From 62a109dd251ef70f8534bc8a0246293b0b0bdd44 Mon Sep 17 00:00:00 2001 From: Philip Hofmann Date: Fri, 5 Jul 2024 18:09:23 +0200 Subject: [PATCH] Add my new models (#438) * Add ATD Arch * Add ATD 4x official pretrain * Create 4x-RealWebPhoto-v3-atd.json * Create 4x-RealWebPhoto-v4-dat2.json * Create 4x-Nomos8k-atd-jpg.json * Create 4x-RGT-S.json * Create 4x-RealWebPhoto-v2-rgt-s.json * Create 1x-Ludvae200.json * Fix 1x-Ludvae200 image links * 1x-Ludvae200 add pth file resource * Add 4x-TextureDAT2-otf * Add 4xTextures_GTAV_rgt-s * Add drct pretrains * Add span pretrains * Add 4xTextures_GTAV_rgt-s * Add 4xRealWebPhoto_v4_drct-l * Create 4x-Textures-GTAV-rgt-s-dither.json * 4x_realplksr_mssim_pretrain * Create 4x-realplksr-gan-pretrain.json * Create 4x-NomosWebPhoto-RealPLKSR.json * Create 4x-NomosWebPhoto-atd.json * Create 4x-NomosWebPhoto-esrgan.json * Create 4x-Nomos2-otf-esrgan.json * Create 4x-mssim-realplksr-dysample-pretrain.json * Create 4x-Nomos2-realplksr-dysample.json * Update 4x-Nomos2-otf-esrgan.json * Update 4x-NomosWebPhoto-atd.json --- data/architectures.json | 8 ++ data/models/4x-Nomos2-otf-esrgan.json | 99 +++++++++++++++++++ data/models/4x-Nomos2-realplksr-dysample.json | 72 ++++++++++++++ data/models/4x-NomosWebPhoto-RealPLKSR.json | 61 ++++++++++++ data/models/4x-NomosWebPhoto-atd.json | 69 +++++++++++++ data/models/4x-NomosWebPhoto-esrgan.json | 72 ++++++++++++++ .../models/4x-Textures-GTAV-rgt-s-dither.json | 77 +++++++++++++++ .../4x-mssim-realplksr-dysample-pretrain.json | 50 ++++++++++ data/models/4x-realplksr-gan-pretrain.json | 35 +++++++ data/models/4x-realplksr-mssim-pretrain.json | 29 ++++++ data/tag-categories.json | 3 +- data/tags.json | 4 + 12 files changed, 578 insertions(+), 1 deletion(-) create mode 100644 data/models/4x-Nomos2-otf-esrgan.json create mode 100644 data/models/4x-Nomos2-realplksr-dysample.json create mode 100644 data/models/4x-NomosWebPhoto-RealPLKSR.json create mode 100644 data/models/4x-NomosWebPhoto-atd.json create mode 100644 data/models/4x-NomosWebPhoto-esrgan.json create mode 100644 data/models/4x-Textures-GTAV-rgt-s-dither.json create mode 100644 data/models/4x-mssim-realplksr-dysample-pretrain.json create mode 100644 data/models/4x-realplksr-gan-pretrain.json create mode 100644 data/models/4x-realplksr-mssim-pretrain.json diff --git a/data/architectures.json b/data/architectures.json index 5330f6e8..7483c699 100644 --- a/data/architectures.json +++ b/data/architectures.json @@ -138,6 +138,14 @@ "onnx" ] }, + "realplksr": { + "name": "RealPLKSR", + "input": "image", + "compatiblePlatforms": [ + "pytorch", + "onnx" + ] + }, "rgt": { "name": "RGT", "input": "image", diff --git a/data/models/4x-Nomos2-otf-esrgan.json b/data/models/4x-Nomos2-otf-esrgan.json new file mode 100644 index 00000000..8ed3658b --- /dev/null +++ b/data/models/4x-Nomos2-otf-esrgan.json @@ -0,0 +1,99 @@ +{ + "name": "4xNomos2_otf_esrgan", + "author": "helaman", + "license": "CC-BY-4.0", + "tags": [ + "compression-removal", + "jpeg", + "photo", + "restoration" + ], + "description": "4xNomos2_otf_esrgan\n\nScale: 4 \nArchitecture: ESRGAN \nArchitecture Option: esrgan \n[Github Release Link](https://github.com/Phhofm/models/releases/tag/4xNomos2_otf_esrgan) \n\nAuthor: Philip Hofmann \nLicense: CC-BY-0.4 \nPurpose: Restoration \nSubject: Photography \nInput Type: Images \nRelease Date: 22.06.2024 \n\nDataset: Nomos-v2 \nDataset Size: 6000 \nOTF (on the fly augmentations): Yes \nPretrained Model: RealESRGAN_x4plus \nIterations: 246'000 \nBatch Size: 8 \nGT Size: 256 \n\nDescription: \n\n4x ESRGAN model