MDP: Multidimensional Vision Model Pruning with Latency Constraint

Figure 1. MDP exhibits Pareto dominance with both CNNs and Transformers in tasks ranging from ImageNet classification to NuScenes 3D detection. Speedup are shown relative to the dense model. [Left] On ImageNet pruning ResNet50, we achieve a 28% speed increase alongside a +1.4 improvement in Top-1 compared with prior art [58]. [Middle] On ImageNet pruning DEIT-Base, compared with very recent Isomorphic Pruning (ECCV'24)[23], our method further accelerates the baseline by an additional 37% while yielding a +0.7 gain in Top-1. [Right] For 3D object detection, we obtain higher speed (×1.18) and mAP (0.451 vs. 0.449) compared to the dense baseline.