trying my_ssd_mobnet_640

VinceJnz · Feb 8, 2023 · 13c7baf · 13c7baf
1 parent 30daf91
commit 13c7baf
Show file tree

Hide file tree

Showing 2 changed files with 49 additions and 25 deletions.
diff --git a/README.md b/README.md
@@ -12,4 +12,34 @@ This fork has the following changes
 
 ## label studio
 
-http://localhost:8080/projects/6/data/local-storage/6/9ee06b40-Day_1_Images_Flight_1_DSC09209_geotag_1.JPG
+http://localhost:8080/projects/6/data/local-storage/6/9ee06b40-Day_1_Images_Flight_1_DSC09209_geotag_1.JPG
+
+
+## Model training
+
+### Training details
+
+<https://www.tensorflow.org/guide/distributed_training>
+<https://www.tensorflow.org/guide/checkpoint>
+
+
+### Execution flow for training
+[model_main.py]
+    main
+        [model_lib_v2.py]
+            train_loop
+                [inputs.py]
+                    train_input
+                        [image_resizer_builder.py]
+                            build
+
+?????????
+                train_step_fn()
+                    eager_train_step
+                        _compute_losses_and_predictions_dicts(...) #model prediction is done here
+
+
+??????????
+
+
+
diff --git a/Training_Customised_Image_Detection_Model-ML4DR_v1.ipynb b/Training_Customised_Image_Detection_Model-ML4DR_v1.ipynb
@@ -43,7 +43,10 @@
     "SEED = 123\n",
     "random.seed(SEED)\n",
     "np.random.seed(SEED)\n",
-    "tf.random.set_seed(SEED)"
+    "tf.random.set_seed(SEED)\n",
+    "\n",
+    "import os\n",
+    "print(f\"Current OS: {os.name}\") "
    ]
   },
   {
@@ -57,22 +60,6 @@
     "Including paths and training hyper-parameters"
    ]
   },
-  {
-   "cell_type": "code",
-   "execution_count": null,
-   "metadata": {
-    "colab": {
-     "base_uri": "https://localhost:8080/"
-    },
-    "id": "vCm7RtDn0za5",
-    "outputId": "9a10857b-fff7-4cfd-c568-318c1658d087"
-   },
-   "outputs": [],
-   "source": [
-    "import os\n",
-    "print(f\"Current OS: {os.name}\") "
-   ]
-  },
   {
    "cell_type": "markdown",
    "metadata": {},
@@ -541,7 +528,7 @@
    "cell_type": "markdown",
    "metadata": {},
    "source": [
-    "### Create TF Record from COCO files"
+    "### Prep COCO file data"
    ]
   },
   {
@@ -626,20 +613,27 @@
     "create_label_map(labels, files['LABELMAP'])"
    ]
   },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "### Create TF Record from COCO files"
+   ]
+  },
   {
    "cell_type": "code",
    "execution_count": null,
    "metadata": {},
    "outputs": [],
    "source": [
-    "#This is used when creating the TF Record files\n",
-    "# in the TF Record file\n",
-    "# Store each image with its annotations\n",
+    "#This is used when creating and reading the TF Record files\n",
     "\n",
     "from object_detection.utils import dataset_util #, label_map_util\n",
     "import io\n",
     "from PIL import Image\n",
     "\n",
+    "#This is used write the TF Records\n",
+    "#Store images with its associated annotations in a TF Record file\n",
     "def create_tf_example(imageInfo, image_path):\n",
     "    with tf.io.gfile.GFile(os.path.join(image_path, '{}'.format(imageInfo['file_name'])), 'rb') as fid:\n",
     "        encoded_jpg = fid.read()\n",
@@ -903,13 +897,13 @@
     "pipeline_config.model.ssd.num_classes = len(labels)\n",
     "#pipeline_config.train_config.batch_size = 4\n",
     "#pipeline_config.train_config.batch_size = 3\n",
-    "#pipeline_config.train_config.batch_size = 2\n",
+    "pipeline_config.train_config.batch_size = 2\n",
     "#pipeline_config.train_config.batch_size = 1\n",
     "\n",
     "#if the batch size is small the learning rate needs to be reduced for the model to converge\n",
     "#https://github.com/tensorflow/models/issues/9151\n",
     "#https://medium.com/mini-distill/effect-of-batch-size-on-training-dynamics-21c14f7a716e\n",
-    "#pipeline_config.train_config.optimizer.momentium_optimizer.learning_rate.cosine_decay_learning_rate.learning_rate_base = 0.05 #0.08\n",
+    "pipeline_config.train_config.optimizer.momentum_optimizer.learning_rate.cosine_decay_learning_rate.learning_rate_base = 0.03 #0.08\n",
     "pipeline_config.train_config.fine_tune_checkpoint = os.path.join(paths['PRETRAINED_MODEL_PATH'], PRETRAINED_MODEL_NAME, 'checkpoint', 'ckpt-0')\n",
     "pipeline_config.train_config.fine_tune_checkpoint_type = \"detection\"\n",
     "pipeline_config.train_input_reader.label_map_path= files['LABELMAP']\n",
@@ -1119,7 +1113,7 @@
     "            category_index,\n",
     "            use_normalized_coordinates=True,\n",
     "            max_boxes_to_draw=5,\n",
-    "            min_score_thresh=.05, #(0.5)\n",
+    "            min_score_thresh=.5, #(0.5)\n",
     "            agnostic_mode=False)\n",
     "\n",
     "plt.imshow(cv2.cvtColor(image_np_with_detections, cv2.COLOR_BGR2RGB))\n",