Object detection quantization

I followed the quantization steps to qunatized faster-rcnn, but I got the following errors in the RegionProposalNetwork part of the network when calling the box_coder.decode(pred_bbox_deltas.detach(), anchors) function.

Could not run ‘aten::empty_strided’ with arguments from the ‘QuantizedCPUTensorId’ backend. ‘aten::empty_strided’ is only available for these backends: [CPUTensorId, VariableTensorId]

Could not run ‘aten::div.Tensor’ with arguments from the ‘QuantizedCPUTensorId’ backend. ‘aten::div.Tensor’ is only available for these backends: [CPUTensorId, SparseCPUTensorId, VariableTensorId].

1 Like

let’s track the progress here: https://github.com/pytorch/pytorch/issues/31316