AI generated textures

.gitignore

@@ -0,0 +1,12 @@
+new
+.idea
+# Byte-compiled / optimized / DLL files
+__pycache__/
+*.py[cod]
+*$py.class
+TEMP/
+*.pth
+
+input
+output
+add

README.md

@@ -3,57 +3,25 @@
  <img width=420px height=150px src="https://upload.wikimedia.org/wikipedia/commons/a/af/Patapon_2_logo.png" alt="Patapon 2 logo"></a>
 </p>
 
-<h3 align="center">Patapon 2 HD</h3>
+<h3 align="center">Patapon 2 HD AI</h3>
 
 <div align="center">
 
-[![Contributors](https://img.shields.io/github/contributors/shockturtle/Patapon2HD)](https://github.com/shockturtle/Patapon2HD/graphs/contributors)
-[![Stargazers](https://img.shields.io/github/stars/shockturtle/Patapon2HD)](https://github.com/shockturtle/Patapon2HD/stargazers)
-[![Issues](https://img.shields.io/github/issues/shockturtle/Patapon2HD.svg)](https://github.com/shockturtle/Patapon2HD/issues)
-[![Pull Requests](https://img.shields.io/github/issues-pr/shockturtle/Patapon2HD.svg)](https://github.com/shockturtle/Patapon2HD/pulls)
 
 </div>
 
 ---
 
-Side Project for HD Patapon 2 textures.
-
-All textures are done entirely by me unless comments in textures.ini state otherwise.
-
-## 🎈 Usage <a name="usage"></a>
-
-To install, enable Texture Replacement in PPSSPP dev menu (Settings->Tools->Developer Tools).
-
-Put both the textures folder and textures.ini files in `PPSSPP/memstick/PSP/TEXTURES/%GAME_ID%`
-
-If the textures folder doesnt exist, make it.
-
-|  Region | Game ID     |
-| ------: | ----------- |
-|      EU | `UCES01177` |
-|      NA | `UCUS98732` |
-|      JP | `UCJS10089` |
-|      AS | `UCAS40239` |
-
-
-## 🎉 Acknowledgements <a name = "acknowledgement"></a>
-BIG Thank you to
-
-- [owocek](https://github.com/owocekTV) (Motivation)
-
-- Xandis/Gat235 (Every single bird and javelin)
-
-- Iracy (Materials and other items)
-
-- [RinCasi](https://twitter.com/I26mc) (Various symbols and high-quality resources)
-
-- [efonte](https://github.com/efonte) (AI Upscaling)
-
-
-Special Thanks to
-
-- [WallSoGB](https://github.com/WallSoGB) (Leader of the [Patapon3Textures](https://github.com/WallSoGB/Patapon3Textures) project)
+AI upscaled Patapon 2 textures.
 
+Contains all exists textures from [Patapon 2 HD](https://github.com/shockturtle/Patapon2HD) repository
 
+If you want generate missing textures, it very easy.
 
+1) Enable saving new nextures in PPSSPP dev settings
+2) Copy all yours textures from `new` to `ai_upscale_workdir/input`
+3) Press `ai_upscale_workdir/generate_ai_textures.bat` and select models
+   * `1_4x_FatalPixels_340000_G_4x_Fatality_01_265000_G_interp_05` works very good, but sometimes `1_4x_Loyaldk-SuperPony_500000_V2.0` work better
+4) After upscale finish first press `copy.bat`, after `paste.bat`
 
