Automate/ skip Aug_Strength variable when resuming from previous pkl

.github/FUNDING.yml

@@ -0,0 +1,12 @@
+# These are supported funding model platforms
+
+github: dvschultz
+patreon: bustbright
+open_collective: # Replace with a single Open Collective username
+ko_fi: # Replace with a single Ko-fi username
+tidelift: # Replace with a single Tidelift platform-name/package-name e.g., npm/babel
+community_bridge: # Replace with a single Community Bridge project-name e.g., cloud-foundry
+liberapay: # Replace with a single Liberapay username
+issuehunt: # Replace with a single IssueHunt username
+otechie: # Replace with a single Otechie username
+custom: ['https://www.youtube.com/channel/UCaZuPdmZ380SFUMKHVsv_AA/join']

Network_Blending_ADA_PT.ipynb

@@ -0,0 +1,302 @@
+{
+  "nbformat": 4,
+  "nbformat_minor": 0,
+  "metadata": {
+    "colab": {
+      "name": "Network-Blending-ADA-PT.ipynb",
+      "private_outputs": true,
+      "provenance": [],
+      "machine_shape": "hm",
+      "mount_file_id": "17KlZS6Ax5HEcJDhW0EVAsrOrXBeTVqxv",
+      "authorship_tag": "ABX9TyN0OSZzbXfWmS2HYwtpQSn3",
+      "include_colab_link": true
+    },
+    "kernelspec": {
+      "name": "python3",
+      "display_name": "Python 3"
+    },
+    "language_info": {
+      "name": "python"
+    },
+    "accelerator": "GPU"
+  },
+  "cells": [
+    {
+      "cell_type": "markdown",
+      "metadata": {
+        "id": "view-in-github",
+        "colab_type": "text"
+      },
+      "source": [
+        "<a href=\"https://colab.research.google.com/github/dvschultz/stylegan2-ada-pytorch/blob/main/Network_Blending_ADA_PT.ipynb\" target=\"_parent\"><img src=\"https://colab.research.google.com/assets/colab-badge.svg\" alt=\"Open In Colab\"/></a>"
+      ]
+    },
+    {
+      "cell_type": "markdown",
+      "metadata": {
+        "id": "CdkWE1TKUWwB"
+      },
+      "source": [
+        "# Network Blending\n",
+        "This demo will show how to combine two separate StyleGAN2-ADA-PyTorch models into one by splitting their weights at a specified layer.\n",
+        "\n",
+        "This example was created by Derrick Schultz for his Advanced StyleGAN2 class. It’s a simpler version of [Justin Pinkney’s Tensorflow version](https://github.com/justinpinkney/stylegan2/blob/master/blend_models.py).\n",
+        "\n",
+        "---\n",
+        "\n",
+        "If you find this notebook useful, consider signing up for my [Patreon](https://www.patreon.com/bustbright) or [YouTube channel](https://www.youtube.com/channel/UCaZuPdmZ380SFUMKHVsv_AA/join). You can also send me a one-time payment on [Venmo](https://venmo.com/Derrick-Schultz).\n"
+      ]
+    },
+    {
+      "cell_type": "code",
+      "metadata": {
+        "id": "aJFXX8WIBeqy"
+      },
+      "source": [
+        "!nvidia-smi -L"
+      ],
+      "execution_count": null,
+      "outputs": []
+    },
+    {
+      "cell_type": "code",
+      "metadata": {
+        "id": "hvwu1LpCOJtz"
+      },
+      "source": [
+        "!git clone https://github.com/dvschultz/stylegan2-ada-pytorch\n",
+        "%cd stylegan2-ada-pytorch\n",
+        "!pip install ninja opensimplex"
+      ],
+      "execution_count": null,
+      "outputs": []
+    },
+    {
+      "cell_type": "markdown",
+      "metadata": {
+        "id": "O09iXoAPQsHX"
+      },
+      "source": [
+        "## Download two models"
+      ]
+    },
+    {
+      "cell_type": "code",
+      "metadata": {
+        "id": "p9dwvbtWp97y"
+      },
+      "source": [
