gpucom.ipynb

{
  "nbformat": 4,
  "nbformat_minor": 0,
  "metadata": {
    "colab": {
      "name": "gpucom.ipynb",
      "provenance": [],
      "collapsed_sections": []
    },
    "kernelspec": {
      "name": "python3",
      "display_name": "Python 3"
    }
  },
  "cells": [
    {
      "cell_type": "code",
      "metadata": {
        "id": "hVvicP6qaVIh",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 98
        },
        "outputId": "85f73ca6-6e5f-4977-fdd3-fd91490353f0"
      },
      "source": [
        "from keras.layers import Input, Dense, Conv2D, Reshape, Lambda\n",
        "from keras.models import Model\n",
        "import numpy as np\n",
        "import tensorflow as tf\n",
        "import cv2, sys\n",
        "import imageio as io\n",
        "from matplotlib import pyplot as plt\n"
      ],
      "execution_count": 1,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Using TensorFlow backend.\n"
          ],
          "name": "stderr"
        },
        {
          "output_type": "display_data",
          "data": {
            "text/html": [
              "<p style=\"color: red;\">\n",
              "The default version of TensorFlow in Colab will soon switch to TensorFlow 2.x.<br>\n",
              "We recommend you <a href=\"https://www.tensorflow.org/guide/migrate\" target=\"_blank\">upgrade</a> now \n",
              "or ensure your notebook will continue to use TensorFlow 1.x via the <code>%tensorflow_version 1.x</code> magic:\n",
              "<a href=\"https://colab.research.google.com/notebooks/tensorflow_version.ipynb\" target=\"_blank\">more info</a>.</p>\n"
            ],
            "text/plain": [
              "<IPython.core.display.HTML object>"
            ]
          },
          "metadata": {
            "tags": []
          }
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "fJMWuk5B4anL",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "def decode_float(fnum):\n",
        "      \n",
        "      # split int and dec parts\n",
        "      fract,integ=np.modf(fnum)\n",
        "      \n",
        "      intstr=bin(np.int(integ))[2:]\n",
        "      floatstr=''\n",
        "\n",
        "      while fnum!=0.0:\n",
        "          fnum=fract*2\n",
        "          fract,integ=np.modf(fnum)\n",
        "          floatstr=floatstr+str(np.int(integ))\n",
        "          fnum=fract\n",
        "\n",
        "      # pad zeros to adjust length\n",
        "      intstr = (8-len(intstr))*'0' + intstr\n",
        "      floatstr += (8-len(floatstr))*'0'\n",
        "      \n",
        "      return intstr+floatstr\n",
        "\n",
        "def float2bin(im,stride=16):\n",
        "    \n",
        "     # Compute size of extracted output image\n",
        "     h,w=im.shape[0], im.shape[1]\n",
        "     result=np.zeros((h,w*stride), dtype=np.uint8)\n",
        "     \n",
        "     # 2D convolution with stride of 16 [kernel shape: (1,16)]\n",
        "     for i in range(0,h):\n",
        "         for j in range(0,w):\n",
        "             ridx=j*stride\n",
        "             bval=decode_float(im[i,j])\n",
        "             result[i,ridx:ridx+stride]=np.uint8(list(bval))*np.uint8([255])\n",
        "             #print(\"BBVAL: \",bval,im[i,j], (i,ridx) )\n",
        "     return result"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "T1vIJ0IYfOk2",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "# Identity kernel (testing)\n",
        "def kernel_init1(shape, dtype=None):\n",
        "    \n",
        "    kernel = np.zeros(shape)\n",
        "    kernel[:,:,0,0] = np.array([[0,0,0],[0,1,0],[0,0,0]])\n",
        "    return kernel\n",
        "\n",
        "# Fixed point fp16 kernel (compression)\n",
        "def kernel_init2(shape, dtype=None):\n",
        "  \n",
        "    base=np.array([7,6,5,4,3,2,1,0,-1,-2,-3,-4,-5,-6,-7,-8],dtype=np.float32)\n",
        "    kernel=np.power(2,base).reshape(shape)\n",
        "    return kernel"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "wP1sdKTZddZ1",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "# Test model 1 (Identity)\n",
        "inputs = Input(shape=(256,256,1))\n",
        "outputs = Conv2D(1, (3, 3), kernel_initializer=kernel_init1, padding=\"same\" )(inputs)\n",
