Skip to content

This is the official page of the object detection network: CAT-YOLO

Notifications You must be signed in to change notification settings

GuanRunwei/MAN-and-CAT

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

42 Commits
 
 
 
 
 
 
 
 

Repository files navigation

MAN and CAT: Mix Attention to NN and Concatenate Attention to YOLO

===This is the official page of Mix Attention, MANet and CAT-YOLO===

CAT-YOLO

Benchmark Results

Mix Attention Block with ResNet-101 and WRN-18 on CIFAR-10

Model Params(M) Top-1 Error(%)
ResNet-101 + Mix Attention 50.07 6.17
WRN-18 + Mix Attention 27.11 4.77

Mix Attention Block with ResNet-101 and WRN-18 on CIFAR-100

Model Params(M) Top-1 Error(%)
ResNet-101 + Mix Attention 50.07 23.19
WRN-18 + Mix Attention 27.11 19.11

MANet on ImageNet

Model Params(M) Top-1 Accuracy(%)
MANet-B 69.3 81.7
MANet-S 23.4 78.3
MANet-T 4.3 73.1

MANet on CIFAR-10

Model Params(M) Top-1 Accuracy(%)
MANet-B 69.3 97.2
MANet-S 23.4 95.1
MANet-T 4.3 93.4

MANet on CIFAR-100

Model Params(M) Top-1 Accuracy(%)
MANet-B 69.3 88.7
MANet-S 23.4 86.5
MANet-T 4.3 81.6

CAT-YOLO on COCO 2017

Model Backbone Params(M) Latency(ms) AP
CAT-YOLO-v1 CSPDarknet53-Tiny 6.16 9.9(TITAN RTX) 24.1
CAT-YOLO-v2 MANet-T 9.17 12.7(TITAN RTX) 25.7
CAT-YOLO-v3 MANet-T 12.5 16.8(TITAN RTX) 33.5

User Guide

MAN includes the plug-and-play modules(Mix Attention) and backbone(MANet).

Note:

  1. We divide the modules and backbones for CIFAR and ImageNet respectively.

  2. The sub-folder named "Big Version" in Modules play the role of one individual layer.

  3. The sub-folder named "Tiny Version" in Modules play the role of the enhance module in the network's bottleneck.

CAT includes the files of CAT-YOLO.

CAT-YOLO

Citation

  @article{guan2022man,
  title={MAN and CAT: mix attention to nn and concatenate attention to YOLO},
  author={Guan, Runwei and Man, Ka Lok and Zhao, Haocheng and Zhang, Ruixiao and Yao, Shanliang and Smith, Jeremy and Lim, Eng Gee and Yue, Yutao},
  journal={The Journal of Supercomputing},
  pages={1--29},
  year={2022},
  publisher={Springer}
  }

About

This is the official page of the object detection network: CAT-YOLO

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages