issue deploying libra-retinanet to onnx due to BFP neck component

[reiffd7]

Checklist

• I have searched related issues but cannot get the expected help.
• 2. I have read the FAQ documentation but cannot get the expected help.
• 3. The bug has not been fixed in the latest version.

Describe the bug

Can't deploy libra-retinanet. Tested with and without the BFP neck component and it works without this component. Wondering if there is a pytorch or onnx versioning problem here leading to not being able to handle this component. Anyone encounter this?

Reproduction

I ran the following mmdeploy command:

python3 ${MMDEPLOY_PATH}/tools/deploy.py \
    ${MMDEPLOY_PATH}/configs/mmdet/detection/detection_tensorrt-int8_static-544x960.py \
    ${ROOT_PATH}/model_data/person_detection/${MODEL_CHECKPOINT_NAME}/configs/libra-retinanet_r50_fpn_1x_coco.py \
    ${ROOT_PATH}/model_data/person_detection/${MODEL_CHECKPOINT_NAME}/models/pytorch/bfp_cross_entropy_epoch_44.pth \
    ${INPUT_IMG} \
    --work-dir ${WORK_DIR} \
    --device cuda \
    --log-level ERROR \
    --dump-info

The model config looks like this:

model = dict(
    type='RetinaNet',
    data_preprocessor=dict(
        type='DetDataPreprocessor',
        mean=[103.53, 116.28, 123.675],
        std=[57.375, 57.12, 58.395],
        bgr_to_rgb=False,
        pad_size_divisor=32),
    backbone=dict(
        type='RegNet',
        arch='regnetx_400mf',
        out_indices=(0, 1, 2, 3),
        frozen_stages=1,
        norm_cfg=dict(type='BN', requires_grad=True),
        norm_eval=True,
        style='pytorch',
        init_cfg=dict(
            type='Pretrained', checkpoint='open-mmlab://regnetx_400mf')
    ),
    neck=[
        dict(
            type='FPN',
            in_channels=[32, 64, 160, 384],
            out_channels=256,
            start_level=1,
            add_extra_convs='on_input',
            num_outs=5),
        dict(
            type='BFP',
            in_channels=256,
            num_levels=5,
            refine_level=1,
            refine_type='non_local')
    ],
    bbox_head=dict(
        type='ATSSHead',
        num_classes=1,
        in_channels=256,
        stacked_convs=4,
        feat_channels=256,
        norm_cfg=None,
        anchor_generator=dict(
            type='AnchorGenerator',
            ratios=[1.0],
            octave_base_scale=8,
            scales_per_octave=1,
            strides=[8, 16, 32, 64, 128]),
        bbox_coder=dict(
            type='DeltaXYWHBBoxCoder',
            target_means=[0.0, 0.0, 0.0, 0.0],
            target_stds=[0.1, 0.1, 0.2, 0.2]),
        loss_cls=dict(
            type='CrossEntropyLoss', use_sigmoid=True, loss_weight=1.0),
        loss_bbox=dict(type='GIoULoss', loss_weight=2.0),
        loss_centerness=dict(
            type='CrossEntropyLoss', use_sigmoid=True, loss_weight=1.0)  ),
    # model training and testing settings
    train_cfg=dict(
        assigner=dict(type='ATSSAssigner', topk=9),
        allowed_border=-1,
        pos_weight=-1,
        debug=False),
    test_cfg=dict(
        nms_pre=1000,
        min_bbox_size=16,
        score_thr=0.25,
        nms=dict(type='nms', iou_threshold=0.5),
        max_per_img=300))

Deployment runs fine without the BFP neck.

Environment

FROM openmmlab/mmdeploy:ubuntu20.04-cuda11.3-mmdeploy1.1.0
#FROM openmmlab/mmdeploy:ubuntu20.04-cuda11.8-mmdeploy

WORKDIR /root/workspace

RUN pip install joblib
RUN pip install torch==1.12.1+cu113 torchvision==0.13.1+cu113 torchaudio==0.12.1 --extra-index-url https://download.pytorch.org/whl/cu113

RUN pip install -U pip
RUN pip install joblib
RUN pip install openmim

RUN pip install fire onnx onnxsim sclblonnx
RUN mim install mmcv==2.1
RUN mim install mmrazor
RUN mim install mmdet>3
RUN mim install mmdeploy==1.3.0

