Receptive Fields As Experts in Convolutional Neural Architectures

Authors: Dongze Lian, Weihao Yu, Xinchao Wang

ICML 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Extensive experiments show that our approach outperforms the baselines in image classification, object detection, and segmentation tasks without significantly increasing the inference time.
Researcher Affiliation Academia Dongze Lian 1 Weihao Yu 1 Xinchao Wang 1 1Department of Electrical and Computer Engineering, National University of Singapore, Singapore. Correspondence to: Xinchao Wang <xinchao@nus.edu.sg>.
Pseudocode No The paper provides mathematical equations and descriptions of the approach but does not include any clearly labeled pseudocode or algorithm blocks.
Open Source Code No The paper does not contain any explicit statements about releasing source code for the described methodology or a link to a code repository.
Open Datasets Yes To evaluate the effectiveness of our proposed Mo RF, we select a wide range of model sizes to conduct experiments with Res Net-18 (He et al., 2016), Res Net-50 (He et al., 2016), and Conv Ne Xt (Liu et al., 2022c) on the Image Net-1K dataset (Deng et al., 2009). The experiments are conducted in the COCO datasets (Lin et al., 2014). We conduct experiments of semantic segmentation on the ADE20K dataset (Zhou et al., 2017).
Dataset Splits Yes We conduct experiments on the Image Net-1K dataset, which contains about 1.28 million training samples and 50,000 validation samples.
Hardware Specification Yes Throughput is measured with a batch size of 64 on a single V100 GPU (32GB).
Software Dependencies No The paper cites various algorithms and techniques from other papers (e.g., Adam W (Loshchilov & Hutter, 2019), mixup (Zhang et al., 2017), cutmix (Yun et al., 2019)), implying their use. However, it does not explicitly list specific software libraries or frameworks with their version numbers (e.g., 'PyTorch 1.9' or 'TensorFlow 2.x').
Experiment Setup Yes We follow the experimental settings of Conv Ne Xt (Liu et al., 2022c), and train our model for 300 epochs with the first 20 epochs used for warm-up. The initial learning rate is set to 0.001 with cosine decay and we use a batch size of 1024. We employ the Adam W (Loshchilov & Hutter, 2019) to optimizer our networks. The more specific details and the hyperparameters can be found in the appendix.