+        "!wget http://d36zk2xti64re0.cloudfront.net/stylegan2/networks/stylegan2-church-config-f.pkl\n",
+        "!wget http://d36zk2xti64re0.cloudfront.net/stylegan2/networks/stylegan2-cat-config-f.pkl"
+      ],
+      "execution_count": null,
+      "outputs": []
+    },
+    {
+      "cell_type": "code",
+      "metadata": {
+        "id": "kRQUzZK2HHSt"
+      },
+      "source": [
+        "!gdown --id 15GpzB-wTwGIZC_Wu0ruaEJi7-giRWOOo -O /content/bone-bone.pkl\n",
+        "!wget http://d36zk2xti64re0.cloudfront.net/stylegan2/networks/stylegan2-ffhq-config-f.pkl"
+      ],
+      "execution_count": null,
+      "outputs": []
+    },
+    {
+      "cell_type": "code",
+      "metadata": {
+        "id": "FnHs7WBuqXDo"
+      },
+      "source": [
+        "!python legacy.py --source=/content/stylegan2-ada-pytorch/stylegan2-ffhq-config-f.pkl --dest=/content/ffhq-pt.pkl\n",
+        "!python legacy.py --source=/content/bone-bone.pkl --dest=/content/bone-bone-pt.pkl"
+      ],
+      "execution_count": null,
+      "outputs": []
+    },
+    {
+      "cell_type": "markdown",
+      "metadata": {
+        "id": "c_kFpXUAvcTP"
+      },
+      "source": [
+        "## Script Example\n",
+        "If you want to simply run the command as a script, you run the cell below."
+      ]
+    },
+    {
+      "cell_type": "code",
+      "metadata": {
+        "id": "s_P9TtAsv_xW"
+      },
+      "source": [
+        "!python blend_models.py --help"
+      ],
+      "execution_count": null,
+      "outputs": []
+    },
+    {
+      "cell_type": "code",
+      "metadata": {
+        "id": "A-S0KjY1vuzj"
+      },
+      "source": [
+        "!python blend_models.py --lower_res_pkl /content/ffhq-pt.pkl --split_res 64 --higher_res_pkl /content/bone-bone-pt.pkl --output_path /content/ffhq-bonebone-split64.pkl"
+      ],
+      "execution_count": null,
+      "outputs": []
+    },
+    {
+      "cell_type": "markdown",
+      "metadata": {
+        "id": "4uDMVyB0UCiS"
+      },
+      "source": [
+        "## Code example\n",
+        "\n",
+        "If you want to see under the hood here’s how this works."
+      ]
+    },
+    {
+      "cell_type": "code",
+      "metadata": {
+        "id": "ENNGfqvKS59e"
+      },
+      "source": [
+        "import os\n",
+        "import copy\n",
+        "import numpy as np\n",
+        "import torch\n",
+        "import pickle\n",
+        "import dnnlib\n",
+        "import legacy\n",
+        "\n",
+        "def extract_conv_names(model, model_res):\n",
+        "    model_names = list(name for name,weight in model.named_parameters())\n",
+        "\n",
+        "    return model_names\n",
+        "\n",
+        "def blend_models(low, high, model_res, resolution, level, blend_width=None):\n",
+        "\n",
+        "    resolutions =  [4*2**x for x in range(int(np.log2(resolution)-1))]\n",
+        "    print(resolutions)\n",
+        "    \n",
+        "    low_names = extract_conv_names(low, model_res)\n",
+        "    high_names = extract_conv_names(high, model_res)\n",
+        "\n",
+        "    assert all((x == y for x, y in zip(low_names, high_names)))\n",
+        "\n",
+        "    #start with lower model and add weights above\n",
+        "    model_out = copy.deepcopy(low)\n",
+        "    params_src = high.named_parameters()\n",
+        "    dict_dest = model_out.state_dict()\n",
+        "\n",
+        "    for name, param in params_src:\n",