for photography, trained using the Real-ESRGAN otf degradation pipeline. \n\nShowcase:\n[Slow Pics 8 Examples](https://slow.pics/s/Y5wdyTPh?canvas-mode=fit-width&image-fit=contain)", + "date": "2024-06-22", + "architecture": "esrgan", + "size": [ + "64nf", + "23nb" + ], + "scale": 4, + "inputChannels": 3, + "outputChannels": 3, + "resources": [ + { + "platform": "pytorch", + "type": "pth", + "size": 33686030, + "sha256": "6c16d8b22b8468e0c0b766f1569e10a77d38fc7a1a929b804806dce84ddb85d3", + "urls": [ + "https://github.com/Phhofm/models/releases/download/4xNomos2_otf_esrgan/4xNomos2_otf_esrgan.pth" + ] + }, + { + "platform": "pytorch", + "type": "safetensors", + "size": 33467822, + "sha256": "12db878907ed3a52ee97de552dfc0ce7ffd38559ca9760a7803cb2d06a737055", + "urls": [ + "https://github.com/Phhofm/models/releases/download/4xNomos2_otf_esrgan/4xNomos2_otf_esrgan.safetensors" + ] + }, + { + "platform": "onnx", + "type": "onnx", + "size": 67003044, + "sha256": "21905712a7963b95bc7e6e0b506e2fc6f353742fa7bc3ccdb81c0ad47744cc8b", + "urls": [ + "https://github.com/Phhofm/models/releases/download/4xNomos2_otf_esrgan/4xNomos2_otf_esrgan_fp32_opset17.onnx" + ] + } + ], + "trainingIterations": 246000, + "trainingBatchSize": 8, + "trainingHRSize": 256, + "trainingOTF": true, + "dataset": "Nomos-v2", + "datasetSize": 6000, + "pretrainedModelG": "4x-realesrgan-x4plus", + "images": [ + { + "type": "paired", + "LR": "https://i.slow.pics/s0aTG2Vz.png", + "SR": "https://i.slow.pics/TOL43AJ8.png" + }, + { + "type": "paired", + "LR": "https://i.slow.pics/556Unq6x.png", + "SR": "https://i.slow.pics/hF8hT44R.png" + }, + { + "type": "paired", + "LR": "https://i.slow.pics/lP3gidyo.png", + "SR": "https://i.slow.pics/nvQUv1Ah.png" + }, + { + "type": "paired", + "LR": "https://i.slow.pics/fBt2h5Lh.png", + "SR": "https://i.slow.pics/5TEHOFxf.png" + }, + { + "type": "paired", + "LR": "https://i.slow.pics/Xw0C7wiT.png", + "SR": "https://i.slow.pics/X4V4vmWk.png" + }, + { + "type": "paired", + "LR": "https://i.slow.pics/4XjROU2k.png", + "SR": "https://i.slow.pics/EMnCftls.png" + }, + { + "type": "paired", + "LR": "https://i.slow.pics/LhqBaoQa.png", + "SR": "https://i.slow.pics/6qr5sQZt.png" + }, + { + "type": "paired", + "LR": "https://i.slow.pics/5YsUu1mQ.png", + "SR": "https://i.slow.pics/gkbocFPC.png" + } + ] +} \ No newline at end of file diff --git a/data/models/4x-Nomos2-realplksr-dysample.json b/data/models/4x-Nomos2-realplksr-dysample.json new file mode 100644 index 00000000..45a3a6c1 --- /dev/null +++ b/data/models/4x-Nomos2-realplksr-dysample.json @@ -0,0 +1,72 @@ +{ + "name": "4xNomos2_realplksr_dysample", + "author": "helaman", + "license": "CC-BY-4.0", + "tags": [ + "compression-removal", + "jpeg", + "photo", + "restoration" + ], + "description": "4xNomos2_realplksr_dysample \nScale: 4 \nArchitecture: [RealPLKSR with Dysample](https://github.com/muslll/neosr/?tab=readme-ov-file#supported-archs) \nArchitecture Option: [realplksr](https://github.com/muslll/neosr/blob/master/neosr/archs/realplksr_arch.py) \n[Github Release](https://github.com/Phhofm/models/releases/tag/4xNomos2_realplksr_dysample) \n\nAuthor: Philip Hofmann \nLicense: CC-BY-0.4 \nPurpose: Pretrained \nSubject: Photography \nInput Type: Images \nRelease Date: 30.06.2024 \n\nDataset: [nomosv2](https://github.com/muslll/neosr/?tab=readme-ov-file#-datasets) \nDataset Size: 6000 \nOTF (on the fly augmentations): No \nPretrained Model: [4xmssim_realplksr_dysample_pretrain](https://github.com/Phhofm/models/releases/tag/4xmssim_realplksr_dysample_pretrain) \nIterations: 185'000 \nBatch