A Transformer-Based Object Detector with Coarse-Fine Crossing Representations

Authors: Zhishan Li, Ying Nie, Kai Han, Jianyuan Guo, Lei Xie, Yunhe Wang

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

Reproducibility Variable Result LLM Response
Research Type Experimental Experimental results on the COCO dataset demonstrate the effectiveness of the proposed method.
Researcher Affiliation Collaboration 1College of Control Science and Engineering, Zhejiang University 2Huawei Noah s Ark Lab
Pseudocode No The paper provides mathematical formulations and architectural diagrams but does not include explicit pseudocode or algorithm blocks.
Open Source Code Yes Code will be available at https: //gitee.com/mindspore/models/tree/master/research/cv/CFDT.
Open Datasets Yes We conduct experiments on Microsoft COCO 2017 benchmark [1].
Dataset Splits Yes Following the usual practice, 118K images are used for training, 5K images for testing.
Hardware Specification Yes We gratefully acknowledge the support of Mind Spore [53], CANN(Compute Architecture for Neural Networks) and Ascend AI Processor used for this research.
Software Dependencies No The paper mentions software frameworks like Mind Spore and CANN but does not provide specific version numbers for them.
Experiment Setup Yes We follow the training strategy provided in Vi DT [14], including Adam W [45] with the initial learning rate of 1 10 4, training with multi-scale input sizes and the total training epoch is set as 50. ... the training batch size per card of P-Tiny and P-Small is set as 2, while that of P-Medium is 1.