Notice: The reproducibility variables underlying each score are classified using an automated LLM-based pipeline, validated against a manually labeled dataset. LLM-based classification introduces uncertainty and potential bias; scores should be interpreted as estimates. Full accuracy metrics and methodology are described in Coakley et alK. L. Coakley, T. Snelleman, H. Hoos, and O. E. Gundersen, "The embrace of open science: An analysis of a decade of AI research and 56 800 conference papers," Under Review, 2026..
Learning Generalizable Shape Completion with SIM(3) Equivariance
Authors: Yuqing Wang, Zhaiyu Chen, Xiaoxiang Zhu
NeurIPS 2025 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We first evaluate on the PCN benchmark [51], which comprises eight categories from ShapeNet [61] with paired partial and complete point clouds. To assess cross-domain transferability, we directly apply PCN-trained models, without further normalization, to real-world scans from KITTI [62] and Omni Object3D [63]. We compare against leading non-equivariant shape completion methods... Table 1 compares our SIM(3)-equivariant model against leading non-equivariant networks trained with augmentation. Our method achieves the lowest average CD-ℓ1 and the highest F1 score, outperforming Ada Poin Tr by 10% and 8%, respectively. |
| Researcher Affiliation | Academia | Yuqing Wang1* Zhaiyu Chen1,2* Xiao Xiang Zhu1,2 1Technical University of Munich 2Munich Center for Machine Learning |
| Pseudocode | No | The paper describes its methodology in detail across sections such as '3.2 SIM(3)-equivariant shape completion' and '3.3 Network architecture', but it does not present any explicitly labeled pseudocode or algorithm blocks. The operations are described in prose and mathematical formulations. |
| Open Source Code | Yes | 3. Protocol and resources. We establish a rigorous evaluation protocol that eliminates hidden pose and scale bias, release code for reproducibility, and provide thorough analyses that pinpoint where equivariance delivers its gains. Under this protocol our method sets a new state of the art. Project page: https://sime-completion.github.io. A Reproducibility. The code repository and demo are publicly accessible via the project page1. Detailed instructions for setup and running the code are described in the repository s README.md file. 1https://sime-completion.github.io |
| Open Datasets | Yes | Datasets and baselines. We first evaluate on the PCN benchmark [51], which comprises eight categories from ShapeNet [61] with paired partial and complete point clouds. To assess cross-domain transferability, we directly apply PCN-trained models, without further normalization, to real-world scans from KITTI [62] and Omni Object3D [63]. |
| Dataset Splits | No | For our model, which requires no training-time augmentation, we adopt the train/test setting of I/SIM(3) where I denotes the identity transform. Each baseline is first evaluated under the group it was designed for (i.e., I/SO(3) for Equiv PCN [13], I/SE(3) for ESCAPE [10], and SE(3)/SE(3) for SCARP [15]) to reveal their upper-bound performance when pose/scale cues are still partly available. We then report their performance under SIM(3)/SIM(3) with additional data augmentation. |
| Hardware Specification | Yes | With a batch size of 40, SIMECO trains at about 1 hour per epoch on two NVIDIA A40 GPUs. Table 8 reports the per-scan latency on PCN. Our model processes a scan in 76 ms end-to-end, about twice as fast as the next-quickest equivariant competitor, ESCAPE [10] (148 ms), and more than twice as fast as Equiv PCN [13] (172 ms) and SCARP [15] (172 ms). Inference denotes the network forward time, measured on an NVIDIA A40 GPU. |
| Software Dependencies | No | SIMECO is implemented in Py Torch and optimized using the Adam optimizer with an initial learning rate of 10 4, a weight decay of 5 10 4, and a learning-rate decay factor of 0.9 every 15 epochs. |
| Experiment Setup | Yes | SIMECO is implemented in Py Torch and optimized using the Adam optimizer with an initial learning rate of 10 4, a weight decay of 5 10 4, and a learning-rate decay factor of 0.9 every 15 epochs. We adopt the same architectural depth and hyperparameters as Ada Poin Tr [7]. The models, including baselines, were trained for 200 epochs on two NVIDIA A40 GPUs. All other completion methods [6, 7, 8, 10, 13, 15, 54, 64] were used with their default settings. With a batch size of 40, SIMECO trains at about 1 hour per epoch on two NVIDIA A40 GPUs. |