self.ensemble_proposals is rpn_proposals and fpr_res are the results from the classification network. (Which are thresholded by cfg['rcnn_test_nms_pre_score_threshold']. So there will be a size mismatch while doing this addition if some of the rpn_proposals were discarded during the above step.
https://github.com/uci-cbcl/NoduleNet/blob/364621c81e899ffaf2de512ba24d25969d8e15c3/net/nodule_net.py#L391
self.ensemble_proposals is rpn_proposals and fpr_res are the results from the classification network. (Which are thresholded by cfg['rcnn_test_nms_pre_score_threshold']. So there will be a size mismatch while doing this addition if some of the rpn_proposals were discarded during the above step.