RUN apt-get clean && rm -rf /var/lib/apt/lists/*

### Error traceback

```Shell
Convert PyTorch model to ONNX graph...
05/09 18:44:21 - mmengine - ERROR - /root/workspace/mmdeploy/mmdeploy/apis/core/pipeline_manager.py - __call__ - 94 - Start pipeline mmdeploy.apis.pytorch2onnx.torch2onnx in subprocess
05/09 18:44:22 - mmengine - WARNING - Failed to search registry with scope "mmdet" in the "Codebases" registry tree. As a workaround, the current "Codebases" registry in "mmdeploy" is used to build instance. This may cause unexpected failure when running the built modules. Please check whether "mmdet" is a correct scope, or whether the registry is initialized.
05/09 18:44:22 - mmengine - WARNING - Failed to search registry with scope "mmdet" in the "mmdet_tasks" registry tree. As a workaround, the current "mmdet_tasks" registry in "mmdeploy" is used to build instance. This may cause unexpected failure when running the built modules. Please check whether "mmdet" is a correct scope, or whether the registry is initialized.
/usr/local/lib/python3.8/dist-packages/mmdet/models/dense_heads/anchor_head.py:108: UserWarning: DeprecationWarning: `num_anchors` is deprecated, for consistency or also use `num_base_priors` instead
  warnings.warn('DeprecationWarning: `num_anchors` is deprecated, '
Loads checkpoint by local backend from path: /opt/ml/forsight-model-optimization/model_data/person_detection/retinanet_regnetx_400mf_fpn_1x8_1x_person_detection_19_04_2024_subset/models/pytorch/bfp_cross_entropy_epoch_44.pth
05/09 18:44:25 - mmengine - WARNING - DeprecationWarning: get_onnx_config will be deprecated in the future. 
05/09 18:44:25 - mmengine - INFO - Export PyTorch model to ONNX: /opt/ml/forsight-model-optimization/model_data/person_detection/retinanet_regnetx_400mf_fpn_1x8_1x_person_detection_19_04_2024_subset/models/onnx/end2end.onnx.
05/09 18:44:25 - mmengine - WARNING - Can not find torch._C._jit_pass_onnx_autograd_function_process, function rewrite will not be applied
05/09 18:44:25 - mmengine - WARNING - Can not find mmdet.models.utils.transformer.PatchMerging.forward, function rewrite will not be applied
/root/workspace/mmdeploy/mmdeploy/codebase/mmdet/models/detectors/single_stage.py:84: TracerWarning: Iterating over a tensor might cause the trace to be incorrect. Passing a tensor of different shape won't change the number of iterations executed (and might lead to errors or silently give incorrect results).
  img_shape = [int(val) for val in img_shape]
/root/workspace/mmdeploy/mmdeploy/codebase/mmdet/models/detectors/single_stage.py:84: TracerWarning: Converting a tensor to a Python integer might cause the trace to be incorrect. We can't record the data flow of Python values, so this value will be treated as a constant in the future. This means that the trace might not generalize to other inputs!
  img_shape = [int(val) for val in img_shape]
/root/workspace/mmdeploy/mmdeploy/core/optimizers/function_marker.py:160: TracerWarning: Converting a tensor to a Python integer might cause the trace to be incorrect. We can't record the data flow of Python values, so this value will be treated as a constant in the future. This means that the trace might not generalize to other inputs!
  ys_shape = tuple(int(s) for s in ys.shape)
/root/workspace/mmdeploy/mmdeploy/codebase/mmdet/models/dense_heads/base_dense_head.py:109: TracerWarning: Converting a tensor to a Python boolean might cause the trace to be incorrect. We can't record the data flow of Python values, so this value will be treated as a constant in the future. This means that the trace might not generalize to other inputs!
  assert cls_score.size()[-2:] == bbox_pred.size()[-2:]
/root/workspace/mmdeploy/mmdeploy/pytorch/functions/topk.py:58: TracerWarning: Converting a tensor to a Python boolean might cause the trace to be incorrect. We can't record the data flow of Python values, so this value will be treated as a constant in the future. This means that the trace might not generalize to other inputs!
  if k > size:
/root/workspace/mmdeploy/mmdeploy/codebase/mmdet/models/task_modules/coders/delta_xywh_bbox_coder.py:38: TracerWarning: Converting a tensor to a Python boolean might cause the trace to be incorrect. We can't record the data flow of Python values, so this value will be treated as a constant in the future. This means that the trace might not generalize to other inputs!
  assert pred_bboxes.size(0) == bboxes.size(0)
