Hilbert Distillation for Cross-Dimensionality Networks

Authors: Dian Qin, Haishuai Wang, Zhe Liu, HONGJIA XU, Sheng Zhou, Jiajun Bu

NeurIPS 2022 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We conduct extensive cross-dimensionality distillation experiments in activity and medical image classification domains... Therefore, we demonstrate the effectiveness of the proposed crossdimensionality distillation method on two datasets: Activity Net [5] dataset for activity classification and Large-COVID-19 [20] dataset for medical imaging classification.
Researcher Affiliation Academia 1Zhejiang Provincial Key Laboratory of Service Robot, College of Computer Science, Zhejiang University, Hangzhou, China 2School of Software Technology, Zhejiang University, Ningbo, China
Pseudocode Yes Algorithm 1: The Mapping Function of Hilbert Curve Hn=2,p
Open Source Code Yes In the supplemental material, we release the source codes as a URL.
Open Datasets Yes Activity Net [5] dataset for activity classification and Large-COVID-19 [20] dataset for medical imaging classification.
Dataset Splits No The paper mentions training and test splits (e.g., 'randomly split the 7000 videos into 2:1 for training and testing', 'randomly select 20% images as the test set'), but does not explicitly specify a separate validation dataset split.
Hardware Specification Yes We train and test models on an NVIDIA GeForce RTX 3090 GPU (24GB).
Software Dependencies No The paper mentions using the Adam optimization algorithm but does not specify version numbers for any software dependencies or libraries (e.g., Python, PyTorch, TensorFlow, etc.).
Experiment Setup Yes For the activity classification task, we randomly crop the input frame into 256 256. The batch size is set to 16. The training process lasts 40 epochs. The hyperparameter α in Eq. 8 is set to 103. For the medical imaging classification task, we resize the input image into 224 224 horizontally. The batch size for training is set to 32. The number of training epochs is 60. Different from the activity classification task, we set α to 10. The models are trained with an initial learning rate 0.01, and the Adam optimization algorithm is employed for both tasks.