Add shapes.ipynb notebook

shapes.ipynb

@@ -0,0 +1,539 @@
+{
+ "cells": [
+  {
+   "cell_type": "code",
+   "execution_count": 416,
+   "metadata": {},
+   "outputs": [
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "\u001b[2mUsing Python 3.12.10 environment at: /Users/andrew/Developer/ML/whale/.venv\u001b[0m\n",
+      "\u001b[2mAudited \u001b[1m3 packages\u001b[0m \u001b[2min 11ms\u001b[0m\u001b[0m\n"
+     ]
+    }
+   ],
+   "source": [
+    "!uv pip install cairosvg torch colour-science"
+   ]
+  },
+  {
+   "attachments": {},
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "Dataset consisting of\n",
+    "- Various single shapes\n",
+    "    - Square\n",
+    "    - Rounded rectangle\n",
+    "    - Circle\n",
+    "- at various coordinates\n",
+    "\n",
+    "Output:\n",
+    "- text prompt\n",
+    "- png images of the shapes\n",
+    "- svg text for the shapes"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 417,
+   "metadata": {},
+   "outputs": [
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "Failures in 1000 attempts: 0\n",
+      "Contrast ratio: 2.72\n"
+     ]
+    },
+    {
+     "data": {
+      "image/svg+xml": [
+       "<svg width=\"100\" height=\"100\" style=\"background-color: #adf3b3;\"><rect x=\"30\" y=\"30\" width=\"40\" height=\"40\" fill=\"#0857fe\"/></svg>"
+      ],
+      "text/plain": [
+       "<IPython.core.display.SVG object>"
+      ]
+     },
+     "execution_count": 417,
+     "metadata": {},
+     "output_type": "execute_result"
+    }
+   ],
+   "source": [
+    "import torch.random\n",
+    "import colour\n",
+    "from IPython.display import SVG\n",
+    "import torch.random\n",
+    "\n",
+    "def uniform(low=0, high=1):\n",
+    "    return torch.rand(1).item() * (high - low) + low\n",
+    "\n",
+    "\n",
+    "def random_color():\n",
+    "    r = uniform(0.0, 1.0)\n",
+    "    g = uniform(0.0, 1.0)\n",
+    "    b = uniform(0.0, 1.0)\n",
+    "    return r, g, b\n",
+    "\n",
+    "def hex(r: float, g: float, b: float):\n",
+    "    return f\"#{int(r*255):02x}{int(g*255):02x}{int(b*255):02x}\"\n",
+    "\n",
+    "ColorType = tuple[float, float, float]\n",
+    "\n",
+    "def contrast_ratio(bg: ColorType, color: ColorType):\n",
+    "    r1, g1, b1 = bg\n",
+    "    r2, g2, b2 = color\n",
+    "    # Convert to CIE XYZ\n",
+    "    _, bg_lum, _ = colour.RGB_to_XYZ([r1, g1, b1], colour.models.RGB_COLOURSPACES['sRGB'].whitepoint, colour.models.RGB_COLOURSPACES['sRGB'].whitepoint, colour.models.RGB_COLOURSPACES['sRGB'].matrix_RGB_to_XYZ)\n",
+    "    _, fg_lum, _ = colour.RGB_to_XYZ([r2, g2, b2], colour.models.RGB_COLOURSPACES['sRGB'].whitepoint, colour.models.RGB_COLOURSPACES['sRGB'].whitepoint, colour.models.RGB_COLOURSPACES['sRGB'].matrix_RGB_to_XYZ)\n",
+    "    # Relative luminance is the Y component in XYZ\n",
+    "    L1, L2 = max(bg_lum, fg_lum), min(bg_lum, fg_lum)\n",
+    "    eps = 1e-6\n",
+    "    return (L1 + eps) / (L2 + eps)\n",
+    "\n",
+    "def contrasting_color(background: ColorType, threshold=2.0, max_attempts=100):\n",
+    "    \"\"\"\n",
+    "    Generate a random color with contrast ratio above the threshold with the given background.\n",
+    "    The threshold has been tuned so that it doesn't fail for random colors, if you have fixed colors you can increase the threshold\n",
