I try to achieve this code
pred = pred[pred[:, 4] > opt.conf_thres]
Pred has [Nx518] where n is the number of detection
auto pred = detection_model.forward_model(im_blob).at(0).squeeze();
//pred has shape [Nx518]
auto conf_mask = (pred.select(1, 4) > opt.conf_thres);
auto filtered_index = conf_mask.nonzero().squeeze();
pred = torch::index_select(pred, 0, filtered_index);
Is my C++ solution considered bad practice ? is there any other solution?