Destroying Big Brain Academy With Image Recognition

I'm using image recognition to beat two of the best minigames in Big Brain Academy on the Nintendo DS, Pathfinder and Matchmaker. The goal is to beat them as fast as possible, without doing any mistakes on the puzzles.

0:00 Intro
0:39 Minigame 1: Pathfinder
6:37 Minigame 2: Matchmaker
11:08 Outro

For those of you who read all of this... You are amazing :)

[guy]

I always found it a bit unfair that the clock would continue while the game was checking your answer. What are we supposed to do? Solve the next problem before it's even on screen???

[TheWorldsLargestOven]

I love seeing people destroy the point system of games.

[pfqniet]

There are definitely improvements that can be made. It shouldn't take multiple seconds for an algorithm to find these solutions, and I'm definitely concerned that your solution is affected by "similar" images or performance is so drastically affected by the images being in colour versus greyscale. This is cool but a video optimising these and going into detail on algorithms would be much more interesting.

[behrmich]

Awesome video. I’d love to see Rythm Paradise if that’s possible. It drove me nuts as a child

[AntjedePantje]

Fun video! I love watching people break games 😂 Can't wait to see what DS games you destroy next, haha

[violet_broregarde]

Does Pathfinder really need a whole-ass computer vision solution? The vertical lines and horizontal connections are in the same place every time, and they're recognizable because of black pixels. Then to find the horizontal lines, you just look between the vertical lines for black pixels. Keep your x position between 2 lanes and look downwards and where you see an RGB(0, 0, 0) that's either a skull or a horizontal line. You distinguish between them by continuing to look downwards; if you see an RGB(255, 255, 255) it's a skull. You stay on the skull until you see a pixel that's neither black nor white.

Once you know how many vertical lines there are, you know exactly where the animals might be. Like at a 3-line puzzle, you know the animals' x-coordinates are gonna be at a, b, and c (they have the same y coordinate every time). If you see a gray pixel at a+10, y+30 it's a koala. If you see a brown pixel at a+14, y+20 it's a bear. etc.

[LrdSqk]

Great video, personally I'd love to see the explanations for how you fixed each problem within the code. You do a great job of identifying what the problems are, but then suddenly they're just fixed

[raaaaaaayden]

Great video! Congratulations on reaching your sub goal :D

[louiscordes]

the sound in the x2 speed section makes me feel like im in dark souls 2

[EeveeRealSenpai]

Hope you reach your sub goal!
W YouTube recommendation

[htspencer]

A tutorial on how you achieve this would be amazing for people who want to do their own image processing gaming!

[bumbobrumbo]

Honestly loved the video! But the repeated game sound effects were really annoying at points

[megaclpb]

for matchmaker you could instead go through each image in the grid and store it's position in a hash table and when there's already something in the hash table you click on the current position and the position in the hash table at that index

[HimmDawg]

I'd have loved to see how far you can go on hard mode. It feels like 1000g are just in reach (if that's even possible). And from this goal onwards, you could show, how you improve the existing algorithm, showing the areas where it performs too slow. Still a cool video though :)

[bs_blackscout]

Great stuff! But I can't help but notice mouth clicks and sounds all over the voice over.

Maybe adjust the mic position a little bit?

[Starxy-jh]

Bro for your work you deserve more subscrive4

[Tyron-sdhg]

Train the switch version and try to get a perfect score multitask AI

[KoffeeB]

Damn, only 700 subs, as underrated as my meme channel