WEBVTT

1
00:00:00.240 --> 00:00:03.720
- For a process that can cost you

2
00:00:03.720 --> 00:00:05.520
or make you hundreds of thousand dollars

3
00:00:05.520 --> 00:00:07.920
or more, typically thousands
of dollars in the traditional,

4
00:00:07.920 --> 00:00:10.110
you know, for modern cards
that needs to be assigned,

5
00:00:10.110 --> 00:00:14.700
that needs to be a standard
for every grader to follow.

6
00:00:14.700 --> 00:00:16.473
And that's what Mike and I created.

7
00:00:20.010 --> 00:00:21.690
Until COVID started,

8
00:00:21.690 --> 00:00:25.170
we didn't realize the importance
that grading had on cards.

9
00:00:25.170 --> 00:00:27.060
You have to have it authenticated

10
00:00:27.060 --> 00:00:29.220
and graded by one of
the grading companies,

11
00:00:29.220 --> 00:00:31.740
and that's how you get the
max value for your card

12
00:00:31.740 --> 00:00:33.270
if you sell in an auction house.

13
00:00:33.270 --> 00:00:34.290
You would've to wait a year

14
00:00:34.290 --> 00:00:36.590
and a half just to get
a card back from some

15
00:00:36.590 --> 00:00:38.580
of these companies while paying hundreds

16
00:00:38.580 --> 00:00:41.100
of dollars per card on
some of these cards.

17
00:00:41.100 --> 00:00:43.020
Humans were just looking at these cards

18
00:00:43.020 --> 00:00:45.480
that could be worth a
million dollars in a 10

19
00:00:45.480 --> 00:00:47.160
and a hundred thousand dollars in a nine

20
00:00:47.160 --> 00:00:49.500
and giving it a grade
based on their opinion.

21
00:00:49.500 --> 00:00:53.310
It's a human bias, a human judgment.

22
00:00:53.310 --> 00:00:55.080
And humans, as we all do,

23
00:00:55.080 --> 00:00:56.980
myself included, make a lot of errors.

24
00:01:01.200 --> 00:01:04.500
- What we generate from
this laser scanner is

25
00:01:04.500 --> 00:01:06.060
firstly luminance image.

26
00:01:06.060 --> 00:01:09.060
We can see the contrast
here, which is very useful

27
00:01:09.060 --> 00:01:11.490
for identifying defects.

28
00:01:11.490 --> 00:01:13.320
However, it's not enough.

29
00:01:13.320 --> 00:01:18.320
So what we need to add to this
is the height map component.

30
00:01:18.600 --> 00:01:21.720
So what we do with these
images is we feed both

31
00:01:21.720 --> 00:01:23.220
of these images into our AI.

32
00:01:23.220 --> 00:01:25.410
Our AI will not run without the height

33
00:01:25.410 --> 00:01:26.700
map and the luminance image.

34
00:01:26.700 --> 00:01:30.240
So an awesome feature of the
height map is that we can,

35
00:01:30.240 --> 00:01:33.300
within 25 microns, identify

36
00:01:33.300 --> 00:01:37.800
spatial differences along
the X and the Y and the Z.

37
00:01:37.800 --> 00:01:39.600
So that means like depth

38
00:01:39.600 --> 00:01:42.630
and along the surface area of the card.

39
00:01:42.630 --> 00:01:46.740
This data alone is enough
to authenticate cards.

40
00:01:46.740 --> 00:01:51.330
This is a Pokemon card, so you
see all the scratches here.

41
00:01:51.330 --> 00:01:52.650
Scratches, scratches, scratches,

42
00:01:52.650 --> 00:01:54.420
all these scratches all over the card.

43
00:01:54.420 --> 00:01:59.340
And we have this, these little
blobs, right, that appear,

44
00:01:59.340 --> 00:02:01.200
these blobs are auto-generated.

45
00:02:01.200 --> 00:02:02.760
If a human came in here and reviewed it,

46
00:02:02.760 --> 00:02:04.410
they could see like this
blob could be better.

47
00:02:04.410 --> 00:02:05.820
We can make it bigger and nicer.

48
00:02:05.820 --> 00:02:07.410
And when we feed the AI over

49
00:02:07.410 --> 00:02:09.480
and over again, one of the pieces of work

50
00:02:09.480 --> 00:02:11.010
that we do is fine tune this.