+        "        if not any(f'synthesis.b{res}' in name for res in resolutions) and not ('mapping' in name):\n",
+        "            # print(name)\n",
+        "            dict_dest[name].data.copy_(param.data)\n",
+        "\n",
+        "    model_out_dict = model_out.state_dict()\n",
+        "    model_out_dict.update(dict_dest) \n",
+        "    model_out.load_state_dict(dict_dest)\n",
+        "    \n",
+        "    return model_out"
+      ],
+      "execution_count": null,
+      "outputs": []
+    },
+    {
+      "cell_type": "code",
+      "metadata": {
+        "id": "ovSNBUBxOeXt"
+      },
+      "source": [
+        "lo_res_pkl = '/content/freagan-pt.pkl'\n",
+        "hi_res_pkl = '/content/ladiescrop.pkl'\n",
+        "model_res = 1024\n",
+        "level = 0\n",
+        "blend_width=None\n",
+        "out = '/content/blend-frea-ladiestransfer-128.pkl'\n",
+        "\n",
+        "G_kwargs = dnnlib.EasyDict()\n",
+        "\n",
+        "with dnnlib.util.open_url(lo_res_pkl) as f:\n",
+        "    # G = legacy.load_network_pkl(f)['G_ema'].to(device) # type: ignore\n",
+        "    lo = legacy.load_network_pkl(f, custom=False, **G_kwargs) # type: ignore\n",
+        "    lo_G, lo_D, lo_G_ema = lo['G'], lo['D'], lo['G_ema']\n",
+        "\n",
+        "with dnnlib.util.open_url(hi_res_pkl) as f:\n",
+        "    # G = legacy.load_network_pkl(f)['G_ema'].to(device) # type: ignore\n",
+        "    hi = legacy.load_network_pkl(f, custom=False, **G_kwargs)['G_ema'] # type: ignore\n",
+        "    #hi_G, hi_D, hi_G_ema = hi['G'], lo['D'], lo['G_ema']\n",
+        "\n",
+        "rezes = [8,16,32,64,128]\n",
+        "for r in rezes: \n",
+        "    model_out = blend_models(lo_G_ema, hi, model_res, r, level, blend_width=blend_width)\n",
+        "\n",
+        "    # for n in model_out.named_parameters():\n",
+        "    #     print(n[0])\n",
+        "\n",
+        "    #save new pkl file\n",
+        "    out = f'/content/blend-frea-ladiestransfer-{r}.pkl'\n",
+        "    data = dict([('G', None), ('D', None), ('G_ema', None)])\n",
+        "    with open(out, 'wb') as f:\n",
+        "        #misc.save_pkl((low_res_G, low_res_D, out), output_pkl)\n",
+        "        data['G'] = lo_G\n",
+        "        data['D'] = lo_D\n",
+        "        data['G_ema'] = model_out\n",
+        "        pickle.dump(data, f)\n",
+        "\n"
+      ],
+      "execution_count": null,
+      "outputs": []
+    },
+    {
+      "cell_type": "code",
+      "metadata": {
+        "id": "M0XUCmyJUJ66"
+      },
+      "source": [
+        "## Test Generating Images With Your New Model"
+      ],
+      "execution_count": null,
+      "outputs": []
+    },
+    {
+      "cell_type": "code",
+      "metadata": {
+        "id": "qEe21is5r1qV"
+      },
+      "source": [
+        "for r in rezes:\n",
+        "    !python generate.py --outdir=/content/out/blended-frea-ladiestransfer2-{r}/ --trunc=0.6 --seeds=0-24 --network=/content/blend-frea-ladiestransfer-{r}.pkl"
+      ],
+      "execution_count": null,
+      "outputs": []
+    },
+    {
+      "cell_type": "code",
+      "metadata": {
+        "id": "B8WmzAmLsL8I"
+      },
+      "source": [
+        "!zip -r transferred-blends_r2.zip /content/out"
+      ],
+      "execution_count": null,
+      "outputs": []
+    },
+    {
+      "cell_type": "code",
+      "metadata": {
+        "id": "OF258_TAuFV1"
+      },
+      "source": [
+        ""
+      ],
+      "execution_count": null,
+      "outputs": []
+    }
+  ]
+}

