search

SysCV / transfiner

Mask Transfiner for High-Quality Instance Segmentation, CVPR 2022
https://www.vis.xyz/pub/transfiner
Apache License 2.0
534 stars 61 forks source link

how to export trained weights to onnx #39

Closed Metal-joker closed 1 year ago

Metal-joker commented 2 years ago

have tried detectron export script but failed.

  File "/workspace/transfiner/detectron2/export/caffe2_modeling.py", line 274, in forward
    detector_results, _ = self._wrapped_model.roi_heads(images, features, proposals)
  File "/opt/conda/lib/python3.8/site-packages/torch/nn/modules/module.py", line 725, in _call_impl
    result = self._slow_forward(*input, **kwargs)
  File "/opt/conda/lib/python3.8/site-packages/torch/nn/modules/module.py", line 709, in _slow_forward
    result = self.forward(*input, **kwargs)
  File "/workspace/transfiner/detectron2/modeling/roi_heads/roi_heads.py", line 752, in forward
    pred_instances = self.forward_with_given_boxes(features, pred_instances)
  File "/workspace/transfiner/detectron2/modeling/roi_heads/roi_heads.py", line 778, in forward_with_given_boxes
    instances = self._forward_mask(features, instances)
  File "/workspace/transfiner/detectron2/modeling/roi_heads/roi_heads.py", line 852, in _forward_mask
    features = self.mask_pooler(features, boxes)
  File "/opt/conda/lib/python3.8/site-packages/torch/nn/modules/module.py", line 725, in _call_impl
    result = self._slow_forward(*input, **kwargs)
  File "/opt/conda/lib/python3.8/site-packages/torch/nn/modules/module.py", line 709, in _slow_forward
    result = self.forward(*input, **kwargs)
  File "/workspace/transfiner/detectron2/export/c10.py", line 360, in forward
    assert roi_feat_shuffled.numel() > 0 and rois_idx_restore_int32.numel() > 0, (
AssertionError: Caffe2 export requires tracing with a model checkpoint + input that can produce valid detections. But no detections were obtained with the given checkpoint and input!
qmcreeper commented 2 years ago

Set weights path in configs/transfiner/xxxx.yaml,then use tools/deploy/export_model.py to deploy onnx.

I found another thing may cause this error, your input image must contain at least one object.I generate some images with np.random and it looks like a big noise with no object.The script cannot export model.