You could reduce the memory requirement to around 1/4 by not using 2 levels of hash. A single level will do the job:

use strict;
use warnings;

open(my $fh, "<", "input.txt") 
   or die "cannot open < input.txt: $!";

my %duplicates;
while (my $line = <$fh>) {
    chomp $line;
    ++$duplicates{$line};
}
[download]

But that will still require around 4GB to build the 50e6 key hash. Better than 16GB, but you will still run out of memory if you are using a 32-bit Perl (unless you have a very high proportion of duplicates. eg. >50%)

As you say the data is presorted, investigate the uniq command