Cross-Layer Retrospective Retrieving via Layer Attention

Authors: Yanwen Fang, Yuxi CAI, Jintai Chen, Jingyu Zhao, Guangjian Tian, Guodong Li

ICLR 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Its effectiveness has been extensively evaluated in image classification, object detection and instance segmentation tasks, where improvements can be consistently observed.
Researcher Affiliation Collaboration 1Department of Statistics & Actuarial Science, The University of Hong Kong 2College of Computer Science and Technology, Zhejiang University 3Huawei Noah s Ark Lab
Pseudocode Yes Pseudo codes of MRLA-base s and MRLA-light s implementations in CNNs and vision transformers are given below.
Open Source Code Yes Our code is available at https://github.com/joyfang1106/MRLA.
Open Datasets Yes We use the middle-sized Image Net-1K dataset (Deng et al., 2009) directly.
Dataset Splits Yes Table 1: Comparisons of single-crop accuracy on the Image Net-1K validation set.
Hardware Specification Yes All models are implemented by Py Torch toolkit on 4 V100 GPUs.
Software Dependencies No The paper mentions software like PyTorch, MMDetection, and timm, but does not provide specific version numbers for these dependencies.
Experiment Setup Yes Specifically, the input images are randomly cropped to 224 × 224 with random horizontal flipping. The networks are trained from scratch using SGD with momentum of 0.9, weight decay of 1e-4, and a mini-batch size of 256. The models are trained within 100 epochs by setting the initial learning rate to 0.1, which is decreased by a factor of 10 per 30 epochs.