DKC Atlas Forum

[Slyhawks]

Hello Simion.

Sorry if it was posted before (I can't understand everything, I'm French :S), but i've question.
With DKCRE, can I modify or add tiles? If we can't do this, what's the utilitie of the extract?
Thanks

[Simion32]

Hello, and welcome to DKC Atlas, slyhawk!

You cannot add or modify tiles because the editor can currently only edit what the game already has in there.

Of course, there probably won't be a graphics/tiles editing feature made due to the graphics being compressed in the ROM. But I will be working on the ability to resize levels and fully delete objects (among other enhancements) in version 0.0.7.0 of the editor.

Extraction Mode is there so that you can get maps/tiles/sprites out of the game. It's not intended for putting anything back in.

[Slyhawks]

Really thanks for your answer

I've an other question, i don't understand why, but, if I change the place of the barrel (the barrel of end of a secret place that bring you back on the map), In the game, it don't change.. I'm the only ho have this bug? ^^' (Sorry if my english isn't perfect :S )


(Oh and, good job for DKCRE, it's a grat editor )

[Super Luigi!]

May I say something? Although I don't know the answer to your question (Perhaps you're forgetting to save changes?), your English is much better than several native speakers I know, so don't worry! Enjoy your first banana!

[Simion32]

The actual locations of bonus returns and their barrels are two different things.

The current release can only edit the barrel objects, not the entry locations. The same goes for the checkpoint barrels, too.


Don't worry - the very next update of DKCRE will have a way to modify these. It's already on my to-do list.

( and, Thanks! )

[Slyhawks]

Ok thanks!

[Blue Banana]

This editor has helped me make soome various hacks, some flawed, some exceeded.

[Simion32]

Spoiler!

CRITICAL COMPONENT FAILURE.
GRAPHICS CARD OUT OF ORDER!!
DEVELOPMENT FULLY HALTED.

The fan on my GPU has just died. (The card itself survived, though.)


I'm posting this notification from my old clunker machine.

[Simion32]

My father happened to have some compressed oxygen (used as pure air to remove dust from inside part of fan) and some thermal compound handy, so...

...for a little while (however long the fan lasts) I'm back in business!

I still plan on getting a new card in case this one permanently bites the dust.

[Crammalamma]

Why not extend the tooltips of the used sprites like you did with the unused ones?

[Simion32]

>>> Level Generator Beta Test #1 (image) <<<

I'm trying out the first bits and pieces of the Level Tile Auto-Filler. The results you see above were obtained from just picking tiles one-by-one for each row based on scanned matching data. Most of what is done so far is just data collection.

I'm very surprised that I got anywhere with this, because it's an extremely hard problem to tackle.

It still needs a lot more coding and algorithm design work, but I'm getting there.

[Qyzbud]

Very exciting to actually see progress on this feature.

I eagerly anticipate your future breakthroughs.

[Super Luigi!]

I agree with the Chief. You're doing just wonderful, Simion32! Keep pounding away at that infernal contraption!

[VideoViking]

A lot of flipped tiles in the beta image. But still, looking good.

[Simion32]

The generator actually has rotated tile compatibility as well as ordinary horizontal and vertical flips.

The fact that the image was made in a totally random manner explains all the odd vertically flipped tiles.

[Simion32]

>>> Level Generator Beta Test #2 (image) <<<

This is a broken test of the fully-random level generation mode.

It's having a lot of trouble not getting stuck, so I added in some lines of code to force it to give up on a tile if it gets stuck.

The result of what it was able to do, is this image.

[Super Luigi!]

After intensely comparing the two images side-by-side for a lengthy amount of time, I have formed my conclusion: I have no idea what I'm looking at.

But all jokes aside, the second picture does indeed look spacier, and somewhat cleaner as a result, smaller, and last but not least, whiter! The best part of the whole scene absolutely has to be the upside-down palm tree. The long, steep cliff on the right takes a close second though.

[Simion32]

I'm having to switch strategies, as it appears my current algorithm just can't satisfy the complex rules needed to generate a "level".

I could use the brute-force "strings-on-poles" algorithm (that is, my original visualization of the problem) to do it, but that would take enormous amounts of RAM (80MB+ for a 16x16 tiles grid) and would likely be quite a bit slower.

It's going to take a miracle and a half to find something elegant that doesn't consume memory like a predator.

[Simion32]

Thought I'd post a quick update on the generator algorithm:

Although the scanner part is perfect, the generation is very difficult to get correct.

I'm now working on a Cellular Automata design document which will specify a way of creating the grid of possible levels, that essentially plays the "game of life" with the grid until only 100% valid solutions exist.