        "model1 = Model(inputs=inputs, outputs=outputs)\n",
        "\n",
        "model1.summary()\n",
        "model1.save('model1.h5')"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "JVGmju5pihPk",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "# Test model 2 (Compression)\n",
        "inputs = Input(shape=(256,256,1))\n",
        "hidden = Conv2D(1, (3, 3), kernel_initializer=kernel_init1, padding=\"same\" )(inputs)\n",
        "outputs = Conv2D(1, (1, 16), kernel_initializer=kernel_init2,strides=(1,16), padding=\"same\" )(hidden)\n",
        "model2 = Model(inputs=inputs, outputs=outputs)\n",
        "\n",
        "model2.summary()\n",
        "model2.save('model2.h5')"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "0CWLDgD0aNU4",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "# Test model 3 (Channel_4X)\n",
        "inputs = Input(shape=(256,256,4))\n",
        "outputs = Conv2D(4, (3, 3), kernel_initializer=kernel_init1, padding=\"same\" )(inputs)\n",
        "model1 = Model(inputs=inputs, outputs=outputs)\n",
        "\n",
        "model1.summary()\n",
        "model1.save('model3.h5')"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "VmnBZ7L-a-aZ",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "# Test model 4 (Comp_Ch4X)\n",
        "inputs = Input(shape=(256,256,3))\n",
        "hidden1 = Conv2D(1, (3, 3), kernel_initializer=kernel_init1, padding=\"same\" )(inputs)\n",
        "hidden2 = Conv2D(1, (1, 16), kernel_initializer=kernel_init2,strides=(1,16), padding=\"same\" )(hidden1)\n",
        "outputs= Reshape((256,16))(hidden2)\n",
        "model2 = Model(inputs=inputs, outputs=outputs)\n",
        "\n",
        "model2.summary()\n",
        "model2.save('model4.h5')"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "MCV3GuZri2kC",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "# Test model 5 (Comp_ResCh4X)\n",
        "inputs=Input(shape=(196608,))\n",
        "reshape = Reshape((256,256,3))(inputs)\n",
        "hidden1 = Conv2D(1, (3, 3), kernel_initializer=kernel_init1, padding=\"same\" )(reshape)\n",
        "hidden2 = Conv2D(1, (1, 16), kernel_initializer=kernel_init2,strides=(1,16), padding=\"same\" )(hidden1)\n",
        "outputs= Reshape((256,16))(hidden2)\n",
        "model2 = Model(inputs=inputs, outputs=outputs)\n",
        "\n",
        "model2.summary()\n",
        "model2.save('model5.h5')"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "UOFflxTsG0I2",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "# Test model 6 (Comp_SliCh4X)\n",
        "inputs = Input(shape=(256,256,4))\n",
        "str_slice=Lambda(lambda inputs: tf.strided_slice(inputs, [0,0, 0, 0], [1,256, 256, 3], [1, 1, 1, 1]))(inputs)\n",
        "hidden1 = Conv2D(1, (3, 3), kernel_initializer=kernel_init1, padding=\"same\" )(str_slice)\n",
        "hidden2 = Conv2D(1, (1, 16), kernel_initializer=kernel_init2,strides=(1,16), padding=\"same\" )(hidden1)\n",
        "outputs= Reshape((256,16))(hidden2)\n",
        "model2 = Model(inputs=inputs, outputs=outputs)\n",
        "\n",
        "model2.summary()\n",
        "model2.save('model6.h5')"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "cufdbvqLB7_B",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "# Test model 7 (Comp_PadCh4X)\n",
        "inputs = Input(shape=(256,256,4))\n",
        "str_slice=Lambda(lambda inputs: tf.strided_slice(inputs, [0,0, 0, 0], [1,256, 256, 3], [1, 1, 1, 1]))(inputs)\n",
        "hidden = Conv2D(1, (3, 3), kernel_initializer=kernel_init1, padding=\"same\" )(str_slice)\n",
        "outputs=Lambda(lambda inputs: tf.pad(inputs, tf.constant([[0,0], [0,0], [0,0], [0,3]])))(hidden)\n",
        "model2 = Model(inputs=inputs, outputs=outputs)\n",
        "\n",
        "model2.summary()\n",
        "model2.save('model7.h5')"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "sN51vycRklji",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "# Read input and preprocess\n",
        "input_img = io.imread('https://drive.google.com/uc?id=1LAfmxWqXVUki71Ha_mWbQ4yrGIOTUGHV')\n",
        "input_img=input_img[:,:,0]\n",
        "img=np.float32(input_img/255.0)\n",
        "img=img.reshape((1,256,256,1))"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "qqHOTN_xlHdA",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "# Perform prediction\n",
