Exploring 3D Dataset Pruning

Dataset pruning remains underexplored for 3D modalities, where inherent class imbalance persists across both training and test sets. This creates a divergence in evaluation: overall accuracy favors natural frequency, reflecting practical usage; while mean accuracy demands balanced generalization. Instead of forcing a premature trade-off, we advocate for base principles that remain universally robust and beneficial across diverse priors. We cast pruning as a quadrature approximation on population risk and decompose the error bound into representation error (fidelity to the underlying manifold) and prior-mismatch bias (distribution shift), clarifying what can be improved jointly across priors. To address prior-mismatch bias, we decouple likelihood from prior in the posterior and transfer the structural likelihood via distillation with a calibrated teacher and geometry-preserving constraints. Simultaneously, to reduce representation error, we audit common pruning signals and choose geometric embedding, which exhibits greater robustness given the high inductive bias of 3D models. We also prioritize a safety floor before selection, capturing high-reward regions beneficial across priors. Finally, acknowledging that no single subset optimally satisfies divergent evaluation priors, we augment these principles with a steering wrapper that interpolates between stratified seeding and global selection. Empirical results demonstrate that our framework elevates the performance floor while offering flexibility for different prior preferences.

Modern 3D object recognition systems learn from collections of digital shapes and scans, but these collections are often uneven: some object categories appear many times, while others are rare. A common way to reduce training cost is to keep only a smaller subset of the data, but doing this for 3D data is difficult. If the subset mostly keeps common objects, the model may perform well on frequent categories but fail on rare ones. If the subset is forced to be too balanced, it may lose useful examples from the natural data distribution. This paper studies how to build smaller training sets for 3D recognition while protecting both common and rare categories. We propose a method that keeps representative examples from the overall dataset, guarantees a minimum level of coverage for each category, and uses guidance from a model trained on the full dataset to make learning from the smaller subset more reliable. Experiments on several 3D recognition benchmarks show that this approach can reduce the amount of training data while improving the balance between overall performance and performance across categories.