def prepare_training(self, output, batch): init = snake_gcn_utils.prepare_training(output, batch) output.update({ 'i_it_4py': init['i_it_4py'], 'i_it_py': init['i_it_py'] }) output.update({ 'i_gt_4py': init['i_gt_4py'], 'i_gt_py': init['i_gt_py'] }) return init
def prepare_training(self, output, batch): init = snake_gcn_utils.prepare_training(output, batch) if cfg.poly_cls_branch: spec_num = 80 init['i_gt_py'] = torch.cat( (init['i_gt_py'][:spec_num], init['neg_i_gt_py'][:spec_num]), dim=0) init['c_gt_py'] = torch.cat( (init['c_gt_py'][:spec_num], init['neg_c_gt_py'][:spec_num]), dim=0) init['i_it_py'] = torch.cat( (init['i_it_py'][:spec_num], init['neg_i_it_py'][:spec_num]), dim=0) init['c_it_py'] = torch.cat( (init['c_it_py'][:spec_num], init['neg_c_it_py'][:spec_num]), dim=0) init['py_ind'] = torch.cat( (init['py_ind'][:spec_num], init['py_ind'][:spec_num]), dim=0) poly_num = len(init['i_gt_py']) poly_cls_labels = torch.ones(poly_num) poly_cls_labels[int(poly_num / 2):] = 0 output.update({'poly_cls_labels': poly_cls_labels.cuda()}) if 0: input_imgs = batch['inp'].detach().cpu().numpy() i_gt_py = init['i_gt_py'].detach().cpu().numpy() i_it_py = init['i_it_py'].detach().cpu().numpy() neg_i_gt_py = init['neg_i_gt_py'].detach().cpu().numpy() neg_i_it_py = init['neg_i_it_py'].detach().cpu().numpy() py_inds = init['py_ind'].detach().cpu().numpy() np.save('input_imgs.npy', input_imgs) np.save('i_gt_py.npy', i_gt_py) np.save('i_it_py.npy', i_it_py) np.save('neg_i_gt_py.npy', neg_i_gt_py) np.save('neg_i_it_py.npy', neg_i_it_py) np.save('py_inds.npy', py_inds) print('saving training data done!') exit() output.update({ 'i_it_4py': init['i_it_4py'], 'i_it_py': init['i_it_py'] }) output.update({ 'i_gt_4py': init['i_gt_4py'], 'i_gt_py': init['i_gt_py'] }) return init