Object detectors used in safety critical or interactive settings should make their decisions understandable. We present AttentionCAM++ (AC++), a lightweight two stage post hoc explainer for DEtection TRansformer (DETR) based detectors such as Real-Time DEtection TRansformer (RTDETR) and RoboFlow DEtection TRansformer (RFDETR). Stage one instruments the Multi-Scale Deformable Attention (MSDA) layers in the decoder with forward callbacks to record sampling locations and attention weights, which are aggregated into class evidence maps. Stage two computes integrated gradients Integrated Gradients (IG) with respect to the decoder box deltas to produce a box attribution map that explains the predicted extent. We evaluate on COCO and on four remote sensing benchmarks, CHAI, DIOR, DOTAv2 and SARDET-100K, following the Object Detection Explainable AI Evaluation Benchmark (ODExAI) protocol. We report localization, PointingGame (PG), Energy-based Pointing-Game (EBPG), faithfulness through Insertion and Deletion, as well as complexity, sparsity and runtime. AC++ delivers near real time class evidence in our setup, while the optional box attribution adds modest latency, and the method requires no changes to detector weights. Code will be released at https://github.com/mburges-cvl/ACPR-ACplusplus