+    "    \"\"\"\n",
+    "    for _ in range(max_attempts):\n",
+    "        color = random_color()\n",
+    "        ratio = contrast_ratio(background, color)\n",
+    "        if ratio >= threshold:\n",
+    "            return color, ratio\n",
+    "    raise ValueError(f'No color found with contrast ratio >= {threshold} for background {background} in {max_attempts} attempts')\n",
+    "\n",
+    "def check_generate_contrast(tests=1000, max_attempts=100):\n",
+    "    failures_per_1000 = 0\n",
+    "    for _ in range(tests):\n",
+    "        try:\n",
+    "            contrasting_color(random_color(), max_attempts=max_attempts)\n",
+    "        except ValueError:\n",
+    "            failures_per_1000 += 1\n",
+    "    return failures_per_1000\n",
+    "\n",
+    "print(f\"Failures in 1000 attempts: {check_generate_contrast(tests=1000, max_attempts=100)}\")\n",
+    "\n",
+    "test_color = random_color()\n",
+    "test_bg_color, ratio = contrasting_color(test_color)\n",
+    "print(f'Contrast ratio: {ratio:.2f}')\n",
+    "SVG(f'<svg width=\"100\" height=\"100\" style=\"background-color: {hex(*test_bg_color)};\"><rect x=\"30\" y=\"30\" width=\"40\" height=\"40\" fill=\"{hex(*test_color)}\" /></svg>')"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 418,
+   "metadata": {},
+   "outputs": [
+    {
+     "name": "stderr",
+     "output_type": "stream",
+     "text": [
+      "100%|██████████| 10000/10000 [00:04<00:00, 2071.32it/s]"
+     ]
+    },
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "Generated 10000 examples\n",
+      "a line from (255.0, 219.0) to (365.0, 347.0) with stroke #210949 and width 3.9 within a 512 x 512 box of color \"white\"\n",
+      "<svg width=\"512\" height=\"512\" viewBox=\"0 0 512 512\" xmlns=\"http://www.w3.org/2000/svg\">\n",
+      "  <rect width=\"512\" height=\"512\" fill=\"#ffffff\" />\n",
+      "  <line x1=\"255.0\" y1=\"219.0\" x2=\"365.0\" y2=\"347.0\" stroke=\"#210949\" stroke-width=\"3.9\" />\n",
+      "</svg>\n",
+      "line\n",
+      "#ffffff\n",
+      "#f00314\n"
+     ]
+    },
+    {
+     "name": "stderr",
+     "output_type": "stream",
+     "text": [
+      "\n"
+     ]
+    },
+    {
+     "data": {
+      "image/svg+xml": [
+       "<svg xmlns=\"http://www.w3.org/2000/svg\" width=\"512\" height=\"512\" viewBox=\"0 0 512 512\">\n",
+       "  <rect width=\"512\" height=\"512\" fill=\"#ffffff\"/>\n",
+       "  <line x1=\"255.0\" y1=\"219.0\" x2=\"365.0\" y2=\"347.0\" stroke=\"#210949\" stroke-width=\"3.9\"/>\n",
+       "</svg>"
+      ],
+      "text/plain": [
+       "<IPython.core.display.SVG object>"
+      ]
+     },
+     "execution_count": 418,
+     "metadata": {},
+     "output_type": "execute_result"
+    }
+   ],
+   "source": [
+    "from typing import Any\n",
+    "from tqdm import tqdm\n",
+    "\n",
+    "shape_types = [\n",
+    "    (0.5, 'circle'),\n",
+    "    (0.8, 'rect'),\n",
+    "    (1, 'rounded-rect'),\n",
+    "    (0.25, 'line'),\n",
+    "]\n",
+    "\n",
+    "backgrounds = [\n",
+    "    (1.0, ('black', (0, 0, 0))),\n",
+    "    (0.5, ('#000000', (0, 0, 0))),\n",
+    "    (1.0, ('white', (1.0, 1.0, 1.0))),\n",
+    "    (0.5, ('#ffffff', (1.0, 1.0, 1.0))),\n",