51
00:02:11.010 --> 00:02:13.770
We constantly do reviews
of the AI to improve it.

52
00:02:13.770 --> 00:02:17.100
And currently this is a card that is very,

53
00:02:17.100 --> 00:02:18.720
there's a lot of defects.

54
00:02:18.720 --> 00:02:20.740
So on such cards we have

55
00:02:21.690 --> 00:02:25.290
basically accuracy like
improvement opportunities.

56
00:02:25.290 --> 00:02:27.420
And also for new cards
like that we just released.

57
00:02:27.420 --> 00:02:31.410
Overall we have around a 95%
accuracy for the overall grade,

58
00:02:31.410 --> 00:02:33.303
which is we're pretty happy with.

59
00:02:38.700 --> 00:02:39.750
- There's no bias involved at all.

60
00:02:39.750 --> 00:02:42.300
There's no human emotion,
there's no human errors.

61
00:02:42.300 --> 00:02:44.700
This is something like this
grade in a card should be

62
00:02:44.700 --> 00:02:46.170
completely scientific.

63
00:02:46.170 --> 00:02:50.850
That's kind of why people are
sharing, going, going with AGS

64
00:02:50.850 --> 00:02:53.010
because they believe and they trust AI

65
00:02:53.010 --> 00:02:55.920
or computers to grade
their cards over humans.

66
00:02:55.920 --> 00:02:58.980
I would think it's actually
more time consuming right now

67
00:02:58.980 --> 00:03:02.970
for us because one, we
have eight subgrades,

68
00:03:02.970 --> 00:03:04.440
which no other company has.

69
00:03:04.440 --> 00:03:05.790
So we have four subgrades on the front

70
00:03:05.790 --> 00:03:09.660
that's surface, centering,
edges and corners.

71
00:03:09.660 --> 00:03:11.580
And the same way we have those
four grades on the front,

72
00:03:11.580 --> 00:03:13.620
we also have the four grades
on the back of the card.

73
00:03:13.620 --> 00:03:17.640
We also have to upload the
scans, the imaging to our server.

74
00:03:17.640 --> 00:03:19.560
We have to process the results.

75
00:03:19.560 --> 00:03:21.810
So there's a lot more
steps involved compared

76
00:03:21.810 --> 00:03:22.950
to a traditional grading company

77
00:03:22.950 --> 00:03:25.140
where they have one human
looking at the card.

78
00:03:25.140 --> 00:03:27.060
Technically our process
is much more tedious

79
00:03:27.060 --> 00:03:28.680
and much more time consuming.

80
00:03:28.680 --> 00:03:30.540
But I think as our technology scales,

81
00:03:30.540 --> 00:03:32.490
I mean continue investing
more and more time

82
00:03:32.490 --> 00:03:34.350
and resources into our tech,

83
00:03:34.350 --> 00:03:37.200
our automation will make our
grading process even faster

84
00:03:37.200 --> 00:03:40.560
than the human grading process
that other companies have.

85
00:03:40.560 --> 00:03:42.810
We see a lot of people, very
smart collectors, they send it

86
00:03:42.810 --> 00:03:45.300
to us 'cause they know they'll
get their cards back quickly

87
00:03:45.300 --> 00:03:46.350
and actually they sell their cards

88
00:03:46.350 --> 00:03:47.610
for much more than someone that sent it

89
00:03:47.610 --> 00:03:48.780
to a different company.

90
00:03:48.780 --> 00:03:50.100
But gets it back two, three months,

91
00:03:50.100 --> 00:03:52.350
even if it sells for 10, 15% more.

92
00:03:52.350 --> 00:03:56.010
For a process that can cost you

93
00:03:56.010 --> 00:03:57.750
or make you hundreds of thousand dollars

94
00:03:57.750 --> 00:04:00.150
or more, typically thousands
of dollars in the traditional,

95
00:04:00.150 --> 00:04:02.390
you know, for modern cards
that needs to be assigned.

96
00:04:02.390 --> 00:04:06.933
It needs to be a standard
for every grader to follow.

97
00:04:08.435 --> 00:04:11.852
(chill electronic music)