that aren't tensors as input. I suggest checking that thread #210

Let me know if that helps!