+    "    (0.5, ('gray', (0.5, 0.5, 0.5))),\n",
+    "    (0.5, (\"RANDOM\", None)),\n",
+    "]\n",
+    "\n",
+    "\n",
+    "def sample(options: list, value: float) -> Any:\n",
+    "    \"\"\" Takes a random sample from a list of (weight, option) tuples\"\"\"\n",
+    "    total_weight = sum(weight for weight, _ in options)\n",
+    "    options_normalized = [(weight / total_weight, option) for weight, option in options]\n",
+    "    v = value\n",
+    "    for weight, option in options_normalized:\n",
+    "        if v < weight:\n",
+    "            return option\n",
+    "        v -= weight\n",
+    "    raise ValueError(f\"Value {v} is out of range for options {options_normalized}\")\n",
+    "\n",
+    "# iterate over the shapes and make random examples\n",
+    "def shape_data(num_examples=1000, use_background=False): \n",
+    "    c_width = 512\n",
+    "    c_height = 512\n",
+    "    margin_x = c_width * 0.02\n",
+    "    margin_y = c_height * 0.02\n",
+    "    def wrap(svg: str, background: ColorType):\n",
+    "        background_str = f'\\n  <rect width=\"{c_width}\" height=\"{c_height}\" fill=\"{hex(*background)}\" />' if background else ''\n",
+    "        return f'<svg width=\"{c_width}\" height=\"{c_height}\" viewBox=\"0 0 {c_width} {c_height}\" xmlns=\"http://www.w3.org/2000/svg\">{background_str}\\n  {svg}\\n</svg>'\n",
+    "    # normalize shape_types\n",
+    "    for _ in tqdm(range(num_examples)):\n",
+    "        shape_name = sample(shape_types, uniform())\n",
+    "        bg_name, bg_color = sample(backgrounds, uniform()) if use_background else (None, None)\n",
+    "        if bg_name == \"RANDOM\":\n",
+    "            bg_color = random_color()\n",
+    "            bg_name = hex(*bg_color)\n",
+    "        color, _ = contrasting_color(background=bg_color) if bg_color else random_color()\n",
+    "        bg_desc = f'within a {c_width} x {c_height} box of color \"{bg_name}\"'\n",
+    "        meta = (shape_name, hex(*bg_color), hex(*color))\n",
+    "        if shape_name == 'circle':\n",
+    "            # Choose a radius > 0, and ensure the circle is strictly inside the bounds\n",
+    "            max_r = min(c_width - 2 * margin_x, c_height - 2 * margin_y) / 2\n",
+    "            r = uniform(8, max_r)  # radius at least 8\n",
+    "            r = round(r, 0)\n",
+    "            # x and y must be at least r from the edge, and at most width - r\n",
+    "            x = uniform(r, c_width - r)\n",
+    "            x = round(x, 0)\n",
+    "            y = uniform(r, c_height - r)\n",
+    "            y = round(y, 0)\n",
+    "            desc = f'a circle with radius {r} at ({x}, {y}) filled with {hex(*color)} {bg_desc}'\n",
+    "            svg = f'<circle cx=\"{x}\" cy=\"{y}\" r=\"{r}\" fill=\"{hex(*color)}\" />'\n",
+    "            yield desc, wrap(svg, bg_color), *meta\n",
+    "        elif shape_name == \"rect\":\n",
+    "            # Constrain to leave at least 2% margin on any side\n",
+    "            max_w = c_width - 2 * margin_x\n",
+    "            max_h = c_height - 2 * margin_y\n",
+    "            w = uniform(max_w * 0.1, max_w)  # width at least 10% of max_w\n",
+    "            w = round(w, 0)\n",