Size: 8 \nGT Size: 256, 512 \n\nDescription: \nA Dysample RealPLKSR 4x upscaling model that was trained with / handles jpg compression down to 70 on the Nomosv2 dataset, preserves DoF. \nBased on the [4xmssim_realplksr_dysample_pretrain](https://github.com/Phhofm/models/releases/tag/4xmssim_realplksr_dysample_pretrain) I released 3 days ago. \nThis model affects / saturate colors, which can be counteracted a bit by using wavelet color fix, as used in these examples. \n\nShowcase:\n[Slowpics](https://slow.pics/s/p3sMnd5l)", + "date": "2024-06-30", + "architecture": "realplksr", + "size": null, + "scale": 4, + "inputChannels": 3, + "outputChannels": 3, + "resources": [ + { + "platform": "pytorch", + "type": "pth", + "size": 29715988, + "sha256": "96c8476ed2c42107628d10562b32ae89e7be3c2f7a53d29504823e9bafe8121d", + "urls": [ + "https://github.com/Phhofm/models/releases/download/4xNomos2_realplksr_dysample/4xNomos2_realplksr_dysample.pth" + ] + }, + { + "platform": "onnx", + "type": "onnx", + "size": 30298952, + "sha256": "f4da353c0a380aa2c53485f2479bfd7959cab071734075d96f53df3233976716", + "urls": [ + "https://github.com/Phhofm/models/releases/download/4xNomos2_realplksr_dysample/4xNomos2_realplksr_dysample_256_fp32_fullyoptimized.onnx" + ] + } + ], + "trainingIterations": 185000, + "trainingBatchSize": 8, + "trainingHRSize": 512, + "trainingOTF": false, + "dataset": "Nomos-v2", + "datasetSize": 6000, + "pretrainedModelG": "4x-mssim-realplksr-dysample-pretrain", + "images": [ + { + "type": "paired", + "LR": "https://i.slow.pics/9ZnUcVDC.png", + "SR": "https://i.slow.pics/XJ5Tt9ja.png" + }, + { + "type": "paired", + "LR": "https://i.slow.pics/wdQgPTgv.png", + "SR": "https://i.slow.pics/1jspoDeR.png" + }, + { + "type": "paired", + "LR": "https://i.slow.pics/7NVM3NoB.png", + "SR": "https://i.slow.pics/481WUhuM.png" + }, + { + "type": "paired", + "LR": "https://i.slow.pics/4PZ6cOrU.png", + "SR": "https://i.slow.pics/xlPjpxUS.png" + }, + { + "type": "paired", + "LR": "https://i.slow.pics/SHPUrMz1.png", + "SR": "https://i.slow.pics/YVMbx6KC.png" + } + ] +} \ No newline at end of file diff --git a/data/models/4x-NomosWebPhoto-RealPLKSR.json b/data/models/4x-NomosWebPhoto-RealPLKSR.json new file mode 100644 index 00000000..3eb8f216 --- /dev/null +++ b/data/models/4x-NomosWebPhoto-RealPLKSR.json @@ -0,0 +1,61 @@ +{ + "name": "4xNomosWebPhoto_RealPLKSR", + "author": "helaman", + "license": "CC-BY-4.0", + "tags": [ + "photo", + "restoration" + ], + "description": "4xNomosWebPhoto_RealPLKSR \n\nScale: 4 \nArchitecture: RealPLKSR \nArchitecture Option: realplksr \n[Link to Github Release](https://github.com/Phhofm/models/releases/tag/4xNomosWebPhoto_RealPLKSR) \n\nAuthor: Philip Hofmann \nLicense: CC-BY-0.4 \nPurpose: Restoration \nSubject: Photography \nInput Type: Images \nRelease Date: 28.05.2024 \n\nDataset: Nomos-v2 \nDataset Size: 6000 \nOTF (on the fly augmentations): No \nPretrained Model: 4x_realplksr_gan_pretrain \nIterations: 404'000, 445'000 \nBatch Size: 12, 4 \nGT Size: 128, 256, 512 \n\nDescription:\n\nshort: 4x RealPLKSR model for photography, trained with realistic noise, lens blur, jpg and webp re-compression.