Darknet: Would it be a good idea to merge Coco | Open Images | Imagenet datasets and train ?

Created on 14 May 2018  路  1Comment  路  Source: AlexeyAB/darknet

I am looking to train a model for Vehicles and Persons. If i increase the dataset by merging (Coco + Open Images + Imagenet). Should it increase accuracy for Yolov3 tiny ?

picture dexception

Most helpful comment

Yes.
Yolo v2 that is trained on VOC+COCO better than trained only on VOC: http://host.robots.ox.ac.uk:8080/leaderboard/displaylb.php?challengeid=11&compid=4

  • YOLOv2 (VOC + COCO)聽 81.5
  • YOLOv2聽 75.4
picture AlexeyAB on 14 May 2018
👍4

>All comments

Yes.
Yolo v2 that is trained on VOC+COCO better than trained only on VOC: http://host.robots.ox.ac.uk:8080/leaderboard/displaylb.php?challengeid=11&compid=4

  • YOLOv2 (VOC + COCO)聽 81.5
  • YOLOv2聽 75.4
AlexeyAB picture AlexeyAB on 14 May 2018
👍4
Was this page helpful?
0 / 5 - 0 ratings

Related issues

qianyunw picture qianyunw  路  3Comments
Mididou picture Mididou  路  3Comments
Cipusha picture Cipusha  路  3Comments
Yumin-Sun-00 picture Yumin-Sun-00  路  3Comments
hemp110 picture hemp110  路  3Comments