+    "            h = uniform(max_h * 0.1, max_h)  # height at least 10% of max_h\n",
+    "            h = round(h, 0)\n",
+    "            x = uniform(margin_x, c_width - margin_x - w)\n",
+    "            x = round(x, 0)\n",
+    "            y = uniform(margin_y, c_height - margin_y - h)\n",
+    "            y = round(y, 0)\n",
+    "            desc = f'a rectangle with width {w} and height {h} at ({x}, {y}) filled with {hex(*color)} {bg_desc}'\n",
+    "            svg = f'<rect x=\"{x}\" y=\"{y}\" width=\"{w}\" height=\"{h}\" fill=\"{hex(*color)}\" />'\n",
+    "            yield desc, wrap(svg, bg_color), *meta\n",
+    "        elif shape_name == 'rounded-rect':\n",
+    "            # Constrain to leave at least 2% margin on any side\n",
+    "            max_w = c_width - 2 * margin_x\n",
+    "            max_h = c_height - 2 * margin_y\n",
+    "            min_w = max_w * 0.1\n",
+    "            min_h = max_h * 0.1\n",
+    "            w = uniform(min_w, max_w)\n",
+    "            w = round(w, 0)\n",
+    "            h = uniform(min_h, max_h)\n",
+    "            h = round(h, 0)\n",
+    "            # x and y must be such that the rect stays within the canvas\n",
+    "            x = uniform(margin_x, c_width - margin_x - w)\n",
+    "            x = round(x, 0)\n",
+    "            y = uniform(margin_y, c_height - margin_y - h)\n",
+    "            y = round(y, 0)\n",
+    "            r = uniform(0, min(w, h) / 2)\n",
+    "            r = round(r, 1)\n",
+    "            desc = f'a rounded rectangle with width {w}, height {h}, and radius {r} at ({x}, {y}) filled with {hex(*color)} {bg_desc}'\n",
+    "            svg = f'<rect x=\"{x}\" y=\"{y}\" width=\"{w}\" height=\"{h}\" rx=\"{r}\" ry=\"{r}\" fill=\"{hex(*color)}\" />'\n",
+    "            yield desc, wrap(svg, bg_color), *meta\n",
+    "        elif shape_name == 'line':\n",
+    "            # Constrain to leave at least 2% margin on any side\n",
+    "            x1 = uniform(margin_x, c_width - margin_x)\n",
+    "            x1 = round(x1, 0)\n",
+    "            y1 = uniform(margin_y, c_height - margin_y)\n",
+    "            y1 = round(y1, 0)\n",
+    "            x2 = uniform(margin_x, c_width - margin_x)\n",
+    "            x2 = round(x2, 0)\n",
+    "            y2 = uniform(margin_y, c_height - margin_y)\n",
+    "            y2 = round(y2, 0)\n",
+    "            color = random_color()\n",
+    "            stroke_width = round(uniform(1, 6), 1)\n",
+    "            desc = f'a line from ({x1}, {y1}) to ({x2}, {y2}) with stroke {hex(*color)} and width {stroke_width} {bg_desc}'\n",
+    "            svg = f'<line x1=\"{x1}\" y1=\"{y1}\" x2=\"{x2}\" y2=\"{y2}\" stroke=\"{hex(*color)}\" stroke-width=\"{stroke_width}\" />'\n",
+    "            yield desc, wrap(svg, bg_color), *meta\n",
+    "    \n",
+    "\n",
+    "examples = list(shape_data(10000, use_background=True))\n",
+    "print(f'Generated {len(examples)} examples')\n",
+    "\n",
+    "print('\\n'.join(examples[0]))\n",
+    "SVG(examples[0][1])"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 419,
+   "metadata": {},
+   "outputs": [
+    {
+     "name": "stderr",
+     "output_type": "stream",
+     "text": [
+      "100%|██████████| 10000/10000 [01:29<00:00, 112.01it/s]\n"
+     ]
+    }
+   ],
