Adding CTC compatiable models

colabs/KERAS_OCR_TFLITE.ipynb

@@ -29,7 +29,7 @@
   },
   {
    "cell_type": "code",
-   "execution_count": 2,
+   "execution_count": null,
    "metadata": {
     "colab": {
      "base_uri": "https://localhost:8080/",
@@ -38,23 +38,7 @@
     "id": "5Mme4MUCxVM8",
     "outputId": "df988251-c816-489b-fa43-58a1096500a3"
    },
-   "outputs": [
-    {
-     "data": {
-      "application/vnd.google.colaboratory.intrinsic+json": {
-       "type": "string"
-      },
-      "text/plain": [
-       "'2.5.0-dev20201128'"
-      ]
-     },
-     "execution_count": 2,
-     "metadata": {
-      "tags": []
-     },
-     "output_type": "execute_result"
-    }
-   ],
+   "outputs": [],
    "source": [
     "import typing\n",
     "import string\n",
@@ -404,8 +388,8 @@
     "                           activation='softmax',\n",
     "                           name='fc_12')(x)\n",
     "    x = keras.layers.Lambda(lambda x: x[:, rnn_steps_to_discard:])(x)\n",
-    "    prediction_model = keras.models.Model(inputs=inputs, outputs=CTCDecoder()(model.output))\n",
     "    model = keras.models.Model(inputs=inputs, outputs=x)\n",
+    "    prediction_model = keras.models.Model(inputs=inputs, outputs=CTCDecoder()(model.output))\n",
     "    return model, prediction_model"
    ]
   },
@@ -611,9 +595,7 @@
     "\n",
     "Refer to this [issue](https://github.com/tensorflow/tensorflow/issues/33494) regarding CTC decoder support in TFLite. \n",
     "\n",
-    "This is the code for [CTC DECODER](https://colab.research.google.com/github/tulasiram58827/ocr_tflite/blob/main/colabs/KERAS_OCR_TFLITE.ipynb#scrollTo=_rdJyCXo2Xzs). By default it is greedy Decoder we can also use Beam Search Decoder by specifying the parameter in the `ctc_decode` function.\n",
-    "\n",
-    "FYI: I am also working on converting Beam Search CTC Decoder to low level language so that we can port entire OCR and use it as offline application combinedly with EAST/CRAFT."
+    "**Update** : CTC Decoder is supported in TFLite now by enabling Built-in-Ops in Tensorflow 2.4. Thanks to TensorFlow team for the support."
    ]
   },
   {
@@ -794,29 +776,6 @@
     "## TFLite Inference"
    ]
   },
-  {
-   "cell_type": "code",
-   "execution_count": 24,
-   "metadata": {
-    "id": "5bxtGxQEf0LJ"
-   },
-   "outputs": [],
-   "source": [
-    "# Code for CTC Decoder \n",
-    "\n",
-    "def decoder(y_pred):\n",
-    "    input_shape = tf.keras.backend.shape(y_pred)\n",
-    "    input_length = tf.ones(shape=input_shape[0]) * tf.keras.backend.cast(\n",
-    "        input_shape[1], 'float32')\n",
-    "    # You can turn on beam search decoding using greedy=False and also play with beam_width parameter.\n",
-    "    unpadded = tf.keras.backend.ctc_decode(y_pred, input_length)[0][0]\n",
-    "    unpadded_shape = tf.keras.backend.shape(unpadded)\n",
-    "    padded = tf.pad(unpadded,\n",
-    "                    paddings=[[0, 0], [0, input_shape[1] - unpadded_shape[1]]],\n",
-    "                    constant_values=-1)\n",
-    "    return padded"
-   ]
-  },
   {
    "cell_type": "code",
    "execution_count": 25,
@@ -895,8 +854,6 @@
    "source": [
     "# Running Dynamic Range Quantization\n",
     "tflite_output = run_tflite_model(image_path, 'dr')\n",
-    "# Running decoder on TFLite Output\n",
-    "# decoded = decoder(tflite_output)\n",
     "final_output = \"\".join(alphabets[index] for index in tflite_output[0] if index not in [blank_index, -1])\n",
     "print(final_output)\n",
     "cv2_imshow(cv2.imread(image_path))"
@@ -937,8 +894,6 @@
    "source": [
     "# Running Float16 Quantization\n",
     "tflite_output = run_tflite_model(image_path, 'float16')\n",
-    "# Running decoder on TFLite Output\n",
-    "# decoded = decoder(tflite_output)\n",
     "final_output = \"\".join(alphabets[index] for index in tflite_output[0] if index not in [blank_index, -1])\n",
     "print(final_output)\n",
     "cv2_imshow(cv2.imread(image_path))"
@@ -954,8 +909,6 @@
    "source": [
     "# Running Integer Quantization\n",
     "tflite_output = run_tflite_model(image_path, 'int8')\n",
-    "# Running decoder on TFLite Output\n",
-    "decoded = decoder(tflite_output)\n",
     "final_output = \"\".join(alphabets[index] for index in decoded[0] if index not in [blank_index, -1])\n",
     "print(final_output)\n",
     "cv2_imshow(cv2.imread(image_path))"
@@ -1020,7 +973,7 @@
     "id": "1g_bWIW5gvni"
    },
    "source": [
-    "**The above benchmarks with respect to Redmi K20 Pro**"
+    "**The above benchmarks with respect to Redmi K20 Pro with 4 threads. **"
    ]
   }
  ],