- detector の中間 activation を compact な channel-wise statistics で要約
- その要約の短い sequence を temporal convolutional network で処理
- 時間方向の文脈を用いて missed-object failure を検出
- detector の変更や retraining を必要としない
- early および intermediate backbone features が最も強い monitoring signal を与える
- temporal inference head は CPU 上で 0.159 ms のみ
4. どうやって有効だと検証した？
- NuScenes 上で実験
- temporal monitoring と single-frame monitoring を比較
- AUROC と 95% recall 時の false alarm rate で評価
- どの feature が monitoring signal として強いかを early/intermediate backbone features について確認
- temporal inference head の CPU 推論時間を計測
5. 議論はある？
- early および intermediate backbone features が最も強い monitoring signal を与える点を報告
- temporal inference head の計算コストは CPU 上で 0.159 ms と軽量
- occlusion、point sparsity、scene complexity の時間的進展を活用できる可能性を示唆
- 具体的な限界や失敗ケース、他データセットへの一般化については要旨からは不明
Reliable perception is essential to intelligent transportation systems, yet deep neural networks can silently miss safety-critical objects while continuing to produce apparently plausible outputs. Existing run-time monitors for 3D object detection generally assess each frame independently and therefore cannot exploit the temporal evolution of occlusion, point sparsity, and scene complexity. We present TRIM, a lightweight temporal monitor that detects missed-object failures without modifying or retraining the underlying detector. TRIM summarises intermediate activations using compact channel-wise statistics and processes short sequences of these summaries with a temporal convolutional network. Experiments on NuScenes show that temporal monitoring improves AUROC from 0.686 to 0.704 and reduces the false alarm rate at 95% recall from 0.818 to 0.759 compared with single-frame monitoring. Early and intermediate backbone features provide the strongest monitoring signal, while the temporal inference head requires only 0.159 ms on a CPU.
