About CondInst Model To Onnx

[wjguo2002]

reference in condinst to onnx： proposals, proposal_losses = self.proposal_generator(images, features, gt_instances, self.controller)

return proposals

    mask_feats, sem_losses = self.mask_branch(features, gt_instances)
    #return mask_feats
    return mask_feats, proposals

I don't know why only return mask_feats, proposals, it donot include : pred_instances_w_masks = self._forward_mask_heads_test(proposals, mask_feats) there are what question that cannot resolve?

the foreward in self._forward_mask_heads_test(proposals, mask_feats) include many compute, it is difficult to C++ or other acc platform

[605436079]

I also encountered the same problem. How to implement this in C + +?
How should this part be implemented as post-processing?

[605436079]

reference in condinst to onnx： proposals, proposal_losses = self.proposal_generator(images, features, gt_instances, self.controller) #return proposals mask_feats, sem_losses = self.mask_branch(features, gt_instances) #return mask_feats return mask_feats, proposals

I don't know why only return mask_feats, proposals, it donot include : pred_instances_w_masks = self._forward_mask_heads_test(proposals, mask_feats) there are what question that cannot resolve?

the foreward in self._forward_mask_heads_test(proposals, mask_feats) include many compute, it is difficult to C++ or other acc platform

Have you solved it

[sssimpleboy]

reference in condinst to onnx： proposals, proposal_losses = self.proposal_generator(images, features, gt_instances, self.controller) #return proposals mask_feats, sem_losses = self.mask_branch(features, gt_instances) #return mask_feats return mask_feats, proposals I don't know why only return mask_feats, proposals, it donot include : pred_instances_w_masks = self._forward_mask_heads_test(proposals, mask_feats) there are what question that cannot resolve? the foreward in self._forward_mask_heads_test(proposals, mask_feats) include many compute, it is difficult to C++ or other acc platform

Have you solved it

Have you solved it ? I can just get 21 tensors from the .onnx.