sharp-training.ipynb

sharp-training.ipynb

{
  "nbformat": 4,
  "nbformat_minor": 0,
  "metadata": {
    "colab": {
      "provenance": [],
      "gpuType": "T4",
      "authorship_tag": "ABX9TyO8nnTCSF1uQuolZRvf9YpG",
      "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/gist/Mistobaan/77ce74663cba2d32682cdacac1ffe735/sharp-training.ipynb\" target=\"_parent\"><img src=\"https://colab.research.google.com/assets/colab-badge.svg\" alt=\"Open In Colab\"/></a>"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "ibLT2fAHeCp4"
      },
      "outputs": [],
      "source": [
        "!pip install condacolab"
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "!rm -fr /content/ml-sharp\n",
        "!git clone https://github.com/Mistobaan/ml-sharp /content/ml-sharp"
      ],
      "metadata": {
        "id": "oGIs7BQheGJ6"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "import condacolab\n",
        "condacolab.install()"
      ],
      "metadata": {
        "id": "H61mA2i1eYHl"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "!conda create -n sharp python=3.13"
      ],
      "metadata": {
        "id": "CVlcayaqeOf4"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "import condacolab\n",
        "condacolab.check()"
      ],
      "metadata": {
        "id": "sr5npHj-eTrs"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "%cd /content/ml-sharp"
      ],
      "metadata": {
        "id": "IlHSv9KWeLWJ"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "%pip install -r requirements.txt"
      ],
      "metadata": {
        "id": "I5f-5QD7efid"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "%env MPLBACKEND=notebook"
      ],
      "metadata": {
        "id": "bs0Fu3Dggazs"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "!sharp --help"
      ],
      "metadata": {
        "id": "5GaiBY1jesXD"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "!wget https://www.hollywoodreporter.com/wp-content/uploads/2018/11/detective_pikachu-screengrab-trailer-3-h_2018.jpg -O /content/input.jpg"
      ],
      "metadata": {
        "id": "Y3txJmVcgQbu"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "!sharp predict -i /content/input.jpg -o /content/examples/01 --render"
      ],
      "metadata": {
        "id": "UE249fZpgy-9"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "from IPython.display import Video\n",
        "\n",
        "Video(\"/content/examples/01/input.mp4\", embed=True)"
      ],
      "metadata": {
        "id": "0qkFOfFWhr0_"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "%cd /content/ml-sharp/\n",
        "!git pull\n",
        "!{sys.executable} -m pip install -e ."
      ],
      "metadata": {
        "id": "zZjfWEdF06MP"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "from pathlib import Path\n",
        "from sharp.utils.gsplat_notebook import show_ply, SplatViewerConfig\n",
        "\n",
        "show_ply(Path(\"/content/examples/01/input.ply\"), config=SplatViewerConfig(load_mode=\"inline\"))"
      ],
      "metadata": {
        "id": "UELj3CPGzWrd"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "import torch\n",
        "import numpy as np\n",
        "from plyfile import PlyData\n",
        "import matplotlib.pyplot as plt\n",
        "\n",
        "def torch_ptp(input):\n",
        "    \"\"\"Calculates peak-to-peak (max - min) for a torch tensor.\"\"\"\n",
        "    return torch.max(input) - torch.min(input)\n",
        "\n",
        "def load_ply_as_torch(path):\n",
        "    \"\"\"\n",
        "    Loads a PLY point cloud (with x,y,z and optional rgb) into a PyTorch tensor.\n",
        "    \"\"\"\n",
        "    ply = PlyData.read(path)\n",
        "    vertex_data = ply['vertex'].data\n",
        "\n",
        "    # Read xyz\n",
        "    xyz = np.vstack([vertex_data['x'], vertex_data['y'], vertex_data['z']]).T\n",
        "\n",
        "    # Optional RGB if present\n",
        "    if {'red','green','blue'}.issubset(vertex_data.dtype.names):\n",
        "        rgb = np.vstack([\n",
        "            vertex_data['red'],\n",
        "            vertex_data['green'],\n",
        "            vertex_data['blue']\n",
        "        ]).T / 255.0\n",
        "        data = np.hstack([xyz, rgb])\n",
        "    else:\n",
        "        data = xyz\n",
        "\n",
        "    return torch.from_numpy(data).float()\n",
        "\n",
        "def simple_project(points, image_size=(512,512)):\n",
        "    \"\"\"\n",
        "    Projects 3D points into 2D with a simple orthographic projection.\n",
        "    - points: Tensor (N,3)\n",
        "    - image_size: (H,W)\n",
        "    \"\"\"\n",
        "    # Center and scale to fit image\n",
        "    pts = points.clone()\n",
        "    pts -= pts.mean(0)\n",
        "    scale = max(torch_ptp(pts[:,0]), torch_ptp(pts[:,1]))\n",
        "    pts[:,:2] /= (scale + 1e-6)\n",
        "\n",
        "    # Convert to pixel coords\n",
        "    H,W = image_size\n",
        "    u = ((pts[:,0] * 0.5 + 0.5) * (W-1)).long()\n",
        "    v = ((-pts[:,1] * 0.5 + 0.5) * (H-1)).long()\n",
        "\n",
        "    return u.clamp(0,W-1), v.clamp(0,H-1)\n",
        "\n",
        "def render_pointcloud(points, colors=None, image_size=(512,512)):\n",
        "    \"\"\"\n",
        "    Renders a simple 2D image from a point cloud.\n",
        "    - points: Tensor (N,3)\n",
        "    - colors: Tensor (N,3) optional\n",
        "    \"\"\"\n",
        "    H,W = image_size\n",
        "    image = torch.zeros((H,W,3))\n",
        "\n",
        "    u,v = simple_project(points, image_size)\n",
        "    for i in range(u.shape[0]):\n",
        "        if colors is not None:\n",
        "            image[v[i], u[i]] = colors[i]\n",
        "        else:\n",
        "            image[v[i], u[i]] = torch.tensor([1.0,1.0,1.0])\n",
        "\n",
        "    return image"
      ],
      "metadata": {
        "id": "Su_I-zIFtpik"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "# Path to your PLY file\n",
        "ply_path = \"/content/examples/01/input.ply\"\n",
        "\n",
        "data = load_ply_as_torch(ply_path)\n",
        "\n",
        "if data.shape[1] == 6:\n",
        "    points, colors = data[:,:3], data[:,3:]\n",
        "else:\n",
        "    points, colors = data[:,:3], None\n",
        "\n",
        "img = render_pointcloud(points, colors)\n",
        "\n",
        "plt.figure(figsize=(6,6))\n",
        "plt.imshow(img.numpy())\n",
        "plt.axis('off')\n",
        "plt.show()"
      ],
      "metadata": {
        "id": "q52VJwfQyeQP"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [],
      "metadata": {
        "id": "rolFkhLc3qwB"
      },
      "execution_count": null,
      "outputs": []
    }
  ]
}