README.md

@@ -1,3 +1,39 @@
+## Add to Colab
+
+### StyleGAN2-ADA PyTorch (Start Here)
+[![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/dvschultz/stylegan2-ada-pytorch/blob/main/SG2_ADA_PyTorch.ipynb)
+
+### StyleGAN2-ADA-PyTorch to Rosinality Convertor
+[![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/dvschultz/stylegan2-ada-pytorch/blob/main/SG2_ADA_PT_to_Rosinality.ipynb)
+
+### StyleGAN2-ADA PyTorch Reactive Audio
+[![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/dvschultz/stylegan2-ada-pytorch/blob/main/SG2-ADA-PT_AudioReactive%2BPitch.ipynb)
+
+### Blending Network Demo/Explainer
+[![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/dvschultz/stylegan2-ada-pytorch/blob/main/Network_Blending_ADA_PT.ipynb)
+
+## Need Help?
+* If you’re new to StyleGAN2-ADA and looking to get started, please check out [this video series](https://www.youtube.com/playlist?list=PLWuCzxqIpJs8ViuBIUtAk-dsAtdrApYoy) from a course Lia Coleman and I taught in October 2020.
+* Interested in contributing? Please submit PRs or discuss changes in the [Artificial Images Slack](https://join.slack.com/t/ml-images/shared_invite/zt-9mxoe7va-eBZ9xzAWpx8VRL~Km_PzUQ) channel
+
+## Edits made to this repo
+
+* **Fakes .jpg**: save yourself a ton of space with fakes during training saved as .jpg instead of .png
+* **Multiple interpolation options**: use `--process="interpolation"`, see `--help` for more options
+* **Easing options for interpolations**: see `--help` for more (this would be a great place for new coders to build additional feautures/options)
+* **Vertical Mirroring**: use `--mirrory=True` to flip training set top to bottom (fixed, thanks [Diego](https://github.com/PDillis/stylegan2-ada-pytorch))
+* **Set Initial Augmentation Strength**: use `--initstrength={float value}` to set the initialized strength of augmentations (really helpful when restarting training)
+* **Set Initial Kimg count**: use `--nkimg={int value}` to set the initial kimg count (helpful with restarts)
+* **Closed Form Factorization:** converted from Rosinality repo by [Philip Bizimis](https://github.com/pbizimis); additional video creation features
+* **Additional Projector Techniques** Thanks to Peter Baylies for his projector code that optionally uses pixel-based loss or CLIP
+* **Interpolate from Projector .npz Files** Use the `combine_npz.py` script to combine multiple .npz files
+* **Convert to Rosinality model structure** Thanks to [Justin Pinkney](https://github.com/justinpinkney) for making this! Converting to Rosinality opens up numerous additional tools for manipulatinng StyleGAN models
+* **Custom data in Tensorflow models will no longer error out when converting to Official PyTorch format** Allows for conversion from many of the custom models on the [Awesome Pretrained SG2 repo](https://github.com/justinpinkney/awesome-pretrained-stylegan2) (This has the potential for problems but I’ve yet to find one.)
+* **Output size modification** Thank you to [Vadim Epstein](https://github.com/eps696/stylegan2ada) for allowing me to port his code for modifying output size for image and video generation. See `--size` and `--scale-type` in the generation script for more details.
+* **Flesh digressions script** based on Aydao script (ported by [grddavies](https://github.com/grddavies), thanks!)
+* **Blend Two Models at specific resolution layer** Based on [Justin Pinkney’s Tensorflow version](https://github.com/justinpinkney/stylegan2/blob/master/blend_models.py)
+* **Experimental Top-K training function** Improve generator training by only propagating gradients from images the discriminator was most unsure of: [Sinha & Zhao](https://arxiv.org/abs/2002.06224). Use `--topk={float value}` to define the decay rate. Ported from code written by [Hans Brouwer](https://github.com/JCBrouwer) from [SG2-ADA Tensorflow version](https://github.com/dvschultz/stylegan2-ada/issues?q=is%3Apr+author%3AJCBrouwer).
+
 ## StyleGAN2-ADA — Official PyTorch implementation
 
 ![Teaser image](./docs/stylegan2-ada-teaser-1024x252.png)