Add conda environment file

README.md

@@ -1 +1,12 @@
-# diaxi-training-tools
+# diaxi-training-tools
+
+Herramientas para la creación del dataset y el entrenamiento del modelo de Desafío IA por la Identidad
+
+## Instalación
+
+> Requiere tener instalado [**Miniconda**](https://docs.conda.io/en/latest/miniconda.html) o una variante de [**Miniforge**](https://github.com/conda-forge/miniforge/releases/latest) (se recomienda `mambaforge`).
+
+```
+$ mamba env create -f environment.yml
+$ conda activate diaxi-training-tools
+```

create_diaxi_dataset.ipynb

@@ -70,6 +70,53 @@
     "len(lost)  # Perdimos 51 archivos en el camino :("
    ]
   },
+  {
+   "attachments": {},
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "## 2. Enderezar las imágenes"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "import os\n",
+    "import math\n",
+    "from typing import Tuple, Union\n",
+    "\n",
+    "import cv2\n",
+    "import numpy as np\n",
+    "from deskew import determine_skew\n",
+    "\n",
+    "THRESHOLD = 10  # límite de 10 grados\n",
+    "\n",
+    "def rotate(image: np.ndarray, angle: float, background: Union[int, Tuple[int, int, int]]) -> np.ndarray:\n",
+    "    old_width, old_height = image.shape[:2]\n",
+    "    angle_radian = math.radians(angle)\n",
+    "    width = abs(np.sin(angle_radian) * old_height) + abs(np.cos(angle_radian) * old_width)\n",
+    "    height = abs(np.sin(angle_radian) * old_width) + abs(np.cos(angle_radian) * old_height)\n",
+    "\n",
+    "    image_center = tuple(np.array(image.shape[1::-1]) / 2)\n",
+    "    rot_mat = cv2.getRotationMatrix2D(image_center, angle, 1.0)\n",
+    "    rot_mat[1, 2] += (width - old_width) / 2\n",
+    "    rot_mat[0, 2] += (height - old_height) / 2\n",
+    "    \n",
+    "    return cv2.warpAffine(image, rot_mat, (int(round(height)), int(round(width))), borderValue=background)\n",
+    "\n",
+    "for i in os.listdir(\"fixed/images/\"):\n",
+    "    image = cv2.imread(f\"fixed/images/{i}\")\n",
+    "    grayscale = cv2.cvtColor(image, cv2.COLOR_BGR2GRAY)\n",
+    "    angle = determine_skew(grayscale)\n",
+    "\n",
+    "    if abs(angle) < THRESHOLD:\n",
+    "        rotated = rotate(image, angle, (0, 0, 0))\n",
+    "        cv2.imwrite(f\"fixed/images/{i}\", rotated)"
+   ]
+  },
   {
    "attachments": {},
    "cell_type": "markdown",
@@ -86,7 +133,6 @@
    "metadata": {},
    "outputs": [],
    "source": [
-    "import os\n",
     "import json\n",
     "from PIL import Image\n",
     "\n",
@@ -310,51 +356,6 @@
     "    json.dump(result, f, indent=2)"
    ]
   },
-  {
-   "attachments": {},
-   "cell_type": "markdown",
-   "metadata": {},
-   "source": [
-    "## 3. Enderezar las imágenes\n",
-    "\n",
-    "Se procede a enderezar las imágenes con el script `deskew.py`."
-   ]
-  },
-  {
-   "cell_type": "code",
-   "execution_count": null,
-   "metadata": {},
-   "outputs": [],
-   "source": [
-    "import math\n",
-    "from typing import Tuple, Union\n",
-    "\n",
-    "import cv2\n",
-    "import numpy as np\n",
-    "from deskew import determine_skew\n",
-    "\n",
-    "\n",
-    "def rotate(image: np.ndarray, angle: float, background: Union[int, Tuple[int, int, int]]) -> np.ndarray:\n",
-    "    old_width, old_height = image.shape[:2]\n",
-    "    angle_radian = math.radians(angle)\n",
-    "    width = abs(np.sin(angle_radian) * old_height) + abs(np.cos(angle_radian) * old_width)\n",
-    "    height = abs(np.sin(angle_radian) * old_width) + abs(np.cos(angle_radian) * old_height)\n",
-    "\n",
-    "    image_center = tuple(np.array(image.shape[1::-1]) / 2)\n",
-    "    rot_mat = cv2.getRotationMatrix2D(image_center, angle, 1.0)\n",
-    "    rot_mat[1, 2] += (width - old_width) / 2\n",
-    "    rot_mat[0, 2] += (height - old_height) / 2\n",
-    "    \n",
-    "    return cv2.warpAffine(image, rot_mat, (int(round(height)), int(round(width))), borderValue=background)\n",
-    "\n",
-    "for i in os.listdir(\"fixed/images\"):\n",
-    "    image = cv2.imread(f\"fixed/images/{i}\")\n",
-    "    grayscale = cv2.cvtColor(image, cv2.COLOR_BGR2GRAY)\n",
-    "    angle = determine_skew(grayscale)\n",
-    "    rotated = rotate(image, angle, (0, 0, 0))\n",
-    "    cv2.imwrite(f\"fixed/images/{i}\", rotated)"
-   ]
-  },
   {
    "attachments": {},
    "cell_type": "markdown",
@@ -466,7 +467,7 @@
     "cp -r fixed/images dataset-diaxi-coco\n",
     "cp fixed/result_final.json dataset-diaxi-coco/result.json\n",
     "cd dataset-diaxi-coco\n",
-    "zip -r ../dataset-diaxi-coco333.zip images result.json"
+    "zip -r ../dataset-diaxi-coco.zip images result.json"
    ]
   }
  ],

environment.yml

@@ -0,0 +1,18 @@
+name: diaxi-training-tools
+
+channels:
+    - pytorch
+    - nvidia
+    - conda-forge
+
+dependencies:
+    - python =3.8
+    - pip
+    - opencv
+    - pytorch >=1.8
+    - torchvision
+    - torchaudio
+    - pytorch-cuda =11.7
+
+    - pip:
+        - -r requirements.txt

requirements.txt

@@ -1,5 +1,3 @@
 deskew
-ipykernel
-pillow
-opencv-python
 pyodi
+git+https://github.com/facebookresearch/[email protected]#egg=detectron2