Sadly, the fitness function (it figures out whether a "cell" lives or dies) is extremely expensive to calculate, and I'm trying to find a fast way to calculate it, but this is proving to be just as difficult as the brute-force method.

The best I've got so far is some "fast boolean math" methods that can be used, but I've not come up with a very efficient way of computing the whole fitness function itself yet.

Whatever I do come up with is going to be quite unorthodox, I can say that.

The number of computations goes something like this:
time_taken = number_of_level_grid_tiles * (number_of_tiles_in_tileset * 8) * (9_spaces_to_evaluate_per_step) * calculations

[Simion32]

I've found a way to compute the data required for the fitness function, and found out a way to reduce its problem size below the usual n^2 complexity, so that's great.

Now it's just down to some very tricky high-performance graph processing code, to get everything going...

[Simion32]

Progress Update: Ah, what a beautiful tapestry of tiles!!

The Pattern Scanner is Done!!
(This is only Part #1 of the algorithm - hold your excitement for a bit )



The full image is actually 1312 x 16384, so I won't be uploading it here - but you can see some of the results of the pattern scanner in that section of the image.

Quite a piece of art, wouldn't you say?

Edit Update:

The Jungles graph started out at 33,102,107 unique edges.

Number of edges in the fully processed Jungles matching graph: 218,597
With a maximum number of edges per subgraph: 1,714
(that is, each subgraph corresponds to a virtual tile, of which there are 3,299)

So in the end, there are FAR LESS than the extreme 33 million edges in the unprocessed graph.

Processing time takes a few minutes but should be improvable with a faster sort algorithm.

[Super Luigi!]

I'm not going to lie; the tapestry is truly terrific! I'll stay calm for now, Simion32, but please tell us when we can unload our joyous praises! See you again, someday, sometime, somewhere!

[Simion32]

Progress Update: Very Preliminary Pattern Filler WORKING


(click to view if whole image is cut off)

This may only work for the base case so far, but what a rousing success... with such a simple algorithm, too.

Starting from a random fill (left) it gradually arrives at the solution (on the far right).

I'm not sure how well this will apply to the actual DKC tile data, but I think i'm actually going somewhere with this.

(You can get a little bit excited now! )

[Simion32]

Several more runs of the algorithm shows that even additional random trials do not error out...

...they ALL have so far resulted in valid solutions!

[Super Luigi!]

Yay! I do enjoy watching unsorted messes form together to create true progressive art. Well done, Simion32!

[Simion32]

Progress Update: Better Yet, Here's A Batch of Random "Levels" For You to Enjoy!











>>> When Taken Up To 11 (CLICK HERE) <<<

WOW.

In memory of Perlin Noise (which can optionally be used here) and to go with a dream I had, ...

I hereby dub this the Perlin Hunter Algorithm.

EDIT: There's some sort of probability bug in the code that causes the algorithm to favor certain solutions over others. I'll be working on a fix for this, but it may take some time.

[edevore]

my head hurts now, that image is too overwhelming for my brain to process its like those psychedelic images that trick your brain into thinking the picture is melting or whatnot.... anyways nice update Simion I will be looking forward to trying the new generation system if you release it in your upcoming update

[Simion32]

It appears that when the complexity goes up, the default algorithm does just fine.

It is in fact possible to provide "focal point" parameters to further randomize/control the output, but this causes the algorithm to take more time than usual (due to increased complexity).

For level generation, the default focus parameters can all be set to zero, since it has little important effect when the complexity of the tileset increases beyond a mere handful of tiles.

Small Example, 10 Tiles (22 x 22):


Large Example, 10 Tiles:
256 x 256 Example

The results here are very promising, and I look forward to seeing Perlin Noise in action on this test tileset (I'm not quite at the point of being able to do actual DKC levels yet).

Furthermore, I think it will all be extremely fast in the end - I just need to buffer all those SuperRandom(); calls instead of calling the function itself the whole time. Choosing the tiles requires some randomness, and with a slow random value chooser, it's really a huge bulk of the time spent.

[Simion32]

There may well be a way to speed the code up even further than it already goes, by heuristically analyzing which parts of the solution are "hot" or "cold" and ignoring areas that have been processed or would be totally random garbage.

The heatmap of the algorithm in action looks like The Matrix running itself on the grid with a randomized "blob" that gradually shrinks into the bottom left corner.

I've got to get an animated GIF of this thing up sometime.

[Qyzbud]

The heatmap of the algorithm in action looks like The Matrix running itself on the grid with a randomized "blob" that gradually shrinks into the bottom left corner.

This I gotta see.

At any rate, it sounds/looks like good progress is being made.

Seeing all these randomly generated patterns with symmetrical sections, harmonious edges, etc. has me a bit excited; I know you're not at the stage yet where level tiles will magically, intelligently fall into place around a physmap, but it's starting to feel like a very real prospect for the not-so-distant future.

