Efficient Net Clarification

Hi Sir,

In Efficient Net lecture vide, having resolution images, how it help neural network scale up or down the Neural Network ? Wil it scale up or down NN size ?

How double the resolution helps NN to scale up or down ?

Here is my understand about need of decrease resolution, if we dont have enough computational resources, smalll NN can be trained to run faster at the cost of little bit accuracy

But not sure why is the need of increase resoltuion ? can u please help to clarify ?

Hi Anbu,

The basic idea is to start from a minimal base model followed by upscaling to the level that fits the device the network is to be implemented on.

About this upscaling, Tan and Le write:

“we investigate the central question: is there a principled method to scale up ConvNets that can achieve better accuracy and efficiency? Our empirical study shows that it is critical to balance all dimensions of network width/depth/resolution, and surprisingly such balance can be achieved by simply scaling each of them with constant ratio. Based on this observation, we propose a simple yet effective compound scaling method. Unlike conventional practice that arbitrary scales these factors, our method uniformly scales network width, depth, and resolution with a set of fixed scaling coefficients.”

I hope this clarifies.

Sir above answer says that uniform scaling can de done . But our doubt is we not sure what is the need of having scaling resolution factor ? How it is going to help increase resolution or decrease resolution to scale up or down NN ?

  1. Should we train big or small NN for high resolution images in general?
  2. which is high resolution images help in prediction or low resolution images? Which helps improve accuracy high or low resolution images?

@reinoudbosch @paulinpaloalto @XpRienzo Can you please share your thoughts ?