torchvision export model error:: torchvision.models.detection.retinanet_resnet50_fpn_v2

[wangqianscu]

🐛 Describe the bug

When I use torchvision.models.detection.retinanet_resnet50_fpn() to generate the model, it raise error.

My code:

    input_=torch.ones(3,300,400)
    input_1 = torch.ones(3,500,400)
    model=torchvision.models.detection.retinanet_resnet50_fpn()
    print(model([input_, input_1]))  # all outputs are empty list '[]'
    model.cpu()
    model.eval()
    traced_model = torch.jit.trace(model, input_data)
    torch.jit.save(traced_model, model_file)

Error log:

[{'boxes': tensor([], size=(0, 4), grad_fn=<StackBackward0>), 'scores': tensor([], grad_fn=<IndexBackward0>), 'labels': tensor([], dtype=torch.int64)}, {'boxes': tensor([], size=(0, 4), grad_fn=<StackBackward0>), 'scores': tensor([], grad_fn=<IndexBackward0>), 'labels': tensor([], dtype=torch.int64)}]
-------------------
Traceback (most recent call last):
  File "/home/wangqian/torch_model_file_gen/run.py", line 43, in <module>
    test_retinanet_resnet50_fpn_v2(weights=sys.argv[1])
  File "/home/wangqian/torch_model_file_gen/run.py", line 41, in test_retinanet_resnet50_fpn_v2
    convert_model(model, weights, [input_, input_1], model_file)
  File "/home/wangqian/torch_model_file_gen/run.py", line 22, in convert_model
    traced_model = torch.jit.trace(model, input_data)
  File "/usr/local/lib/python3.10/dist-packages/torch/jit/_trace.py", line 806, in trace
    return trace_module(
  File "/usr/local/lib/python3.10/dist-packages/torch/jit/_trace.py", line 1074, in trace_module
    module._c._create_method_from_trace(
  File "/usr/local/lib/python3.10/dist-packages/torch/nn/modules/module.py", line 1511, in _wrapped_call_impl
    return self._call_impl(*args, **kwargs)
  File "/usr/local/lib/python3.10/dist-packages/torch/nn/modules/module.py", line 1520, in _call_impl
    return forward_call(*args, **kwargs)
  File "/usr/local/lib/python3.10/dist-packages/torch/nn/modules/module.py", line 1501, in _slow_forward
    result = self.forward(*input, **kwargs)
  File "/usr/local/lib/python3.10/dist-packages/torchvision/models/detection/retinanet.py", line 606, in forward
    images, targets = self.transform(images, targets)
  File "/usr/local/lib/python3.10/dist-packages/torch/nn/modules/module.py", line 1511, in _wrapped_call_impl
    return self._call_impl(*args, **kwargs)
  File "/usr/local/lib/python3.10/dist-packages/torch/nn/modules/module.py", line 1520, in _call_impl
    return forward_call(*args, **kwargs)
  File "/usr/local/lib/python3.10/dist-packages/torch/nn/modules/module.py", line 1501, in _slow_forward
    result = self.forward(*input, **kwargs)
  File "/usr/local/lib/python3.10/dist-packages/torchvision/models/detection/transform.py", line 131, in forward
    for k, v in t.items():
AttributeError: 'Tensor' object has no attribute 'items'. Did you mean: 'item'?

Versions

torch 2.2.1+cpu
torchaudio 2.2.1+cpu
torchvision 0.17.1+cpu

cc @EikanWang @jgong5 @wenzhe-nrv @sanchitintel