Rethinking 3D Convolution in $\ell_p$-norm Space
Authors: Li Zhang, Yan Zhong, Jianan Wang, Zhe Min, RujingWang , Liu Liu
NeurIPS 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Experimental results indicate that our approach exhibits competitive performance with traditional CNNs, with lower energy consumption and instruction latency. We apply our methods to classic networks and conduct related experiments. |
| Researcher Affiliation | Collaboration | Li Zhang1,2,4, Yan Zhong3, Jianan Wang4, Zhe Min5, Rujing Wang1,2, Liu Liu6 1 Hefei Institute of Physical Science, Chinese Academy of Sciences 2 University of Science and Technology of China, Hefei, China 3 School of Mathematical Sciences, Peking University. Beijing, China 4 Astribot, Shenzhen, China 5 Shandong University, Jinan, China 6 Hefei University of Technology, Hefei, China |
| Pseudocode | Yes | Algorithm 1 OMD Input: Initial learning rate α, hyper-parameters p1, p2, p3, referred by Eq. 13 and Eq. 14. q0 and q in (0, 1). 1: m0 = 0, α(0) = α, x1 = 0. 2: Set the functions α1(k) and α2(k) by hyper-parameters p1, p2, p3. 3: for k = 1 to T do 4: gk L K (xk) # Consider the gradient L K as an vector here. L K(i,j) = L Y , Y K(i,j) . L Y only depends on the choice of loss function. See Eq. 12 for Y K(i,j). 5: qk = q0 qk. 6: mk = qk mk 1 + (1 qk) gk 7: ˆα(k) min max{α1(k), A[α(k)]}, α2(k) 8: α(k) ˆα(k)/ k 9: xk+1 = ΠF,α(k) 1/2 (xk α(k) mk) 10: end for |
| Open Source Code | Yes | Question: Does the paper provide open access to the data and code, with sufficient instructions to faithfully reproduce the main experimental results, as described in supplemental material? Answer: [Yes] Justification: see the zip files of codes in the supplemental material. |
| Open Datasets | Yes | Dataset. 1) Shape Net. In Shape Net, there are 16,881 shapes from 16 categories, which are annotated with 50 parts in total. ... 2) S3DIS. The Stanford Large-Scale 3D Indoor Spaces Dataset, which encompasses 3D scans obtained from Matterport scanners across 6 distinct areas, comprising a total of 271 rooms. ... 3) Garment Nets Simulation. Garment Nets Simulation is a large-scale dataset proposed by [44]. |
| Dataset Splits | No | The paper mentions 'K-fold strategy' and 'randomly sampling up to 4096 points during training' for specific datasets, and that 'The best model on the validation set is selected for testing.' However, it does not provide explicit numerical training, validation, or test dataset splits (e.g., 80/10/10 percentages or specific sample counts). |
| Hardware Specification | Yes | All experiments are conducted on a computer workstation with three Ge Force GTX 3090 GPUs using the Py Torch deep learning framework. |
| Software Dependencies | No | The paper mentions 'Py Torch deep learning framework' and the 'Adam W optimizer' but does not specify version numbers for these software components or any other libraries. |
| Experiment Setup | Yes | We train our frameworks using Cross Entropy loss and the Adam W optimizer [45], with an initial learning rate of 0.001, a weight decay of 10 4, Cosine Decay, and a batch size of 32. The total training consists of 200 epochs. All tasks use the same settings unless otherwise specified. |