search

giddyyupp / coco-minitrain

a subset of coco dataset for faster experimentation
236 stars 34 forks source link

About training settings #12

Closed kcs6568 closed 2 years ago

kcs6568 commented 2 years ago

Hi. Thank you for the great project!

I'm curious about the training settings for each model epoch, learning rate, etc.

Can I get information about each model? (Faster RCNN, RetinaNet, etc)

nerminsamet commented 2 years ago

Hello,

In order to train the Faster RCNN, MaskRCNN and RetinaNet models we used MMDetection with the default parameter settings. Working with Detectron2 also should be OK. For the training of ExtremeNet, CornerNet and HoughNet we used the official repositories again with the default parameter settings. For some models we have to work with smaller batch sizes. In such cases we followed the linear learning rate rule to scale the learning rate according to batch size.

kcs6568 commented 2 years ago

Great! Thanks for the reply!