[Bug] Conversion fails on Grayscale (PseudoClass Gray) JPEGs

[MarcG2]

Please carefully review each item in the checklist below | 请认真检查以下清单中的每一项 | 以下のチェックリストの各項目を注意深く確認してください

• Searched and didn't find a similar issue | 已经搜索过，没有发现类似issue | 類似の問題が見つかりませんでした
• Searched documentation and didn't find relevant content | 已经搜索过文档，没有发现相关内容 | ドキュメントを検索して関連する内容が見つかりませんでした
• Tried with the latest version and the issue still exists | 已经尝试使用过最新版，问题依旧存在 | 最新バージョンを試しましたが問題は解消されませんでした

Software Version | 软件版本 | ソフトウェアバージョン

1.1.5

Operating System | 操作系统 | オペレーティングシステム

Windows x64

System Version | 系统版本 | システムバージョン

Windows 10

Describe the bug | 描述错误 | BUGの説明

Final2x conversions seems to always fail if inputting a grayscale JPEGs. ImageMagick identifies them as being "PseudoClass Gray".

To reproduce | 复现步骤 | 再現方法

I've uploaded an example of a grayscale image that won't convert.

Error log | 报错日志 | ログ

Log
[{"label":"CPU: Core™ i7-8750H","value":-1},{"label":"GPU 0: NVIDIA GeForce GTX 1070","value":0},{"label":"GPU: Auto","value":114514}]
{"gpuid":0,"inputpath":["D:\_Junk\_Cache\test\Claymore v1-001.jpg"],"model":"Waifu2x-cunet","modelscale":2,"modelnoise":0,"outputpath":"D:\_Junk\_Cache\test","targetscale":2,"tta":false}
2023-09-25 20:33:22.697 | INFO | main:main:58 - projectPATH: C:\Users\mpgri\Downloads\Final2x-windows-x64-unpacked\resources\app\resources\Final2x-core
2023-09-25 20:33:22.698 | INFO | main:main:71 - config loaded
2023-09-25 20:33:22.698 | INFO | src.SRFactory.SRBaseClass:_set_sr_n:53 - sr_n set to 1
2023-09-25 20:33:22.698 | INFO | src.SRFactory.SRBaseClass:init:35 - SRBaseClass init
[0 NVIDIA GeForce GTX 1070] queueC=2[8] queueG=0[16] queueT=1[2]
[0 NVIDIA GeForce GTX 1070] bugsbn1=0 bugbilz=0 bugcopc=0 bugihfa=0
[0 NVIDIA GeForce GTX 1070] fp16-p/s/a=1/1/0 int8-p/s/a=1/1/1
[0 NVIDIA GeForce GTX 1070] subgroup=32 basic=1 vote=1 ballot=1 shuffle=1
2023-09-25 20:33:23.450 | INFO | src.SRFactory.WAIFU2X:_init_SR_class:74 - Waifu2x model initialized
2023-09-25 20:33:23.451 | INFO | src.SRqueue:SR_queue:25 - Processing------[ 0.0% ]
2023-09-25 20:33:23.473 | ERROR | src.SRqueue:SR_queue:67 - tuple index out of range
2023-09-25 20:33:23.473 | WARNING | src.SRqueue:SR_queue:68 - CV2 load image failed: D:_Junk_Cache\test\Claymore v1-001.jpg, skip.
2023-09-25 20:33:23.473 | WARNING | src.SRqueue:SR_queue:69 - Skip_Image: D:_Junk_Cache\test\Claymore v1-001.jpg
2023-09-25 20:33:23.474 | INFO | src.utils.progressLog:printProgress:34 - Processing------[ 100.0% ]
2023-09-25 20:33:23.479 | SUCCESS | main:main:81 - SR_COMPLETED
CLOSE CODE:3221226505

Additional context | 附加内容 | 追加コンテキスト

No response

[Tohrusky]

At present, we can only convert it to rgb/rgba and then process it.

in https://github.com/Tohrusky/Final2x-core/blob/main/src/SRqueue.py#L59 I use this to compatible with special characters
img = cv2.imdecode(np.fromfile(img_path, dtype=np.uint8), cv2.IMREAD_UNCHANGED)

but cv.imread can load a grayscale image directly into 3 channels, when img is None, try to use it may work…
And the output of the neural network is in rgb. Wanna convert it back to grayscale?

[MarcG2]

I'm not sure I understand your comment. For reference, I use Final2x-core with the compiled Windows EXE.

I understand that the upscalers are intended for sRGB. So it's fine with me if the 1-channel grayscale images have to first be converted to RGB.

Having a built-in option for converting the final output back to 1-channel grayscale would be nice. Though I don't require it because I convert the upscaled images to JPEG afterwards.

[lclrc]

having the same problem. is there any solutions?

[Tohrusky]

having the same problem. is there any solutions?

I'll fix it in a few days