cleaned up

Author: msokoloff1

examples/integrations/tlt/detectnet_v2_bounding_box.ipynb

@@ -6,13 +6,12 @@
    "source": [
     "# Create training data using Labelbox\n",
     "* Download images and animal annotations\n",
-    "* Upload them to labelbox using MAL\n",
-    "* Label to add any missing classes"
+    "* Upload them to labelbox using MAL"
    ]
   },
   {
    "cell_type": "code",
-   "execution_count": 1,
+   "execution_count": null,
    "metadata": {},
    "outputs": [],
    "source": [
@@ -40,7 +39,7 @@
   },
   {
    "cell_type": "code",
-   "execution_count": 2,
+   "execution_count": null,
    "metadata": {},
    "outputs": [],
    "source": [
@@ -53,45 +52,44 @@
   },
   {
    "cell_type": "code",
-   "execution_count": 3,
+   "execution_count": null,
    "metadata": {},
    "outputs": [],
    "source": [
     "# Download the annotations\n",
-    "if not os.path.exists('eccv_18_annotation_files'):\n",
-    "    !wget http://www.vision.caltech.edu/~sbeery/datasets/caltechcameratraps18/eccv_18_annotations.tar.gz\n",
-    "    !tar -zxf eccv_18_annotations.tar.gz"
+    "if not os.path.exists('eccv_18_all_annotations'):\n",
+    "    !wget http://www.vision.caltech.edu/~sbeery/datasets/caltechcameratraps18/eccv_18_all_annotations.tar.gz\n",
+    "    !tar -zxf eccv_18_all_annotations.tar.gz"
    ]
   },
   {
    "cell_type": "markdown",
    "metadata": {},
    "source": [
     "## Preprocess Data\n",
+    "* Sample images from video sequences\n",
     "* Select only day time images and a subset of possible animals\n",
-    "* Since the images are coming from video frames we split into train and eval datasets to account for this.\n",
     "* Convert the data into a format that is compatible with labelbox"
    ]
   },
   {
    "cell_type": "code",
-   "execution_count": 4,
+   "execution_count": null,
    "metadata": {},
    "outputs": [],
    "source": [
-    "data = json.load(open('eccv_18_annotation_files/train_annotations.json'))\n",
+    "data = json.load(open('CaltechCameraTrapsECCV18.json'))\n",
     "data['categories'] = {d['id'] : d for d in data['categories']}\n",
     "annotations = defaultdict(lambda: [])"
    ]
   },
   {
    "cell_type": "code",
-   "execution_count": 5,
+   "execution_count": null,
    "metadata": {},
    "outputs": [],
    "source": [
-    "#Unique sequence of frames so the that the same animal isn't in the train and eval set by chance\n",
-    "#We also want different seq_ids so that they are all from different sequences (not same animal)\n",
+    "# One image per video sequence to reduce correlation between training/testing images.\n",
     "\n",
     "images = {}\n",
     "ids = set()\n",
@@ -113,19 +111,25 @@
   },
   {
    "cell_type": "code",
-   "execution_count": 6,
+   "execution_count": null,
    "metadata": {},
    "outputs": [],
    "source": [
+    "# These ids correspond to locations with a lot of people in the images that we can label\n",
+    "target_locations = {0,125,120}\n",
     "target_classes = {'dog', 'cat', 'deer', 'bobcat', 'fox'}\n",
+    "min_border_distance = 50\n",
     "\n",
     "\n",
-    "def process_image(image, min_bbox_height_px = 50 , min_bbox_width_px = 50):\n",
+    "def process_image(image):\n",
     "    date_time_obj = datetime.datetime.strptime(image['date_captured'], '%Y-%m-%d %H:%M:%S')\n",
     "    if (not ((18 > date_time_obj.hour > 7)) or (date_time_obj.hour == 12)):\n",
     "        #Only train on day time images\n",
     "        return\n",
     "    \n",
+    "    if image['location'] not in target_locations:\n",
+    "        return\n",
+    "    \n",
     "    annots = annotations[image['id']]\n",
     "    im = None \n",
     "    box_coords = []\n",
@@ -138,11 +142,11 @@
     "        bbox = annot.get('bbox')\n",
     "        assert bbox is not None\n",
     "        \n",
-    "        if bbox[0] < min_bbox_width_px or bbox[1] < min_bbox_height_px:\n",
-    "            #Ignore tiny bboxes\n",
+    "        # Don't train on images where the animal is on the edge of the image\n",
+    "        if bbox[0] < min_border_distance or bbox[1] < min_border_distance:\n",
     "            return\n",
     "        \n",