\n\nfull: My newest version of my RealWebPhoto series, this time I used the newly released Nomos-v2 dataset by musl. \nI then made 12 different low resolution degraded folders, using kim's datasetdestroyer for scaling and compression, my ludvae200 model for realistic noise, and umzi's wtp_dataset_destroyer with its floating point lens blur implementation for better control. \nI then mixed them together in a single lr folder and trained for 460'000 iters, checked the results, and upon kims suggestion of using interpolation, I tested and am releasing this interpolation between the checkpoints 404'000 and 445'000. \n\nThis model has been trained on neosr using mixup, cutmix, resizemix, cutblur, nadam, unet, multisteplr, mssim, perceptual, gan, dists, ldl, ff, color and lumaloss, and interpolated using the current chaiNNer nightly version. \n\nThis model took multiple retrainings and reworks of the dataset, until I am now satisfied enough with the quality to release this version. \n\nFor more details on the whole process see the pdf file in the attachement. \n\nI am also attaching the 404'000, 445'000 and 460'000 checkpoints for completeness. \n\nPS in general degradation strengths have been reduced/adjusted in comparison to my previous RealWebPhoto models \n\nShowcase:\n[Slow Pics 10 Examples](https://slow.pics/s/euvEv4hL)", + "date": "2024-05-28", + "architecture": "realplksr", + "size": [ + "64dim", + "28nb", + "17ks", + "Real" + ], + "scale": 4, + "inputChannels": 3, + "outputChannels": 3, + "resources": [ + { + "platform": "pytorch", + "type": "pth", + "size": 29683482, + "sha256": "a9db66c9b674c6a5025b6ef3bee71a57c33b8605d8a2de0980470f89002efbbe", + "urls": [ + "https://github.com/Phhofm/models/releases/download/4xNomosWebPhoto_RealPLKSR/4xNomosWebPhoto_RealPLKSR.pth" + ] + } + ], + "trainingIterations": 445000, + "trainingBatchSize": 4, + "trainingHRSize": 512, + "trainingOTF": false, + "dataset": "Nomos-v2", + "datasetSize": 6000, + "pretrainedModelG": "4x-realplksr-gan-pretrain", + "images": [ + { + "type": "paired", + "LR": "https://i.slow.pics/ttBmEsyM.png", + "SR": "https://i.slow.pics/yHKf35eW.png" + }, + { + "type": "paired", + "LR": "https://i.slow.pics/68yeIooo.png", + "SR": "https://i.slow.pics/Z6vmBN69.png" + }, + { + "type": "paired", + "LR": "https://i.slow.pics/LUc7ikyl.png", + "SR": "https://i.slow.pics/IhI3wtGC.png" + }, + { + "type": "paired", + "LR": "https://i.slow.pics/mljAdTUv.png", + "SR": "https://i.slow.pics/XtfSCf8c.png" + } + ] +} \ No newline at end of file diff --git a/data/models/4x-NomosWebPhoto-atd.json b/data/models/4x-NomosWebPhoto-atd.json new file mode 100644 index 00000000..596f7841 --- /dev/null +++ b/data/models/4x-NomosWebPhoto-atd.json @@ -0,0 +1,69 @@ +{ + "name": "4xNomosWebPhoto_atd", + "author": "helaman", + "license": "CC-BY-4.0", + "tags": [ + "photo", + "restoration" + ], + "description": "4xNomosWebPhoto_atd\n\nScale: 4 \nArchitecture: [ATD](https://github.com/LabShuHangGU/Adaptive-Token-Dictionary) \nArchitecture Option: atd \n[Github Release Link](https://github.com/Phhofm/models/releases/tag/4xNomosWebPhoto_atd) \n\n\nAuthor: Philip Hofmann \nLicense: CC-BY-0.4 \nPurpose: Restoration \nSubject: Photography \nInput Type: Images \nRelease Date: 07.06.2024 \n\nDataset: Nomos-v2 \nDataset Size: 6000 \nOTF (on the fly augmentations): No \nPretrained Model: 003_ATD_SRx4_finetune.pth \nIterations: 460'000 \nBatch Size: 6, 2 \nGT Size: 128, 192 \n\nDescription: \n\n4x ATD model for photography, trained with realistic noise, lens blur, jpg and webp re-compression. \nATD version of [4xNomosWebPhoto_RealPLKSR](https://github.com/Phhofm/models/releases/tag/4xNomosWebPhoto_RealPLKSR), trained on the same dataset and in the same way. For more information look into the [4xNomosWebPhoto_RealPLKSR](https://github.com/Phhofm/models/releases/tag/4xNomosWebPhoto_RealPLKSR) release, and the pdf file in its attachments.