Tfhub object detection improvements

deployment/tfhub/object-detection/base64/.gitignore

@@ -1 +1,2 @@
 .env/
+images/

deployment/tfhub/object-detection/base64/object-detection-base64-deploy.ipynb

@@ -2,17 +2,22 @@
  "cells": [
   {
    "cell_type": "markdown",
-   "metadata": {},
+   "metadata": {
+    "id": "99FgJuGtHbK-"
+   },
    "source": [
-    "# Object Identification (TensorFlow Hub) - Base64 Images"
+    "# Object Detection - SSD Mobilenet V2 from TensorFlow Hub (Base64 Images)"
    ]
   },
   {
    "cell_type": "markdown",
-   "metadata": {},
+   "metadata": {
+    "id": "aYVICIPxIqAM"
+   },
    "source": [
-    "This notebook identifies whether a given object is present in a set of images. If the object is found, it returns a dataframe with the file name, score and bounding boxes.\n",
-    "We are using a dataset from [UCF](https://www.crcv.ucf.edu/data/GMCP_Geolocalization/#Dataset) that contains 62,058 high quality Google Street View images, and [TensorFlow Hub](https://www.tensorflow.org/hub), a library and platform for transfer learning."
+    "This model identifies objects present in images, returning its scores, labels and bounding boxes.\n",
+    "\n",
+    "We are using a dataset from [UCF](https://www.crcv.ucf.edu/data/GMCP_Geolocalization/#Dataset) and the model [SSD Mobilenet V2](https://tfhub.dev/google/openimages_v4/ssd/mobilenet_v2/1) from [TensorFlow Hub](https://www.tensorflow.org/hub)."
    ]
   },
   {
@@ -33,7 +38,7 @@
    "cell_type": "markdown",
    "metadata": {},
    "source": [
-    "This notebook has been tested with Python 3.10.6 and the following package versions:"
+    "This notebook has been tested with **Python 3.8.15** and the following package versions:"
    ]
   },
   {
@@ -42,8 +47,9 @@
    "metadata": {},
    "outputs": [],
    "source": [
+    "%%capture\n",
     "!pip install dill==0.3.6\n",
-    "!pip install Pillow==9.3.0\n",
+    "!pip install Pillow==7.1.2\n",
     "!pip install tensorflow==2.11.0\n",
     "!pip install tensorflow-hub==0.12.0\n",
     "!pip install verta==0.21.1\n",
@@ -111,8 +117,8 @@
     "        img_arr = np.array(Image.open(img_bytes), dtype=np.uint8)\n",
     "        img = Image.fromarray(img_arr)\n",
     "        img = ImageOps.grayscale(img)\n",
-    "        img.thumbnail((width, height), Image.Resampling.LANCZOS)\n",
-    "        img.save(path, format = 'JPEG', quality = 90)\n",
+    "        img.thumbnail((width, height), Image.LANCZOS)\n",
+    "        img.save(path)\n",
     "        \n",
     "        print(f\"Image downloaded to {path}.\")\n",
     "        return path\n",
@@ -179,38 +185,7 @@
     "        file, img_arr = data\n",
     "        result = self.detect_objects(file, img_arr)\n",
     "\n",
-    "        return result\n",
-    "\n",
-    "    def describe(self):\n",
-    "        \"\"\"Return a description of the service.\"\"\"        \n",
-    "        return {\n",
-    "            \"method\": \"predict\",\n",
-    "            \"args\": f\"{self.example()}\",\n",
-    "            \"returns\": \"file, has_car, score, ymin, xmin, ymax, xmax\",\n",
-    "            \"description\": \"\"\"\n",
-    "                Identify whether a given object is present in the image, with its score and bounding boxes.\n",
-    "            \"\"\",\n",
-    "            \"input_description\": \"\"\"\n",
-    "                A list with base64 images.\n",
-    "            \"\"\",\n",
-    "            \"output_description\": \"\"\"\n",
-    "                A CSV file with information about all the processed images.\n",
-    "                The columns presents the file names, if the object was found (boolean), its score and bounding boxes.\n",
-    "            \"\"\"\n",
-    "        }\n",
-    "    \n",
-    "    def example(self):\n",
-    "        \"\"\"Return example input json-serializable data.\"\"\"\n",
-    "        url = 'http://s3.amazonaws.com/verta-starter/street-view-images/000001_0.jpg'\n",
-    "        file = url.split('/')[-1]\n",
-    "        wget.download(url, file)\n",
-    "\n",
-    "        with open(file, 'rb') as img:\n",
-    "            img_bytes = base64.b64encode(img.read())\n",
-    "            img_str = img_bytes.decode('utf-8')\n",
-    "            img_str = json.dumps(img_str)\n",
-    "        \n",
-    "        return [\"000001_0.jpg\", img_str]"
+    "        return result"
    ]
   },
   {
@@ -349,7 +324,7 @@
    "source": [
     "assert status['status'] == \"active\"\n",
     "url = 'http://s3.amazonaws.com/verta-starter/street-view-images/000001_0.jpg'\n",
-    "file = url.split('/')[-1]\n",
+    "file = os.path.basename(url)\n",
     "wget.download(url, file)\n",
     "\n",
     "with open(file, 'rb') as img:\n",
@@ -364,7 +339,7 @@
  ],
  "metadata": {
   "kernelspec": {
-   "display_name": "Python 3 (ipykernel)",
+   "display_name": "Python 3.8.15 ('.env': venv)",
    "language": "python",
    "name": "python3"
   },
@@ -378,11 +353,11 @@
    "name": "python",
    "nbconvert_exporter": "python",
    "pygments_lexer": "ipython3",
-   "version": "3.10.6"
+   "version": "3.8.15 (default, Oct 12 2022, 19:15:16) \n[GCC 11.2.0]"
   },
   "vscode": {
    "interpreter": {
-    "hash": "2bb62b3221a3e8eb4d56ef3abf33b59c1166039d4d316f70559cb86bddd6b450"
+    "hash": "8fcee05e428355ba9289075b405db6efadfd667fc0be61680b205b2b303cc40a"
    }
   }
  },