-    "        if (w - (bbox[0] + bbox[2])) < min_bbox_width_px or (h - (bbox[1] + bbox[3])) < min_bbox_height_px:\n",
+    "        if (w - (bbox[0] + bbox[2])) < min_border_distance or (h - (bbox[1] + bbox[3])) < min_border_distance:\n",
     "            return \n",
     "        \n",
     "        if im is None:\n",
@@ -158,17 +162,9 @@
   },
   {
    "cell_type": "code",
-   "execution_count": 7,
+   "execution_count": null,
    "metadata": {},
-   "outputs": [
-    {
-     "name": "stdout",
-     "output_type": "stream",
-     "text": [
-      "191\n"
-     ]
-    }
-   ],
+   "outputs": [],
    "source": [
     "examples = [process_image(ex) for ex in data['images'].values()]\n",
     "examples = [ex for ex in examples if ex is not None]\n",
@@ -184,7 +180,7 @@
   },
   {
    "cell_type": "code",
-   "execution_count": 8,
+   "execution_count": null,
    "metadata": {},
    "outputs": [],
    "source": [
@@ -195,6 +191,7 @@
     "    os.mkdir(\"labels\")\n",
     "    \n",
     "image_paths = []\n",
+    "\n",
     "for idx, example in enumerate(examples):\n",
     "    imm, coords, location = example\n",
     "    image_path = os.path.join(\"uploaded_images\", f\"{idx}.jpg\")\n",
@@ -219,29 +216,18 @@
   },
   {
    "cell_type": "code",
-   "execution_count": 9,
+   "execution_count": null,
    "metadata": {},
    "outputs": [],
    "source": [
-    "client = Client(api_key = \"eyJhbGciOiJIUzI1NiIsInR5cCI6IkpXVCJ9.eyJ1c2VySWQiOiJja2s0cTF2Z3djMHZwMDcwNHhoeDdtNHZrIiwib3JnYW5pemF0aW9uSWQiOiJja2s0cTF2Z2Fwc2F1MDczMjRhd25zanEyIiwiYXBpS2V5SWQiOiJja2t6bjd5dG5pZHNjMDcwNjczazIyamF1IiwiaWF0IjoxNjEyOTc0MjQ3LCJleHAiOjIyNDQxMjYyNDd9.GrGjHbN1w1X5-qLzlzp9UKCnkSffKqTQWEWIRyegHGg\")"
+    "client = Client()"
    ]
   },
   {
    "cell_type": "code",
-   "execution_count": 10,
+   "execution_count": null,
    "metadata": {},
-   "outputs": [
-    {
-     "data": {
-      "text/plain": [
-       "True"
-      ]
-     },
-     "execution_count": 10,
-     "metadata": {},
-     "output_type": "execute_result"
-    }
-   ],
+   "outputs": [],
    "source": [
     "project = client.create_project(name = \"animal_demo_proj\")\n",
     "dataset = client.create_dataset(name = \"animal_demo_ds\")\n",
@@ -252,7 +238,7 @@
   },
   {
    "cell_type": "code",
-   "execution_count": 11,
+   "execution_count": null,
    "metadata": {},
    "outputs": [],
    "source": [
@@ -273,7 +259,7 @@
   },
   {
    "cell_type": "code",
-   "execution_count": 12,
+   "execution_count": null,
    "metadata": {},
    "outputs": [],
    "source": [
@@ -286,7 +272,7 @@
   },
   {
    "cell_type": "code",
-   "execution_count": 13,
+   "execution_count": null,
    "metadata": {},
    "outputs": [],
    "source": [
@@ -295,7 +281,7 @@
   },
   {
    "cell_type": "code",
-   "execution_count": 14,
+   "execution_count": null,
    "metadata": {},
    "outputs": [],
    "source": [
@@ -319,7 +305,7 @@
   },
   {
    "cell_type": "code",
-   "execution_count": 16,
+   "execution_count": null,
    "metadata": {},
    "outputs": [],
    "source": [
@@ -331,24 +317,21 @@
    "cell_type": "markdown",
    "metadata": {},
    "source": [
-    "## Go to labelbox and label\n",
-    "* Most of the data is prelabeled so we just need to go through and make sure everything is correct\n",
+    "## Go to Labelbox and Label\n",
+    "* Most of the animal data is prelabeled we want to go through and make sure everything is correct\n",
+    "    * Make sure to use the hot keys to label quickly!\n",
+    "    * 'e' submits the image\n",
+    "    * '1' selects the person bounding box\n",
+    "    * '2' selects the animal bounding box\n",
+    "    * There are other helpful ones too! Check out the keyboard shortcuts panel in the top right of the editor.\n",
     "* None of the people in the images have been labeled so we are also going to add those annotations"
    ]
   },
   {
    "cell_type": "code",
-   "execution_count": 17,
+   "execution_count": null,
    "metadata": {},
-   "outputs": [
-    {
-     "name": "stdout",
-     "output_type": "stream",
-     "text": [
-      "https://app.labelbox.com/projects/ckq6zvmwm8sko0ya4fevdgsbf/overview\n"
-     ]
-    }
-   ],
+   "outputs": [],
    "source": [
     "print(f\"https://app.labelbox.com/projects/{project.uid}/overview\")"
    ]