You are viewing a plain text version of this content. The canonical link for it is here.
Posted to user@hadoop.apache.org by venito camelas <ro...@gmail.com> on 2016/10/06 13:14:30 UTC

Hadoop: precomputing data

I'm designing a prototype using *Hadoop* for video processing to do face
recognition. I thought of 2 ways of doing it.

*Approach 1:*

I was thinking of doing something in 2 steps:

   1. A map that receives frames and if a face is found it gets stored for
   the next step.
   2. A map that receives the frames from step 1 (all frames containing 1
   face at least) and does face recognition.

Step 1 would be ran only once while step 2 runs every time I want recognize
a new face.


*Approach 2:*

The other approach I thought about is to do face recognition to all the
data every time

The first approach saves time because I don't have to process faceless
frames every time I want to do face recognition, it also uses more disk
space (and it could be a lot of space).


I'm not sure whats better. Is it a bad thing to leave that precomputed
frames there forever?

Re: Hadoop: precomputing data

Posted by Ravi Prakash <ra...@gmail.com>.
I guess one of the questions is what is your false negative rate in
Approach 1 Step 1?

Ofcourse if you are limited by resources you may have to go with Approach 1.

On Thu, Oct 6, 2016 at 6:14 AM, venito camelas <ro...@gmail.com>
wrote:

> I'm designing a prototype using *Hadoop* for video processing to do face
> recognition. I thought of 2 ways of doing it.
>
> *Approach 1:*
>
> I was thinking of doing something in 2 steps:
>
>    1. A map that receives frames and if a face is found it gets stored
>    for the next step.
>    2. A map that receives the frames from step 1 (all frames containing 1
>    face at least) and does face recognition.
>
> Step 1 would be ran only once while step 2 runs every time I want
> recognize a new face.
>
>
> *Approach 2:*
>
> The other approach I thought about is to do face recognition to all the
> data every time
>
> The first approach saves time because I don't have to process faceless
> frames every time I want to do face recognition, it also uses more disk
> space (and it could be a lot of space).
>
>
> I'm not sure whats better. Is it a bad thing to leave that precomputed
> frames there forever?
>