\n\nShowcase:\n[Slow Pics 18 Examples](https://slow.pics/s/GkS39XYi)", + "date": "2024-06-07", + "architecture": "atd", + "size": [ + "210dim", + "16w", + "256cat" + ], + "scale": 4, + "inputChannels": 3, + "outputChannels": 3, + "resources": [ + { + "platform": "pytorch", + "type": "pth", + "size": 81959074, + "sha256": "f1af8e13824e44dfaefd9954e4dd9a5abe916f40971ed323f63727735aa0983c", + "urls": [ + "https://github.com/Phhofm/models/releases/download/4xNomosWebPhoto_atd/4xNomosWebPhoto_atd.pth" + ] + }, + { + "platform": "pytorch", + "type": "safetensors", + "size": 81689540, + "sha256": "ca0e7a3ecbb042dad1695d4dd3b73cf355af6db822984b61ac363e42fac7dff1", + "urls": [ + "/home/phhofm/Downloads/4xNomosWebPhoto_atd.safetensors" + ] + } + ], + "trainingIterations": 460000, + "trainingBatchSize": 2, + "trainingHRSize": 192, + "trainingOTF": false, + "dataset": "Nomos-v2", + "datasetSize": 6000, + "pretrainedModelG": "4x-003-ATD-SRx4-finetune", + "images": [ + { + "type": "paired", + "LR": "https://i.slow.pics/cJW6aPS0.png", + "SR": "https://i.slow.pics/CLWwbc3K.png" + }, + { + "type": "paired", + "LR": "https://i.slow.pics/OuZiJlE6.png", + "SR": "https://i.slow.pics/11QRFEdo.png" + }, + { + "type": "paired", + "LR": "https://i.slow.pics/Nls9WLwm.png", + "SR": "https://i.slow.pics/vbYgeTL8.png" + }, + { + "type": "paired", + "LR": "https://i.slow.pics/U06F3Fm2.png", + "SR": "https://i.slow.pics/RXBNdhut.png" + } + ] +} \ No newline at end of file diff --git a/data/models/4x-NomosWebPhoto-esrgan.json b/data/models/4x-NomosWebPhoto-esrgan.json new file mode 100644 index 00000000..e75f0363 --- /dev/null +++ b/data/models/4x-NomosWebPhoto-esrgan.json @@ -0,0 +1,72 @@ +{ + "name": "4xNomosWebPhoto_esrgan", + "author": "helaman", + "license": "CC-BY-4.0", + "tags": [ + "photo", + "restoration" + ], + "description": "I simply wanted to release an ESRGAN model just because I had not trained one for quite a while and just wanted to revisit this older arch for the current series.\n\n4xNomosWebPhoto_esrgan\n\nScale: 4 \nArchitecture: ESRGAN \nArchitecture Option: esrgan \n[Github Release Link](https://github.com/Phhofm/models/releases/tag/4xNomosWebPhoto_esrgan) \n\nAuthor: Philip Hofmann \nLicense: CC-BY-0.4 \nPurpose: Restoration \nSubject: Photography \nInput Type: Images \nRelease Date: 16.06.2024 \n\nDataset: Nomos-v2 \nDataset Size: 6000 \nOTF (on the fly augmentations): No \nPretrained Model: RealESRGAN_x4plus \nIterations: 210'000 \nBatch Size: 12 \nGT Size: 256 \n\nDescription:\n\n4x ESRGAN model for photography, trained with realistic noise, lens blur, jpg and webp re-compression. \nESRGAN version of 4xNomosWebPhoto_RealPLKSR, trained on the same dataset and in a similiar way. \nFor more information look into the 4xNomosWebPhoto_RealPLKSR release, and the pdf file in its attachments.", + "date": "2024-06-16", + "architecture": "esrgan", + "size": [ + "64nf", + "23nb" + ], + "scale": 4, + "inputChannels": 3, + "outputChannels": 3, + "resources": [ + { + "platform": "pytorch", + "type": "pth", + "size": 33688212, + "sha256": "131bb044af9fff0c748e460df1dce5aff018e2cd44340bbc6d97e683cdc0d491", + "urls": [ + "https://github.com/Phhofm/models/releases/download/4xNomosWebPhoto_esrgan/4xNomosWebPhoto_esrgan.pth" + ] + }, + { + "platform": "pytorch", + "type": "safetensors", + "size": 33467822, + "sha256": "2f49c3065db8aaeb64adb08e8879001fe2f1460df5655499707efedad8a3d556", + "urls": [ + "https://github.com/Phhofm/models/releases/download/4xNomosWebPhoto_esrgan/4xNomosWebPhoto_esrgan.safetensors" + ] + }, + { + "platform": "onnx", + "type": "onnx", + "size": 67003044, + "sha256": "e29f473727ee0e32416d3a2d2642568e0159e2608d67236193772fced8696121", + "urls": [ + "https://github.com/Phhofm/models/releases/download/4xNomosWebPhoto_esrgan/4xNomosWebPhoto_esrgan_fp32_opset17.onnx" + ] + } + ], + "trainingIterations": 210000, + "trainingBatchSize": 12, + "trainingHRSize": 256, + "trainingOTF": false, + "dataset": "Nomos-v2", + "datasetSize": 6000, + "pretrainedModelG": "4x-realesrgan-x4plus", + "images": [ + { + "type": "paired", + "LR": "https://i.slow.pics/Q20xKISa.png", + "SR": "https://i.slow.pics/JFKE6GKT.png" + }, + { + "type": "paired", + "LR": "https://i.slow.pics/fY0zlrme.png", + "SR": "https://i.slow.pics/5yq29FLI.png" + }, + { + "type": "paired", + "LR": "https://i.slow.pics/ORvP5MW6.png", + "SR": "https://i.slow.pics/fwrZBVdt.png" + } + ] +} \ No newline at end of file diff --git a/data/models/4x-Textures-GTAV-rgt-s-dither.json b/data/models/4x-Textures-GTAV-rgt-s-dither.json new file mode 100644 index 00000000..ba3509ea --- /dev/null +++ b/data/models/4x-Textures-GTAV-rgt-s-dither.json @@ -0,0 +1,77 @@ +{ + "name": "4xTextures_GTAV_rgt-s_dither", + "author": "helaman", + "license": "CC-BY-4.0", + "tags": [ + "compression-removal", + "dedither", + "game-textures", + "jpeg", + "restoration" + ], + "description": "## 4xTextures_GTAV_rgt-s_dither\n\n**Scale:** 4 \n**Architecture:** [RGT](https://github.com/zhengchen1999/RGT) \n**Architecture Option:** RGT-S \n\n**Author:** Philip Hofmann \n**License:** CC-BY-0.4 \n**Purpose:** Restoration \n**Subject:** Game Textures \n**Input Type:** Images \n**Release Date:** 08.05.2024 \n\n**Dataset:** [GTAV_512_Textures](https://discord.com/channels/547949405949657098/905446120333930566/1132367991808466975) \n**Dataset Size:** 7061 \n**OTF (on the fly augmentations):** No \n**Pretrained Model:** 4xTextures_GTAV_rgt-s \n**Iterations:** 128'000 \n**Batch Size:** 6,4 \n**GT Size:** 128,256 \n\n**Description:**\nA model to upscale game textures, trained on GTAV Textures, handles jpg compression down to 80 and was trained with dithering. Basically the previous 4xTextures_GTAV_rgt-s model but extended to handle dithering.\n\n**Showcase:** \n[Slow Pics 25 Examples](https://slow.pics/s/EW7Ifiuw)", + "date": "2024-05-08", + "architecture": "rgt", + "size": null, + "scale": 4, + "inputChannels": 3, + "outputChannels": 3, + "resources": [ + { + "platform": "pytorch", + "type": "pth", + "size": 135994918, + "sha256": "49e4b4aaac843839219d509242298c9e9b1a5d5294e9bad1154f05aa732a6a09", + "urls": [ + "https://github.com/Phhofm/models/releases/download/4xTextures_GTAV_rgt-s_dither/4xTextures_GTAV_rgt-s_dither.pth" + ] + }, + { + "platform": "pytorch", + "type": "safetensors", + "size": 135601676, + "sha256": "ef850c4fd1d1ab81eb8cf63a0e11c62adfead07ec3a34f81675e986e76800f3f", + "urls": [ + "https://github.com/Phhofm/models/releases/download/4xTextures_GTAV_rgt-s_dither/4xTextures_GTAV_rgt-s_dither.safetensors" + ] + }, + { + "platform": "onnx", + "type": "onnx", + "size": 46068836, + "sha256": "40bcebc87f5fd2308763fba5d3e71d9cebead68f0e0771fbc53bb78914b84b27", + "urls": [ + "https://github.com/Phhofm/models/releases/download/4xTextures_GTAV_rgt-s_dither/4xTextures_GTAV_rgt-s_dither_fp32_opset17.onnx" + ] + } + ], + "trainingIterations": 128000, + "trainingBatchSize": 4, + "trainingHRSize": 256, + "trainingOTF": false, + "dataset": "GTAV_512_Textures", + "datasetSize": 7061, + "pretrainedModelG": "4x-Textures-GTAV-rgt-s", + "images": [ + { + "type": "paired", + "LR": "https://i.slow.pics/47Est1eG.png", + "SR": "https://i.slow.pics/0mT3MX6v.png" + }, + { + "type": "paired", + "LR": "https://i.slow.pics/sYlGUhTR.png", + "SR": "https://i.slow.pics/6QzxAMM1.png" + }, + { + "type": "paired", + "LR": "https://i.slow.pics/HTBuOdVM.png", + "SR": "https://i.slow.pics/ccSW80P4.png" + }, + { + "type": "paired", + "LR": "https://i.slow.pics/IHo1iJkB.png", + "SR": "https://i.slow.pics/FNkq7062.png" + } + ] +} \ No newline at end of file diff --git a/data/models/4x-mssim-realplksr-dysample-pretrain.json b/data/models/4x-mssim-realplksr-dysample-pretrain.json new file mode 100644 index 00000000..69f6101b --- /dev/null +++ b/data/models/4x-mssim-realplksr-dysample-pretrain.json @@ -0,0 +1,50 @@ +{ + "name": "4xmssim_realplksr_dysample_pretrain", + "author": "helaman", + "license": "CC-BY-4.0", + "tags": [ + "photo", + "pretrained" + ], + "description": "4xmssim_realplksr_dysample_pretrain \nScale: 4 \nArchitecture: [RealPLKSR with Dysample](https://github.com/muslll/neosr/?tab=readme-ov-file#supported-archs) \nArchitecture Option: [realplksr](https://github.com/muslll/neosr/blob/master/neosr/archs/realplksr_arch.py) \n\nAuthor: Philip Hofmann \nLicense: CC-BY-0.4 \nPurpose: Pretrained \nSubject: Photography \nInput Type: Images \nRelease Date: 27.06.2024 \n\nDataset: [nomosv2](https://github.com/muslll/neosr/?tab=readme-ov-file#-datasets) \nDataset Size: 6000 \nOTF (on the fly augmentations): No \nPretrained Model: None (=From Scratch) \nIterations: 200'000 \nBatch Size: 8 \nGT Size: 192, 512 \n\nDescription: [Dysample](https://arxiv.org/pdf/2308.15085) had been recently added to RealPLKSR, which from what I had seen can resolve or help avoid the checkerboard / grid pattern on inference outputs. So with the [commits from three days ago, the 24.06.24, on neosr](https://github.com/muslll/neosr/commits/master/?since=2024-06-24&until=2024-06-24), I wanted to create a 4x photo pretrain I can then use to train more realplksr models with dysample specifically to stabilize training at the beginning.\n\nShowcase: \n[Imgsli](https://imgsli.com/Mjc0OTA1) \n[Slowpics](https://slow.pics/c/I9grkcqM)", + "date": "2024-06-27", + "architecture": "realplksr", + "size": null, + "scale": 4, + "inputChannels": 3, + "outputChannels": 3, + "resources": [ + { + "platform": "pytorch", + "type": "pth", + "size": 29715988, + "sha256": "0a8706be7239d571a700b4c5bde2ca565d1f78a9386807e197fc02854e26862f", + "urls": [ + "https://github.com/Phhofm/models/releases/download/4xmssim_realplksr_dysample_pretrain/4xmssim_realplksr_dysample_pretrain.pth" + ] + } + ], + "trainingIterations": 200000, + "trainingBatchSize": 8, + "trainingHRSize": 512, + "trainingOTF": false, + "dataset": "Nomos-v2", + "datasetSize": 6000, + "images": [ + { + "type": "paired", + "LR": "https://i.slow.pics/Gp03YPR1.png", + "SR": "https://i.slow.pics/TRrZx7uR.png" + }, + { + "type": "paired", + "LR": "https://i.slow.pics/WeKmX7jp.png", + "SR": "https://i.slow.pics/WEdJkyBU.png" + }, + { + "type": "paired", + "LR": "https://i.slow.pics/gdVBJEs1.png", + "SR": "https://i.slow.pics/xIuYfymN.png" + } + ] +} \ No newline at end of file diff --git a/data/models/4x-realplksr-gan-pretrain.json b/data/models/4x-realplksr-gan-pretrain.json new file mode 100644 index 00000000..0f500c53 --- /dev/null +++ b/data/models/4x-realplksr-gan-pretrain.json @@ -0,0 +1,35 @@ +{ + "name": "4x_realplksr_gan_pretrain", + "author": "musl", + "license": "CC0-1.0", + "tags": [ + "photo", + "pretrained" + ], + "description": "## RealPLKSR GAN pretrain\n\n**Scale:** 4x \n**Architecture:** RealPLKSR \n**Download:** [GDrive](https://drive.google.com/file/d/1iQmsnMhWXsHLYuYEwwRhdj9fMg71h-IU/view) | Training files: [GDrive](https://drive.google.com/drive/folders/1OWLGBce6oe8W6kARINFCWG7o0O5LeC_h?usp=sharing) \n\n**Author:** musl \n**License:** [CC0](https://creativecommons.org/public-domain/cc0/) \n**Purpose:** Pretrain \n**Subject:** Multipurpose \n**Date:** 15 May 2024 \n\n**Size:** default Real-PLKSR \n**I/O Channels:** 3(RGB)->3(RGB) \n\n**Dataset:** Nomos-v2 \n**Dataset Size:** 6000 \n**OTF (on the fly augmentations):** No \n**Pretrained Model:** 4x_realplksr_mssim_pretrain \n**Iterations:** ~450k \n**Batch Size:** 2-6 \n**GT Size:** 128-416 \n\n**Description:** Pretrained GAN models for RealPLKSR network. Trained on downsampling-only (nearest, bilinear, bicubic, lanczos and mitchell).", + "date": "2024-05-15", + "architecture": "realplksr", + "size": [ + "64dim", + "28nb", + "17ks", + "Real" + ], + "scale": 4, + "inputChannels": 3, + "outputChannels": 3, + "resources": [ + { + "platform": "pytorch", + "type": "pth", + "size": 29678402, + "sha256": "bc7ed4f8fb54ebd7721e07bbf275109c1b7a19886b152808db4647f4f411b220", + "urls": [ + "https://drive.google.com/file/d/1iQmsnMhWXsHLYuYEwwRhdj9fMg71h-IU/view" + ] + } + ], + "trainingOTF": false, + "pretrainedModelG": "4x-realplksr-mssim-pretrain", + "images": [] +} \ No newline at end of file diff --git a/data/models/4x-realplksr-mssim-pretrain.json b/data/models/4x-realplksr-mssim-pretrain.json new file mode 100644 index 00000000..058f3659 --- /dev/null +++ b/data/models/4x-realplksr-mssim-pretrain.json @@ -0,0 +1,29 @@ +{ + "name": "4x_realplksr_mssim_pretrain", + "author": "musl", + "license": "CC0-1.0", + "tags": [ + "photo", + "pretrained" + ], + "description": "## RealPLKSR mssim pretrains\n\n**Scale:** 2x and 4x \n**Architecture:** RealPLKSR \n**Links:** \n\n**`4x_realplksr_mssim_pretrain.pth`** | [GDrive](https://drive.google.com/file/d/12ek1vitEporWc5qqaYo6AMy0-RYlRqu8/view?usp=drive_link) \n**`2x_realplksr_mssim_pretrain.pth`** | [GDrive](https://drive.google.com/file/d/1GAdf5VOqYa5ntswT9sYsKKZ2Z7OQp7gO/view?usp=drive_link) \n\n**Author:** musl \n**License:** [CC0](https://creativecommons.org/public-domain/cc0/) \n**Purpose:** Pretrain \n**Subject:** Multipurpose \n**Date:** 08 May 2024 \n\n**Size:** default Real-PLKSR \n**I/O Channels:** 3(RGB)->3(RGB) \n\n**Dataset:** Nomos-v2 \n**Dataset Size:** 6000 \n**OTF (on the fly augmentations):** No \n**Pretrained Model:** scratch \n**Iterations:** ~260k \n**Batch Size:** 2-6 \n**GT Size:** 128-256 \n\n**Description:** Pretrained models for RealPLKSR network. Trained on downsampling-only (nearest, bilinear, bicubic and lanczos).", + "date": "2024-05-09", + "architecture": "realplksr", + "size": null, + "scale": 4, + "inputChannels": 3, + "outputChannels": 3, + "resources": [ + { + "platform": "pytorch", + "type": "pth", + "size": 29678402, + "sha256": "37a1a6245ea29dc17655b1df64ba5090b0d5ed9d4b50525a722c53e378d552f1", + "urls": [ + "https://drive.google.com/file/d/12ek1vitEporWc5qqaYo6AMy0-RYlRqu8/view" + ] + } + ], + "trainingOTF": false, + "images": [] +} \ No newline at end of file diff --git a/data/tag-categories.json b/data/tag-categories.json index 443ae3fe..c35a26a8 100644 --- a/data/tag-categories.json +++ b/data/tag-categories.json @@ -76,7 +76,8 @@ "arch:swift-srgan", "arch:swinir", "arch:atd", - "arch:drct" + "arch:drct", + "arch:realplksr" ] }, "platform": { diff --git a/data/tags.json b/data/tags.json index 43a4e880..0fc71b74 100644 --- a/data/tags.json +++ b/data/tags.json @@ -216,6 +216,10 @@ "name": "Real-CUGAN", "description": "" }, + "arch:realplksr": { + "name": "RealPLKSR", + "description": "" + }, "arch:rgt": { "name": "RGT", "description": ""