PDO-s3DCNNs: Partial Differential Operator Based Steerable 3D CNNs
Authors: Zhengyang Shen, Tao Hong, Qi She, Jinwen Ma, Zhouchen Lin
ICML 2022 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Extensive experiments show that our models can preserve equivariance well in the discrete domain, and outperform previous works on SHREC 17 retrieval and ISBI 2012 segmentation tasks with a low network complexity. We perform extensive experiments to evaluate the performance of our models. |
| Researcher Affiliation | Collaboration | 1School of Mathematical Sciences, Peking University, Beijing, China 2Bytedance AI Lab, Haidian District, Beijing, China 3Key Lab. of Machine Perception (Mo E), School of Artificial Intelligence, Peking University, Beijing, China 4Institute for Artificial Intelligence, Peking University, Beijing, China 5Pazhou Lab, Guangzhou, China. |
| Pseudocode | No | The paper does not contain any structured pseudocode or algorithm blocks. |
| Open Source Code | No | The paper does not provide concrete access to source code for the methodology described. |
| Open Datasets | Yes | The SHREC 17 retrieval task (Savva et al., 2017) contains 51,162 models of 3D shapes belonging to 55 classes. This dataset is divided into 35,764 training samples, 5,133 validation samples, and 10,265 test samples. We focus on the perturbed version where models are arbitrarily rotated. ISBI 2012 Challenge (Arganda-Carreras et al., 2015) is a volumetric boundary segmentation benchmark, and the target is to segment Drosophila ventral nerve cords from a serial-section transmission electron microscopy image. |
| Dataset Splits | Yes | The SHREC 17 retrieval task... This dataset is divided into 35,764 training samples, 5,133 validation samples, and 10,265 test samples. We take the first 25 slices of the training image as our training set, and the last 5 as the validation set. |
| Hardware Specification | Yes | Our experiments are implemented using Pytorch, and each one is run using one single Tesla-V100 GPU. |
| Software Dependencies | No | The paper mentions 'Pytorch' as the implementation framework but does not provide specific version numbers for PyTorch or any other software dependencies. |
| Experiment Setup | Yes | We train for 200 epochs with an initial learning rate of 0.01 and an exponential decay of 0.98 after 50 epochs. We train for 2000 epochs with a batch size of 32. The initial learning rate is set to 0.01 and is divided by 10 at 700 and 1, 400 epochs. We train for 4, 000 iterations with a batch size of 4. We use an initial learning rate of 0.001 and an exponential decay of 0.99. |