Keep up the excellent work, you great ape.

[Simion32]

A quick update: Perlin Noise, or any noise for that matter, doesn't have much effect on the overall output because the tiles become white-noise randomized as soon as the algorithm begins iteration #2.

The only real requirement in that respect is that the gird starts out as random garbage, so that it has something to cause the algorithm to do actual work.

To apply Perlin Noise here, will require more than just a simple random function substitution.

[Simion32]

Animations of Perlin Hunter (Normal Mode):



(Shrinking Mode, Unfixed):


The animation is taken after each base iteration of the algorithm.

[VideoViking]

This looks like something out of a Jurassic Park movie. Fractals.

[Simion32]

Progress Update: New Techniques Required!
Introducing: Perlin Hunter v2.0

I've been hard at work on Perlin Hunter 2, which is intended to be capable of generating complex DKC levels, but it's extremely difficult to do this. I combined the original Perlin Hunter algorithm with a cellular automation approach, in order to more easily allow the grid to converge on a solution (which plain doesn't happen in v1.0, at all).

The algorithm is extremely stubborn to work with - and I've only been able to get large contiguous "chunks" of terrain to appear, so far.

No full levels have been generated yet.

Here's a few samples of the progress that has been made on the algorithm...

Perlin Hunter v1.0 Fails using Jungle Tileset
Image 0, Image 1, Image 2, Image 3,
Image 4, Image 5, Image 6, Image 7

A more detailed look at the generator, with red lines showing non-matches*
(*according to the algorithm, anyway :P )

A few tweaks got better results after I had a lucky test run.
Looks awesome, but nowhere near complete ;)

[Simion32]

Back to Main DKCRE Development (In The Meantime, Anyway!)

I've thought out a new technique which would be better than PH2, but unfortunately it will take some time before I am able to develop it.

Development on Perlin Hunter 3 has been postponed. DKCRE development resumes.

I wasn't able to find much of a method to keep development going, so, for the moment, I'm going back to RE development.

[Simion32]

Progress Update: Some New Optimizations for Perlin Hunter 3 FOUND

I've finally cracked the shell of huge computation times with some very clever optimization designs.

What I've designed here (not coded yet) is a fast boolean processing algorithm based on my previous research, with a special twist: The information of where to copy and/or manipulate bit information is stored in code that is compiled at runtime and then executed - this means that information can stay inside the CPU's code cache, without disturbing the data cache as the algorithm works on the data.

Additionally, by reducing a tile bit table down to its group IDs first, it is possible to reduce the NxN bit matrix "data smash" operation. That would would ordinarily take ages to compute per tile (one scan would take minutes), but this trick allows me to turn it into only 2 linear operations instead (which is on the order of sub-second scanning times).

Altogether, this means that, if I program all of it *VERY* carefully, we might be seeing a fully working version of the algorithm in the near future.

[Markster]

That's genius Simion32.

[Cyclone]

^ +

I'm not sure if this has been suggested but what about having the option to mirror a level. Playing the level backwards would be a new challenge. The controls and tiles would have to be hacked of course.

[Simion32]

Not really possible with DKCRE due to most objects only facing one direction. For DELTA, though, I can implement it as a Mirror Mode by rendering everything reversed and swapping controls.

You can achieve mirroring the tiles already - selection mode, ctrl+A, ctrl+F. I used this feature to create that Carnage Cart Mines video.

[Simion32]

NitroGUI Save & Open Dialog is COMPLETED!!!


After all this time I've finally had a chance to get some work done. And it's a big one - the file save and open dialogs are COMPLETE!!!

Now, you can open and most-of-the-way manage your files in epic NitroGUI style.

nitrogui_open_dlg_demo.png (74.1 KiB) Viewed 363324 times

By the way, copy and paste works fully here, as if you were in Windows Explorer.

[Super Luigi!]

Well done, Simion32! Although I am unable to grasp the true concept of NitroGUI file support, I appreciate the feature and your hard work that went behind implementing it. All I can give you is my praise and a banana, but I'll always support your tireless efforts!

[Markster]

Great job Simion32, it looks great!

--
I'm requesting a feature for NitroGUI, I love to snap windows to the side using aero snap (I'm on Win8.1, it's a feature on Windows 7 and newer), but you can't use the aero snap feature with NitroGUI windows, can you fix that, FormatFactory has a custom interface that can snap w/ Win7+.

[Simion32]

Sorry, I can't. It's because the program runs on top of DirectX, and therefore I can't do anything about whether it's screen-shottable. FormatFactory is likely a WinAPI interface, whereas NitroGUI has been made completely from scratch (like a game engine).