/root/workspace/mmdeploy/mmdeploy/codebase/mmdet/models/task_modules/coders/delta_xywh_bbox_coder.py:40: TracerWarning: Converting a tensor to a Python boolean might cause the trace to be incorrect. We can't record the data flow of Python values, so this value will be treated as a constant in the future. This means that the trace might not generalize to other inputs!
  assert pred_bboxes.size(1) == bboxes.size(1)
/root/workspace/mmdeploy/mmdeploy/mmcv/ops/nms.py:451: TracerWarning: Converting a tensor to a Python integer might cause the trace to be incorrect. We can't record the data flow of Python values, so this value will be treated as a constant in the future. This means that the trace might not generalize to other inputs!
  int(scores.shape[-1]),
/root/workspace/mmdeploy/mmdeploy/mmcv/ops/nms.py:148: TracerWarning: Converting a tensor to a Python boolean might cause the trace to be incorrect. We can't record the data flow of Python values, so this value will be treated as a constant in the future. This means that the trace might not generalize to other inputs!
  out_boxes = min(num_boxes, after_topk)
Process Process-2:
Traceback (most recent call last):
  File "/usr/local/lib/python3.8/dist-packages/torch/onnx/symbolic_opset9.py", line 1187, in symbolic_fn
    output_size = symbolic_helper._parse_arg(output_size, "is")
  File "/usr/local/lib/python3.8/dist-packages/torch/onnx/symbolic_helper.py", line 97, in _parse_arg
    raise RuntimeError(
RuntimeError: Failed to export an ONNX attribute 'onnx::Gather', since it's not constant, please try to make things (e.g., kernel size) static if possible

During handling of the above exception, another exception occurred:

Traceback (most recent call last):
  File "/usr/lib/python3.8/multiprocessing/process.py", line 315, in _bootstrap
    self.run()
  File "/usr/lib/python3.8/multiprocessing/process.py", line 108, in run
    self._target(*self._args, **self._kwargs)
  File "/root/workspace/mmdeploy/mmdeploy/apis/core/pipeline_manager.py", line 107, in __call__
    ret = func(*args, **kwargs)
  File "/root/workspace/mmdeploy/mmdeploy/apis/pytorch2onnx.py", line 98, in torch2onnx
    export(
  File "/root/workspace/mmdeploy/mmdeploy/apis/core/pipeline_manager.py", line 356, in _wrap
    return self.call_function(func_name_, *args, **kwargs)
  File "/root/workspace/mmdeploy/mmdeploy/apis/core/pipeline_manager.py", line 326, in call_function
    return self.call_function_local(func_name, *args, **kwargs)
  File "/root/workspace/mmdeploy/mmdeploy/apis/core/pipeline_manager.py", line 275, in call_function_local
    return pipe_caller(*args, **kwargs)
  File "/root/workspace/mmdeploy/mmdeploy/apis/core/pipeline_manager.py", line 107, in __call__
    ret = func(*args, **kwargs)
  File "/root/workspace/mmdeploy/mmdeploy/apis/onnx/export.py", line 131, in export
    torch.onnx.export(
  File "/usr/local/lib/python3.8/dist-packages/torch/onnx/__init__.py", line 350, in export
    return utils.export(
  File "/usr/local/lib/python3.8/dist-packages/torch/onnx/utils.py", line 163, in export
    _export(
  File "/usr/local/lib/python3.8/dist-packages/torch/onnx/utils.py", line 1074, in _export
    graph, params_dict, torch_out = _model_to_graph(
  File "/root/workspace/mmdeploy/mmdeploy/apis/onnx/optimizer.py", line 27, in model_to_graph__custom_optimizer
    graph, params_dict, torch_out = ctx.origin_func(*args, **kwargs)
  File "/usr/local/lib/python3.8/dist-packages/torch/onnx/utils.py", line 731, in _model_to_graph
    graph = _optimize_graph(
  File "/usr/local/lib/python3.8/dist-packages/torch/onnx/utils.py", line 308, in _optimize_graph
    graph = _C._jit_pass_onnx(graph, operator_export_type)
  File "/usr/local/lib/python3.8/dist-packages/torch/onnx/__init__.py", line 416, in _run_symbolic_function
    return utils._run_symbolic_function(*args, **kwargs)
  File "/usr/local/lib/python3.8/dist-packages/torch/onnx/utils.py", line 1406, in _run_symbolic_function
    return symbolic_fn(g, *inputs, **attrs)
  File "/usr/local/lib/python3.8/dist-packages/torch/onnx/symbolic_helper.py", line 308, in wrapper
    return fn(g, *args, **kwargs)
  File "/usr/local/lib/python3.8/dist-packages/torch/onnx/symbolic_opset9.py", line 1189, in symbolic_fn
    return symbolic_helper._onnx_unsupported(
  File "/usr/local/lib/python3.8/dist-packages/torch/onnx/symbolic_helper.py", line 454, in _onnx_unsupported
    raise RuntimeError(
RuntimeError: Unsupported: ONNX export of operator adaptive pooling, since output_size is not constant.. Please feel free to request support or submit a pull request on PyTorch GitHub.