error: no suitable constructor exists to convert from "c10::ScalarType" to "at::Type"

[cslxiao]

I encounter this error when executing

python setup.py build develop

---

det3d/ops/sigmoid_focal_loss/src/sigmoid_focal_loss_cuda.cu(121): error: no suitable constructor exists to convert from "c10::ScalarType" to "at::Type"

[a157801]

I encounter this error when executing

python setup.py build develop

det3d/ops/sigmoid_focal_loss/src/sigmoid_focal_loss_cuda.cu(121): error: no suitable constructor exists to convert from "c10::ScalarType" to "at::Type"

Please check the version of cuda cudnn and gcc. I use gcc 5.4.0.

[poodarchu]

you need to use gpu to compile. or it will raise error when executing. Please follow instructions.

[cslxiao]

I use cuda 10, cudnn 7.5 and gcc 7.4. I strictly followed the instructions but still got the error

[a157801]

I use cuda 10, cudnn 7.5 and gcc 7.4. I strictly followed the instructions but still got the error

Please use gcc 5.4.0 and try again

[muzi2045]

If you are using pytorch 1.0, you will face this error(same as me). there is some api changed between torch 1.0 and 1.3 check this part in the sigmoid_focal_loss_cuda.cu file

/** pytorch 1.1+ **/
  // AT_DISPATCH_FLOATING_TYPES_AND_HALF(
  //     logits.scalar_type(), "SigmoidFocalLoss_forward", [&] {
  //       SigmoidFocalLossForward<scalar_t><<<grid, block>>>(
  //           losses_size, logits.contiguous().data<scalar_t>(),
  //           targets.contiguous().data<int64_t>(), num_classes, gamma, alpha,
  //           num_samples, losses.data<scalar_t>());
  //     });

  /** pytorch 1.0 **/
  AT_DISPATCH_FLOATING_TYPES_AND_HALF(
      logits.type(), "SigmoidFocalLoss_forward", [&] {
        SigmoidFocalLossForward<scalar_t><<<grid, block>>>(
            losses_size, logits.contiguous().data<scalar_t>(),
            targets.contiguous().data<int64_t>(), num_classes, gamma, alpha,
            num_samples, losses.data<scalar_t>());
      });

[cslxiao]

@muzi2045 Thanks. Updating torch solves this problem.