The first of each identity in the VGG face dataset.

vgg_full_scrape.sh

#!/bin/bash
# Author: Boaz Sender
#
# Script for downloading all the good images in the VGG data set.
#
# VGG is available for download from www.robots.ox.ac.uk/~vgg/data/vgg_face/
#
# usage: from the root of a vgg_face_dataset download, run:
# mkdir full_scrape && ./vgg_full_scrape.sh

# Loop over all the files in the vgg data set that we downloaded
for file in `ls -1 files`
  do
    # Figure out the name of the person who we are downloading an image for
    # from the file name
    identity=$(echo $file | rev | cut -c5- | rev)
    declare -i count=1

    # make a file to log all the dead images in
    touch dead_images.txt

    function scrape_image()
    {
      # download the image
      wget $1 --output-document=scrape_full/$2 --timeout=6 --tries 2;

      # If the download creates a recognizeable image, then figure out the format,
      #  give it the correct extension, and move on
      if file "scrape_full/$2" | grep "image"; then
        format=$(file -0 -F" " "scrape_full/$2" | grep -aPo '\0\s*\K\S+' | tr '[:upper:]' '[:lower:]')
        mv "scrape_full/$2" "scrape_full/$2.$format"

        echo "******************************************************************"
        echo "Images successful! for ${identity} at count number ${count}!"
        echo "******************************************************************"

        # Otherwise, if we got an unrecognizeable image delete that garbo
        # and log that is was a bad image
      else
        echo "******************************************************************"
        echo "Images download failed for ${identity} at count number ${count}! Recording and moving on"
        echo "******************************************************************"
        echo $identity $2 >> dead_images.txt
        rm scrape_full/$2
      fi

      # Increment the counter, and call this function again with the next line
      ((count++))
      arg=$count
      arg+="p"
      nextarr=(`sed -n $arg 'files/'$file`)
      scrape_image ${nextarr[1]} ${identity}_${nextarr[0]}
    }

    # Kickoff this shindig with the first line of the current file
    arr=(`sed -n 1p 'files/'$file`)
    scrape_image ${arr[1]} ${identity}_${arr[0]}
done;

vgg_process_files.sh

#!/bin/bash
# usage: from the root of a vgg_face_dataset download run
# ./vgg_process_files.sh > vgg_face_dataset_first_image_per_identity.txt
for i in `ls -1 files`
do
   head 'files/'$i -n1 -v
done;

vgg_process_files_and_scrape.sh

#!/bin/bash
# Author: Boaz Sender
#
# Script for processing the data in vgg_face_dataset and downloading a single
# image for each identity (person) in the data.
#
# The VGG data is available for download from
# www.robots.ox.ac.uk/~vgg/data/vgg_face/
#
# At the time of this writing, about 12% of the first images of each identity
# are no longer available, or borken in some way. This script tries
# downloading until it finds a good image.
#
# usage: from the root of a vgg_face_dataset download, run:
# mkdir scrape && ./vgg_process_files_and_scrape.sh

# Define our scraper function
function scrape_image()
{
  # download the image, and if it downloads and creates a recognizeable image,
  # then figure out the format, give it the correct extension, and move on
  if wget $1 --output-document=scrape/$2 -q --timeout=6 --tries 2 && file "scrape/$2" | grep "image" && (( $(identify -format %n scrape/$2) < 2)); then
    format=$(file -0 -F" " "scrape/$2" | grep -aPo '\0\s*\K\S+' | tr '[:upper:]' '[:lower:]')

    mv "scrape/$2" "scrape/$2.$format"
    echo "******************************************************************"
    echo "Images successful! for ${identity} at count number ${count}!"
    echo "******************************************************************"

  # otherwise, if the wget fails, or downloads and empty file, or something
  # that is not an image delete that garbo, and try again with the next line
  # of the identity's file
  else
    ((count++))
    echo "******************************************************************"
    echo "Image failed, moving on to image number ${count} for ${identity}"
    echo "******************************************************************"
    rm scrape/$2

    arg=$count
    arg+="p"
    nextarr=(`sed -n $arg 'files/'$file`)
    scrape_image ${nextarr[1]} ${identity}_${nextarr[0]}
  fi
}


