AttributeError: 'list' object has no attribute 'val'. When set flexible input shapes

[ThuyyTran]

🐞Describing the bug

Traceback (most recent call last):
  File "/media/anlab/data-2tb/ANLAB_THUY/ImageSearcher/ConvertSolar2Coreml.py", line 119, in <module>
    mlprogram = ct.convert(
  File "/home/anlab/anaconda3/envs/convertmodel1/lib/python3.9/site-packages/coremltools/converters/_converters_entry.py", line 551, in convert
    mlmodel = mil_convert(
  File "/home/anlab/anaconda3/envs/convertmodel1/lib/python3.9/site-packages/coremltools/converters/mil/converter.py", line 188, in mil_convert
    return _mil_convert(model, convert_from, convert_to, ConverterRegistry, MLModel, compute_units, **kwargs)
  File "/home/anlab/anaconda3/envs/convertmodel1/lib/python3.9/site-packages/coremltools/converters/mil/converter.py", line 212, in _mil_convert
    proto, mil_program = mil_convert_to_proto(
  File "/home/anlab/anaconda3/envs/convertmodel1/lib/python3.9/site-packages/coremltools/converters/mil/converter.py", line 286, in mil_convert_to_proto
    prog = frontend_converter(model, **kwargs)
  File "/home/anlab/anaconda3/envs/convertmodel1/lib/python3.9/site-packages/coremltools/converters/mil/converter.py", line 108, in __call__
    return load(*args, **kwargs)
  File "/home/anlab/anaconda3/envs/convertmodel1/lib/python3.9/site-packages/coremltools/converters/mil/frontend/torch/load.py", line 75, in load
    return _perform_torch_convert(converter, debug)
  File "/home/anlab/anaconda3/envs/convertmodel1/lib/python3.9/site-packages/coremltools/converters/mil/frontend/torch/load.py", line 114, in _perform_torch_convert
    prog = converter.convert()
  File "/home/anlab/anaconda3/envs/convertmodel1/lib/python3.9/site-packages/coremltools/converters/mil/frontend/torch/converter.py", line 484, in convert
    convert_nodes(self.context, self.graph)
  File "/home/anlab/anaconda3/envs/convertmodel1/lib/python3.9/site-packages/coremltools/converters/mil/frontend/torch/ops.py", line 93, in convert_nodes
    add_op(context, node)
  File "/home/anlab/anaconda3/envs/convertmodel1/lib/python3.9/site-packages/coremltools/converters/mil/frontend/torch/ops.py", line 3923, in avg_pool2d
    _avg_pool(context, node, inputs)
  File "/home/anlab/anaconda3/envs/convertmodel1/lib/python3.9/site-packages/coremltools/converters/mil/frontend/torch/ops.py", line 3876, in _avg_pool
    strides = mb.const(val=kernel_sizes.val, name=strides.name)
AttributeError: 'list' object has no attribute 'val'

To Reproduce

• I try convert model SOLAR- (https://github.com/tonyngjichun/SOLAR/tree/master)

Init model pytorch:

class Network(nn.Module):
    def __init__(self, model):
        super().__init__()
        self.model = model.cpu()
        self.mean = torch.tensor([0.485, 0.456, 0.406]).view(3, 1, 1)
        self.std = torch.tensor([0.229, 0.224, 0.225]).view(3, 1, 1)
    def forward(self,x):
        out1 = self.model(x)   
        reshaped_tensor1 = out1.view(1, 2048)
        return reshaped_tensor1
state = torch.load(os.path.join(get_data_root(), 'networks/model_best.pth.tar'),map_location=torch.device('cpu'))
net_params = {}
net_params['architecture'] = state['meta']['architecture']
net_params['pooling'] = state['meta']['pooling'] 
net_params['local_whitening'] = state['meta'].get('local_whitening', False)
net_params['regional'] = state['meta'].get('regional', False)
net_params['whitening'] = state['meta'].get('whitening', True)
net_params['mean'] = state['meta']['mean']
net_params['std'] = state['meta']['std']
net_params['pretrained'] = False
net = load_network('model_best.pth.tar')
net.load_state_dict(state['state_dict'])
net.eval()
test_model = Network(net)

Convert pytorch to coreml

scale = 1/(0.226*255.0)
bias = [- 0.485/(0.229) , - 0.456/(0.224), - 0.406/(0.225)]
input_shape = ct.Shape(shape=(1, 3, ct.RangeDim(lower_bound=100, upper_bound=800),
                              ct.RangeDim(lower_bound=100, upper_bound=800)))
dummy_input = torch.rand(1,3,300,300)
input_tensor = ct.ImageType(name="my_input", shape=input_shape,scale=scale, bias=bias)
traced_model = torch.jit.trace(test_model.eval(), dummy_input)
traced_model.eval()

mlprogram = ct.convert(
    traced_model,
    minimum_deployment_target=ct.target.iOS13,
    inputs=[input_tensor],
    outputs=[ct.TensorType(name="embeddings")],
    convert_to="neuralnetwork",
    compute_units=ct.ComputeUnit.CPU_ONLY,
)
saved_model = 'ModelConvert/TestModel/Solar300_image_CPU_FlexibleInput.mlmodel'
outputmodel.save(saved_model)

System environment

• coremltools version: 7.0
• OS (e.g. MacOS version or Linux type): Linux
• Any other relevant version information (e.g. PyTorch or TensorFlow version): Torch 1.9.1

[xorange]

Could you try #2050 to see if it fixes this problem ?
Or could you provide a standalone minimum example for reproduce ?

[ThuyyTran]

Could you try #2050 to see if it fixes this problem ? Or could you provide a standalone minimum example for reproduce ?

I tried to change the code in file coremltools/converters/mil/frontend/torch/ops.py similar to #2050. But when I convert the model, it always pauses at a certain threshold. I tried waiting but it didn't seem to work.
you can reproduce using my colab code here: https://colab.research.google.com/drive/1jqT_K3ValOyfc_o_2od44eiz4UiE8xAk?usp=sharing

[xorange]

Could you try #2050 to see if it fixes this problem ? Or could you provide a standalone minimum example for reproduce ?

I tried to change the code in file coremltools/converters/mil/frontend/torch/ops.py similar to #2050. But when I convert the model, it always pauses at a certain threshold. I tried waiting but it didn't seem to work. you can reproduce using my colab code here: https://colab.research.google.com/drive/1jqT_K3ValOyfc_o_2od44eiz4UiE8xAk?usp=sharing

OK. I also notice the pause and has a plan yet to be implemented. I'll try your code after the optimization.

[ThuyyTran]

@xorange Thanks. Looking forward to hearing from you.

[xorange]

My PR focused on "padding" and failed to fix this problem. I've encountered the same situation on another net structure with a similar root cause.

I'll try to generalize my fix for these.