Darknet with differential learning rate

tcapelle · November 9, 2018, 7:42pm

I want some tips on how to create a Darknet with differential group layers to be able to use differential learning rate.
I am training in images that are big, so I would like to train first with small ones, and increase size progressively. To do this, I would like to be able to train the bottom, and the head. I just want the same behavior that you get with a resnet34 model, but with a custom darknet or wideresnet, what should I need to add to this classes to be able to do so in V1?

tcapelle · November 11, 2018, 4:10am

The answer to this is here:

github.com

fastai/fastai_docs/blob/master/dev_nb/004a_discriminative_lr.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "%reload_ext autoreload\n",
    "%autoreload 2"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "#export\n",
    "from nb_004 import *"

This file has been truncated. show original