+And it's done, you generated new textures

ai_upscale_workdir/README.md

@@ -0,0 +1,42 @@
+# Fork of [BlueAmulet's fork](https://github.com/BlueAmulet/ESRGAN) of [ESRGAN by Xinntao](https://github.com/xinntao/ESRGAN)
+
+This fork ports features over from my ESRGAN-Bot repository and adds a few more. It natively allows:
+
+* In-memory splitting/merging functionality (fully seamless, recently revamped for the third time and no longer requires tile size)
+* Seamless texture preservation (both tiled and mirrored)
+* Model chaining
+* Transparency preservation (3 different modes)
+* 1-bit transparency support (with half transparency as well)
+* Both new-arch and old-arch models
+* Regular and SPSR models, of any scale or internal model settings
+* On-the-fly interpolation
+
+**Tile size was recently removed! It is no longer needed for split/merge functionality!**
+
+To set your textures to seamless, use the `--seamless` flag. For regular tiled seamless mode, use `--seamless tile`. For mirrored seamless mode, use `--seamless mirror`. You can also add pixel-replication padding using `--seamless replicate` and alpha padding using `--seamless alpha_pad`.
+
+To chain models, simply put one model name after another with a `>` in between (you can also use `+` if using bash to avoid issues), such as `1xDeJpeg.pth>4xESRGAN.pth` **note: To use model chaining, model names must be the complete full name without the path included, and the models must be present in your `/models` folder. You can still use full model paths to upscale with a single model.**
+
+For on-the-fly interpolation, you use this syntax: `<model1_name>:<##>&<model2_name>:<##>`, where the model name is the path to the model and ## is the numerical percentage to interpolate by. For example, `model1:50&model2:50` would interpolate model1 and model2 by 50 each. The numbers should add up to 100. If you have trouble using `:` or `&`, either try putting the interpolation string in quotes or use `@` or `|` respectively (`"model1@50|model2@50"`).
+
+To use 1 bit binary alpha transparency, set the `--binary-alpha` flag to True. When using `--binary-alpha` transparency, provide the optional `--alpha-threshold` to specify the alpha transparency threshold. 1 bit binary transparency is useful when upscaling images that require that the end result has 1 bit transparency, e.g. PSX games. If you want to include half transparency, use `--ternary-alpha` instead, which allows you to set the `--alpha-boundary-offset` threshold.
+
+The default alpha mode is now 0 (ignore alpha). There are also now 3 other modes to choose from:
+
+* `--alpha-mode 1`: Fills the alpha channel with both white and black and extracts the difference from each result.
+* `--alpha-mode 2`: Upscales the alpha channel by itself, as a fake 3 channel image (The IEU way) then combines with result.
+* `--alpha-mode 3`: Shifts the channels so that it upscales the alpha channel along with other regular channels then combines with result.
+
+To process images in reverse order, use `--reverse`. If needed, you can also skip existing files by using `--skip-existing`.
+
+If you're upscaling images of the same size, you can do `--cache-max-split-depth` to only calculate the automatic tile size once to improve performance.
+
+Examples:
+
+* `python upscale.py 4xBox.pth --seamless tile`
+* `python upscale.py 1xSSAntiAlias9x.pth>4xBox.pth`
+* `python upscale.py 4xBox.pth --binary-alpha --alpha-threshold .2`
+* `python upscale.py /models/4xBox.pth`
+* `python upscale.py "1x_model1.pth@50|1x_model2.pth@50>2x_model3.pth"`
+
+If you want a GUI for ESRGAN and if you're on Windows, check out [Cupscale](https://github.com/n00mkrad/cupscale/). It implements most of this fork's features as well as other utilities around it.

ai_upscale_workdir/generate_ai_textures.bat

@@ -0,0 +1,2 @@
+py generate_ai_textures.py
+PAUSE

ai_upscale_workdir/generate_ai_textures.py

@@ -0,0 +1,73 @@
+import os
+import shutil
+from pathlib import Path
+
+import upscale
+
+processAlpha = None
+
+inputVal=input("Need process alpha channel ? (Y,n)")
+if inputVal.lower() == "n":
+    processAlpha = False
+elif inputVal.lower() == "y" or not inputVal:
+    processAlpha = True
+else:
+    print("Error input")
+    exit(0)
+
+print("Select upscale engine, or if you want use multiple, write multiple numbers like 123")
+
+models = os.listdir("models")
+
+if len(models) == 0:
+    print("Download models and place in it models before use")
+    exit(0)
+
+for i in range(len(models)):
+    print(str(i + 1) + ". " + models[i])
+
+select = input("Select: ")
+if not select.isdigit():
+    print("Write digit")
+    exit(0)
+
+selectedModels = []
+
+for i in select:
+    num = int(i) - 1
+    if num < 0 or num >= len(models):
+        print("Number bigger of models size or lower 0")
+        exit(0)
+    else:
+        selectedModels.append(models[num])
+
+print("Start: " + "    ->     ".join(selectedModels))
+
+
+def deleteFolder(path: str):
+    if os.path.exists(path):
+        shutil.rmtree(path)
+
+
+tempFolder = "TEMP"
+finalOutputFolder = "output"
+
+deleteFolder(tempFolder)
+
+inputFolder = "input"
+
+for i in range(len(selectedModels)):
+    model = selectedModels[i]
+    modelResultFolder = "TEMP" + os.sep + str(i + 1) + "(" + model[0:20] + ")"
+
+    upscale.Upscale(
+        model=model,
+        input=Path(inputFolder),
+        output=Path(modelResultFolder),
+        alpha_mode=upscale.AlphaOptions.alpha_separately if processAlpha else upscale.AlphaOptions.no_alpha
+    ).run()
+
+    inputFolder = modelResultFolder
+
+deleteFolder(finalOutputFolder)
+shutil.copytree(inputFolder, finalOutputFolder)

ai_upscale_workdir/models/README.md

@@ -0,0 +1,7 @@
+Im use next models, you can find it in the internet:
+* [4x_FatalPixels_340000_G](https://de-next.owncube.com/index.php/s/gPjswdm6gCegQdz)
+* [4x_Loyaldk-SuperPony_500000_V2.0](https://mega.nz/file/IBhElRYJ#C_lVc8Vr2q8X6hEltJKD6I63U7oL_FDLeJgPh72AZmg)
+* [1x_NoiseTonerV1_110000_G](https://drive.google.com/drive/u/0/folders/12j6BvuR7SU48ui9DHjBPB1vzPpQri9bD)
+* [1x_Fatality_DeBlur_270000_G](https://de-next.owncube.com/index.php/s/aAojXwLTPZto8rP)
+
+Very cool works Pony model, but some time combine fatalpixel->deblur->noiseToner give also very cool results

ai_upscale_workdir/requirements.txt

@@ -0,0 +1,5 @@
+opencv_python
+numpy
+torch
+rich
+typer