        "out1=model1.predict(img)\n",
        "out2=model2.predict(img)"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "urikIKo-lhe9",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "# Plot output of model1\n",
        "plt.imshow(out1.squeeze())"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "xnJiSuOUnvm4",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "output_img=float2bin(out2.squeeze()) # Extracted image [UINT8]\n",
        "\n",
        "print(\"\\nConverting the extracted image back to UINT8...\")\n",
        "\n",
        "print(\"\\nExtracted image:- \")\n",
        "print(\"Shape: \",output_img.shape)\n",
        "print(\"Type: \", output_img.dtype)\n",
        "print(\"Size: \",sys.getsizeof(output_img))\n",
        "print(\"Unique values: \", np.unique(output_img).shape[0])\n",
        "\n",
        "# Check if image was successfully extracted\n",
        "if np.array_equal(input_img,output_img):\n",
        "   print(\"\\nExtraction successful !!!\")\n",
        "else:\n",
        "   print(\"\\nExtraction failed !!!\")\n",
        "\n",
        "# Save the output image\n",
        "cv2.imwrite(\"decoded2.png\",output_img)\n",
        "plt.imshow(output_img)"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "D_NStg14-X_Z",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "# Export mode1-1 as tflite\n",
        "converter = tf.lite.TFLiteConverter.from_keras_model_file('/content/model1.h5',custom_objects={'kernel_init1':kernel_init1})\n",
        "tflite_model1 = converter.convert()\n",
        "open(\"model1.tflite\", \"wb\").write(tflite_model1)\n"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "jGlJRypD-dZO",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "# Export model-2 as tflite\n",
        "converter = tf.lite.TFLiteConverter.from_keras_model_file('/content/model2.h5', custom_objects={'kernel_init1':kernel_init1, 'kernel_init2':kernel_init2})\n",
        "tflite_model2 = converter.convert()\n",
        "open(\"model2.tflite\", \"wb\").write(tflite_model2)"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "gINXeVuQafgh",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "# Export model-3 as tflite\n",
        "converter = tf.lite.TFLiteConverter.from_keras_model_file('/content/model3.h5', custom_objects={'kernel_init1':kernel_init1, 'kernel_init2':kernel_init2})\n",
        "tflite_model2 = converter.convert()\n",
        "open(\"model3.tflite\", \"wb\").write(tflite_model2)"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "v8JNzsUrf75y",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "# Export model-4 as tflite\n",
        "converter = tf.lite.TFLiteConverter.from_keras_model_file('/content/model4.h5', custom_objects={'kernel_init1':kernel_init1, 'kernel_init2':kernel_init2})\n",
        "tflite_model2 = converter.convert()\n",
        "open(\"model4.tflite\", \"wb\").write(tflite_model2)"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "QLGFwjjijvbR",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "# Export model-5 as tflite\n",
        "converter = tf.lite.TFLiteConverter.from_keras_model_file('/content/model5.h5', custom_objects={'kernel_init1':kernel_init1, 'kernel_init2':kernel_init2})\n",
        "tflite_model2 = converter.convert()\n",
        "open(\"model5.tflite\", \"wb\").write(tflite_model2)"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "iebC5kqF4nx5",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "# Export model-6 as tflite\n",
        "converter = tf.lite.TFLiteConverter.from_keras_model_file('/content/model6.h5', custom_objects={'tf':tf,'kernel_init1':kernel_init1, 'kernel_init2':kernel_init2})\n",
        "tflite_model2 = converter.convert()\n",
        "open(\"model6.tflite\", \"wb\").write(tflite_model2)"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "-9nO3XGpDjjx",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "# Export model-7 as tflite\n",
        "converter = tf.lite.TFLiteConverter.from_keras_model_file('/content/model7.h5', custom_objects={'tf':tf,'kernel_init1':kernel_init1, 'kernel_init2':kernel_init2})\n",
        "tflite_model2 = converter.convert()\n",
        "open(\"model7.tflite\", \"wb\").write(tflite_model2)"
      ],
      "execution_count": 0,
      "outputs": []
    }
  ]
}