+   "source": [
+    "import os\n",
+    "from tqdm import tqdm\n",
+    "import json\n",
+    "os.environ[\"DYLD_LIBRARY_PATH\"] = \"/opt/homebrew/lib\"\n",
+    "import cairosvg\n",
+    "\n",
+    "\n",
+    "def generate_images(output_dir=\"./data/shapes\"):\n",
+    "    os.makedirs(output_dir, exist_ok=True)\n",
+    "\n",
+    "    # delete all the old files\n",
+    "    for file in os.listdir(output_dir):\n",
+    "        os.remove(f\"{output_dir}/{file}\")\n",
+    "\n",
+    "    def render_example(svg, i):\n",
+    "        file_name = f\"{i:06d}.png\"\n",
+    "        temp_path = f\"{output_dir}/{file_name}\"\n",
+    "        cairosvg.svg2png(svg, write_to=temp_path)\n",
+    "        return file_name\n",
+    "\n",
+    "\n",
+    "    with open(f\"{output_dir}/metadata.jsonl\", \"w\") as f:\n",
+    "        for i, example in tqdm(enumerate(examples), total=len(examples)):\n",
+    "            #print(\"rendering\", example[1])\n",
+    "            img = render_example(example[1], i)\n",
+    "            # Add metadata to metadata.jsonl\n",
+    "            f.write(json.dumps({\"desc\": example[0], \"svg\": example[1], \"file_name\": img, \"shape_name\": example[2], \"bg_color\": example[3], \"color\": example[4]}) + \"\\n\")\n",
+    "            if i%100 == 0:\n",
+    "                f.flush()\n",
+    "        #print(example[0], render_example(example[1]))\n",
+    "generate_images()"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 423,
+   "metadata": {},
+   "outputs": [
+    {
+     "data": {
+      "image/png": "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",
+      "text/plain": [
+       "<IPython.core.display.Image object>"
+      ]
+     },
+     "execution_count": 423,
+     "metadata": {},
+     "output_type": "execute_result"
+    }
+   ],
+   "source": [
+    "from IPython.display import Image\n",
+    "cairosvg.svg2png(examples[0][1], write_to=\"./output.png\")\n",
+    "Image(filename=\"./output.png\")"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 421,
+   "metadata": {},
+   "outputs": [
+    {
+     "data": {
+      "application/vnd.jupyter.widget-view+json": {
+       "model_id": "ee645e163c8144498689b13747ecd2f2",
+       "version_major": 2,
+       "version_minor": 0
+      },
+      "text/plain": [
+       "Resolving data files:   0%|          | 0/10001 [00:00<?, ?it/s]"
+      ]
+     },
+     "metadata": {},
+     "output_type": "display_data"
+    },
+    {
+     "data": {
+      "application/vnd.jupyter.widget-view+json": {
+       "model_id": "d79680a43a5d4de0a917de4c03da6bb2",
+       "version_major": 2,
+       "version_minor": 0
+      },
+      "text/plain": [
+       "Downloading data:   0%|          | 0/10001 [00:00<?, ?files/s]"
+      ]
+     },
+     "metadata": {},
+     "output_type": "display_data"
+    },
+    {
+     "data": {
+      "application/vnd.jupyter.widget-view+json": {
+       "model_id": "637f82e0971543f5929c786c8c40b047",
+       "version_major": 2,
+       "version_minor": 0
+      },
+      "text/plain": [
+       "Generating train split: 0 examples [00:00, ? examples/s]"
+      ]
+     },
+     "metadata": {},
+     "output_type": "display_data"
+    }
+   ],
+   "source": [
+    "# Create Dataset object\n",
+    "from datasets import load_dataset\n",
