Skip to content

Files

Latest commit

375f5d4 · Apr 19, 2025

History

History

object_detection

Folders and files

NameName
Last commit message
Last commit date

parent directory

..
Jul 10, 2023
Jun 16, 2023
Apr 9, 2024
Apr 17, 2024
Feb 11, 2025
Feb 2, 2024
Feb 2, 2024
Oct 13, 2023
Aug 2, 2024
Aug 6, 2024
Apr 9, 2024
Feb 2, 2024
Nov 3, 2024
Feb 2, 2024
Oct 12, 2023
Jul 10, 2023
Jul 10, 2023
Feb 2, 2024
Apr 11, 2023
Aug 31, 2024
Jan 9, 2025
Feb 2, 2024
Feb 2, 2024
Apr 11, 2023
Jul 10, 2023
Jul 4, 2023
Jul 10, 2023
Nov 18, 2024
Nov 21, 2024
Jul 10, 2023
Jul 10, 2023
Jul 10, 2023
Jul 10, 2023
Feb 2, 2024
Feb 2, 2024
Feb 2, 2024
Jul 10, 2023
Jul 10, 2023
Feb 2, 2024
Feb 29, 2024
Feb 2, 2024
Jul 10, 2023
Aug 25, 2022

ailia MODELS : Object Detection

Models for bounding box

Detect bounding box of objects from single image.

Name mAP75 mAP50 GFlops Resolution Publish Date
yolov7 51.72 66.17 104.7 640 2022.7
yolox_l 48.98 64.45 155.6 640 2021.8
yolox_m 46.28 62.12 73.8 640 2021.8
yolov4 43.02 64.38 129.5 416 2020.4
yolov6_t 40.76 57.06 36.7 640 2022.6
yolox_s 39.53 56.35 26.8 640 2021.8
yolov3 36.94 65.23 65.86 416 2018.4
yolov5s6 35.12 51.90 16.8 640 2021.10
yolov5s 34.11 51.89 16.5 640 2020.6
yolov7_tiny 32.34 46.96 5.8 416 2022.7
yolox_tiny 31.36 47.04 6.45 416 2021.8
yolov6_n 29.69 43.79 4.7 416 2022.6
yolox_nano 24.15 39.03 1.08 416 2021.8
yolov4_tiny 16.13 36.31 6.92 416 2020.4
yolov3_tiny 12.65 35.76 5.56 416 2018.4

Metrics

mAP (Accuracy)

Basically the accuracy of object detection algorithm is calculated by mAP. In this page, mAP was calculated using this repository.

https://github.com/axinc-ai/ailia-models-measurement/tree/main/object_detection

The repository uses Object-Detection-Metrics for mAP calculation.

https://github.com/rafaelpadilla/Object-Detection-Metrics

We used COCO2017 val images for testing. We set parameters, iou = 0.5 (mAP50) and iou = 0.75 (mAP75), detection threshold = 0.01 (Because small value achieves high accuracy).

GFlops (Computing cost)

GFlops was referred from below site.

Models for segmentation mask

Detect bounding box and mask of objects from single image.

Name maskAP Classes Resolution Publish Date
detic swinB 41.3 (LVIS-all) 1000+ 800 2022.1
detic Res50 33.2 (LVIS-all) 1000+ 800 2022.1
maskrcnn R_50_FPN 34.2 (coco) 80 800 2019.9

Metrics

maskAP

maskAP was referred from below site.

Leader board

Object Detection https://paperswithcode.com/task/object-detection