# Loop over all the files in the vgg data set that we downloaded
for file in `ls -1 files`
  do
    # Figure out the name of the person who we are downloading an image for
    # from the file name
    identity=$(echo $file | rev | cut -c5- | rev)
    declare -i count=1


    # Kickoff this shindig with the first line of the current file
    arr=(`sed -n 1p 'files/'$file`)
    scrape_image ${arr[1]} ${identity}_${arr[0]}
done;

vgg_scrape.sh

#!/bin/bash
# usage: from the root of a vgg_face_dataset download run
# mkdir scrape && ./vgg_scrape.sh
while read p; do
  arr=(`echo ${p}`)
  wget ${arr[2]} --output-document=scrape/${arr[0]}.jpg --timeout=6 --tries 2
done < vgg_face_dataset_first_image_per_identity.txt

It would be slightly more performant if you didn't define your functions repeatedly in the loop. I'm sure the performance difference is totally irrelevant for your use case since so much time is spent fetching images, but I thought you might be interested in seeing this anyway.

Here are the commands to test:

time ./timing.sh declare-repeatedly
./timing.sh declare-repeatedly  5.70s user 0.06s system 99% cpu 5.800 total

time ./timing.sh declare-once
./timing.sh declare-once  3.30s user 0.04s system 99% cpu 3.362 total

...here is the code to put in timing.sh:

#!/bin/bash 

if [[ $1 == "declare-once" ]]; then
  scrape_image() {
    # i've turned this function into a noop but bash still has to parse the whole thing
    continue
    # download the image
    wget $1 --output-document=scrape_full/$2 --timeout=6 --tries 2;

    # If the download creates a recognizeable image, then figure out the format,
    #  give it the correct extension, and move on
    if file "scrape_full/$2" | grep "image"; then
      format=$(file -0 -F" " "scrape_full/$2" | grep -aPo '\0\s*\K\S+' | tr '[:upper:]' '[:lower:]')
      mv "scrape_full/$2" "scrape_full/$2.$format"

      echo "******************************************************************"
      echo "Images successful! for ${identity} at count number ${count}!"
      echo "******************************************************************"

      # Otherwise, if we got an unrecognizeable image delete that garbo
      # and log that is was a bad image
    else
      echo "******************************************************************"
      echo "Images download failed for ${identity} at count number ${count}! Recording and moving on"
      echo "******************************************************************"
      echo $identity $2 >> dead_images.txt
      rm scrape_full/$2
    fi

    # Increment the counter, and call this function again with the next line
    ((count++))
    arg=$count
    arg+="p"
    nextarr=(`sed -n $arg 'files/'$file`)
    scrape_image ${nextarr[1]} ${identity}_${nextarr[0]}
  }
  for i in {1..10000}; do
    scrape_image "$i"
  done
fi

if [[ $1 == "declare-repeatedly" ]]; then
  for i in {1..10000}; do
    scrape_image() {
      # i've turned this function into a noop but bash still has to parse the whole thing
      continue
      # download the image
      wget $1 --output-document=scrape_full/$2 --timeout=6 --tries 2;

      # If the download creates a recognizeable image, then figure out the format,
      #  give it the correct extension, and move on
      if file "scrape_full/$2" | grep "image"; then
        format=$(file -0 -F" " "scrape_full/$2" | grep -aPo '\0\s*\K\S+' | tr '[:upper:]' '[:lower:]')
        mv "scrape_full/$2" "scrape_full/$2.$format"

        echo "******************************************************************"
        echo "Images successful! for ${identity} at count number ${count}!"
        echo "******************************************************************"

        # Otherwise, if we got an unrecognizeable image delete that garbo
        # and log that is was a bad image
      else
        echo "******************************************************************"
        echo "Images download failed for ${identity} at count number ${count}! Recording and moving on"
        echo "******************************************************************"
        echo $identity $2 >> dead_images.txt
        rm scrape_full/$2
      fi

      # Increment the counter, and call this function again with the next line
      ((count++))
      arg=$count
      arg+="p"
      nextarr=(`sed -n $arg 'files/'$file`)
      scrape_image ${nextarr[1]} ${identity}_${nextarr[0]}
    }
    scrape_image "$i"
  done
fi

Great point, updated!