+    "\n",
+    "shapes_dataset = load_dataset(\"imagefolder\", data_dir=\"./data/shapes\")\n"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 422,
+   "metadata": {},
+   "outputs": [
+    {
+     "data": {
+      "application/vnd.jupyter.widget-view+json": {
+       "model_id": "a271274070c84c7fb7d2fe8cdb7d471c",
+       "version_major": 2,
+       "version_minor": 0
+      },
+      "text/plain": [
+       "Uploading the dataset shards:   0%|          | 0/1 [00:00<?, ?it/s]"
+      ]
+     },
+     "metadata": {},
+     "output_type": "display_data"
+    },
+    {
+     "data": {
+      "application/vnd.jupyter.widget-view+json": {
+       "model_id": "119330c1626b4ea28fea8418df8210de",
+       "version_major": 2,
+       "version_minor": 0
+      },
+      "text/plain": [
+       "Map:   0%|          | 0/10000 [00:00<?, ? examples/s]"
+      ]
+     },
+     "metadata": {},
+     "output_type": "display_data"
+    },
+    {
+     "data": {
+      "application/vnd.jupyter.widget-view+json": {
+       "model_id": "32e80b41ef544d64b5b3a5f21e4ef15b",
+       "version_major": 2,
+       "version_minor": 0
+      },
+      "text/plain": [
+       "Creating parquet from Arrow format:   0%|          | 0/100 [00:00<?, ?ba/s]"
+      ]
+     },
+     "metadata": {},
+     "output_type": "display_data"
+    },
+    {
+     "data": {
+      "application/vnd.jupyter.widget-view+json": {
+       "model_id": "8c2a3a613ea047059ead76da4b191636",
+       "version_major": 2,
+       "version_minor": 0
+      },
+      "text/plain": [
+       "Uploading...:   0%|          | 0.00/29.5M [00:00<?, ?B/s]"
+      ]
+     },
+     "metadata": {},
+     "output_type": "display_data"
+    },
+    {
+     "data": {
+      "application/vnd.jupyter.widget-view+json": {
+       "model_id": "27e4f630d9b340fe80353ed8b11b91f8",
+       "version_major": 2,
+       "version_minor": 0
+      },
+      "text/plain": [
+       "README.md:   0%|          | 0.00/347 [00:00<?, ?B/s]"
+      ]
+     },
+     "metadata": {},
+     "output_type": "display_data"
+    },
+    {
+     "data": {
+      "text/plain": [
+       "CommitInfo(commit_url='https://huggingface.co/datasets/darknoon/simple-shapes-svg/commit/d9f47eeaa9b84d78c01cbe0df90d3413ea57ce62', commit_message='Upload dataset', commit_description='', oid='d9f47eeaa9b84d78c01cbe0df90d3413ea57ce62', pr_url=None, repo_url=RepoUrl('https://huggingface.co/datasets/darknoon/simple-shapes-svg', endpoint='https://huggingface.co', repo_type='dataset', repo_id='darknoon/simple-shapes-svg'), pr_revision=None, pr_num=None)"
+      ]
+     },
+     "execution_count": 422,
+     "metadata": {},
+     "output_type": "execute_result"
+    }
+   ],
+   "source": [
+    "shapes_dataset.push_to_hub(\"darknoon/simple-shapes-svg\")"
+   ]
+  }
+ ],
+ "metadata": {
+  "kernelspec": {
+   "display_name": ".venv",
+   "language": "python",
+   "name": "python3"
+  },
+  "language_info": {
+   "codemirror_mode": {
+    "name": "ipython",
+    "version": 3
+   },
+   "file_extension": ".py",
+   "mimetype": "text/x-python",
+   "name": "python",
+   "nbconvert_exporter": "python",
+   "pygments_lexer": "ipython3",
+   "version": "3.12.10"
+  },
+  "orig_nbformat": 4
+ },
+ "nbformat": 4,
+ "nbformat_minor": 2
+}