The program now handles its own screenshots via print screen key, though, so it shouldn't be too hard to just use that.

[Kingizor]

It's related to window positioning, not screenshots.

Windows 7+ have a feature where the edges of windows will "snap" to the edge of the screen when you resize the borders. It's also possible to resize and reposition using the Windows key in combination with the arrow keys.

Win + Up = Maximize
Win + Left = Fill the left half of the screen
Win + Right = Fill the right half of the screen
Win + Down = Minimize (Only this one works with DKCRE)

There are programs that replicate this behaviour on XP.

[Simion32]

In that case, it is entirely possible.

Perhaps I can do that later on, but it's not on my priority list at the moment. I can add the relevant windows key shortcuts relatively quickly, though (thanks to actually being able to somewhat detect winkey).

[Simion32]

K. Rool Running Frames Bug Cause Report

This post is to report on the K. Rool Running Frames interpretation bug so that we are not in doubt about what caused it ever again.

It's quite simple, actually (and, the bug actually happened in multiple places outside the sprite renderer):

int pc_offset = ((rom_data_offset - 0xC00000) & 0xFFFFFF); //WRONG
int pc_offset = (rom_data_offset & 0x3FFFFF); //CORRECT

The bug happens when the ROM's data offset already has some of the high bits subtracted out or missing. The SNES doesn't care, but a PC program sure would.

Furthermore, I believe this bug was carried over from someone's ROM address calculation advice, suggesting to subtract 0xC00000, which is in fact not the most correct (and bug-proofed) way of doing things.

I don't know why I never wrote this down, but here it is, for future reference.

[Simion32]

Major Features Update:
DKC Resource Editor v0.0.6.9

>>> DOWNLOAD LATEST VERSION HERE <<<

There are many updates, which you can check for in the readme.html file, but here are the most notable updates:

Spoiler!

Vertical Levels Lock - Unfortunately, the vertical levels are too buggy to reliably edit, but can still be viewed. I've locked the levels from being modified to prevent bugs from appearing in those levels (and preventing them from being impassable by players).

Significant Updates in This Version:
Bring Back the Sprite Viewer Mode!! (Special Request by Cyclone)
Improved Controls - The controls now use the numpad keys for navigation and etc.
Jump to the Palette Editor - You can click the palette bar to go directly to a palette in the Palette Editor mode!
Sprite Names - With more INI magic, RE is able to use defined ranges of sprites to show sprite names.
Sprite Zoom - You are able to zoom in and out of the displayed sprite.
Copy Sprite Info - You can use Ctrl+C or Ctrl+X to copy a sprite's text information to the windows clipboard.
Copy or Save Bitmap with Key Shortcuts - You can save out a PNG to disk. You can also copy a 24bit bitmap to the clipboard.
Go to Sprite Number - By holding Enter and typing 3 hex digits.
INI-Based Sprite-to-Palette Table - Now supported!!

Start Points, Entrance Styles, and Smart Checkpoint Barrels
JSON Layout Generator (with Bananas), & Banana Group Image Extraction

Loading and Saving, ROMs/Patches, and Data Formats
The NGUI save and open dialogs.
Supporting multiple ROM versions via GameConfig INI Files
Save to a Patching Format by Default & Minimize Patching Confusion
Original ROM(s) Backup Cache Folder +Restore Function, ...
......Silent Conversion Between ROM/Patch Formats

Making Things Easier & Interface Revisions
Usage Meter - Placed in status bar to the far right, next to the sizing grip.
Make the Mode Menu More Obvious - Attention Getter Button!!
Logarithmic Smooth Scrolling in Level Editor
Arrow Keys Smooth Scroll if Nothing is Selected - because RE should have had this from the beginning.

Improvements
DKCRE now comes with a handful of File Type Icons to make opening frequently opened files easier.
DKCRE is now optimized with dirty rectangles. The IU will be able to run much faster and on a wider range of computers with this improvement!!
All of the stability and timing improvements from DELTA. You can now run multiple copies of DKCRE without lag problems!!

Bug Fixes
All of the input bugfixes and updates from DELTA. Fixes menu unresponsiveness/lags.
Further NGUI fixes to completely exterminate all input lag. NGUI is now single threaded. Mostly the lag was happening after heavy processing. Bug also affects DELTA.

[Simion32]

Kingizor has just pointed out a crash bug that happens when you hover over an object that's at the top of the viewing box.

As a workaround until I can provide a last-minute patch, just use it in maximized mode to avoid the crash bug.

[VideoViking]

ZSNES functionality is not working. Pushing F5 or F4 gives me nothing. This is even with the correct paths.

[Simion32]

That's really strange. Another bug to add to the last minute patching list.