Also download the utilities code in python before running the code in this notebook.
" ] }, { "cell_type": "code", "execution_count": 1, "metadata": {}, "outputs": [], "source": [ "import warnings\n", "warnings.filterwarnings('ignore')\n", "warnings.simplefilter('ignore')" ] }, { "cell_type": "code", "execution_count": 2, "metadata": {}, "outputs": [ { "name": "stderr", "output_type": "stream", "text": [ "2025-08-16 13:28:46.739489: E external/local_xla/xla/stream_executor/cuda/cuda_fft.cc:477] Unable to register cuFFT factory: Attempting to register factory for plugin cuFFT when one has already been registered\n", "WARNING: All log messages before absl::InitializeLog() is called are written to STDERR\n", "E0000 00:00:1755331126.821084 72924 cuda_dnn.cc:8310] Unable to register cuDNN factory: Attempting to register factory for plugin cuDNN when one has already been registered\n", "E0000 00:00:1755331126.845849 72924 cuda_blas.cc:1418] Unable to register cuBLAS factory: Attempting to register factory for plugin cuBLAS when one has already been registered\n", "2025-08-16 13:28:47.047048: I tensorflow/core/platform/cpu_feature_guard.cc:210] This TensorFlow binary is optimized to use available CPU instructions in performance-critical operations.\n", "To enable the following instructions: AVX2 FMA, in other operations, rebuild TensorFlow with the appropriate compiler flags.\n" ] } ], "source": [ "# Import all the necessary modules for Deep Learning - uses Theano backend\n", "##########################################################################\n", "import numpy as np\n", "import pickle, gzip, os\n", "import cv2\n", "import tensorflow as tf\n", "from tensorflow import keras\n", "from tensorflow.keras import layers\n", "from matplotlib.pyplot import imshow\n", "import cb_dataset_utils as dsu\n", "\n", "MNIST70K = '/usr/local/Datasets/mnist.pkl.gz'\n", "TELOCR73K = 'amit_73k.pkl'\n", "TELMAPFILE = 'HL-UNICODE.TXT'\n", "TELUGU_SCRIPT = True\n", "np.set_printoptions(suppress=True, precision=3)" ] }, { "cell_type": "code", "execution_count": 3, "metadata": {}, "outputs": [ { "name": "stderr", "output_type": "stream", "text": [ "I0000 00:00:1755331143.484946 72924 gpu_device.cc:2022] Created device /job:localhost/replica:0/task:0/device:GPU:0 with 1730 MB memory: -> device: 0, name: NVIDIA GeForce MX250, pci bus id: 0000:02:00.0, compute capability: 6.1\n" ] } ], "source": [ "from tensorflow.compat.v1 import ConfigProto\n", "from tensorflow.compat.v1 import InteractiveSession\n", "\n", "config = ConfigProto()\n", "config.gpu_options.allow_growth = True\n", "session = InteractiveSession(config=config)" ] }, { "cell_type": "code", "execution_count": 4, "metadata": {}, "outputs": [], "source": [ "def TF_preprocess_data(X, Y, r, c, ch, cat=False, nc=0) :\n", " newX = X.reshape((X.shape[0], r, c, ch))\n", " if newX.max() > 1.0 :\n", " newX /= 255.\n", " newX = newX.astype('float32')\n", " newY = Y.astype('int32')\n", " if cat :\n", " if nc == 0 :\n", " print ('Warning: Number of classes = 0')\n", " else :\n", " newY = tf.keras.utils.to_categorical(Y, nc)\n", "\n", " return newX, newY\n", "\n", "def make_telugu_dictionary(fname) :\n", " fp = open(TELMAPFILE, 'r', encoding='utf-8', errors='ignore')\n", " ttext = fp.read()\n", " fp.close()\n", " codelines = ttext.splitlines()\n", " codes = codelines[0].split('\\t')\n", " teldict = {codes[0]:codes[2]}\n", " for i in range(1, len(codelines)) :\n", " codes = codelines[i].split('\\t')\n", " teldict[codes[0]] = codes[2]\n", " print('Dictionary Size: ', len(teldict))\n", " return teldict" ] }, { "cell_type": "code", "execution_count": 5, "metadata": {}, "outputs": [ { "name": "stdout", "output_type": "stream", "text": [ "Training Set Shape: (55000, 1024)\n", "Validation Set Shape: (8000, 1024)\n", "Testing Set Shape: (10000, 1024)\n", "Trg. Data Shape: (55000, 32, 32, 1)\n", "Val. Data Shape: (8000, 32, 32, 1)\n", "Test Data Shape: (10000, 32, 32, 1)\n" ] } ], "source": [ "# Read dataset into Training, Validation and Test sets.\n", "# Each set contains the input data and corresponding output class labels.\n", "##########################################################################\n", "ts, vs, xs = dsu.pickle_read_data(TELOCR73K, gz=False) # MNIST dataset\n", " # For Telugu OCR, use km.TELOCR73K\n", "nrows = 32 # Change to 32 for Telugu OCR\n", "ncols = 32 # Change to 32 for Telugu OCR\n", "numclasses = 508 # Num Classes; 508 for Telugu\n", "TELUGU_SCRIPT = True # Set this if using Telugu dataset\n", "# Preprocess data into required shapes for Keras layers\n", "td, tlab = TF_preprocess_data(ts[0], ts[1], \n", " nrows, nrows, 1, \n", " cat=True, nc=numclasses)\n", "vd, vlab = TF_preprocess_data(vs[0], vs[1], \n", " nrows, ncols, 1, \n", " cat=True, nc=numclasses)\n", "xd, xlab = TF_preprocess_data(xs[0], xs[1], \n", " nrows, ncols, 1, \n", " cat=True, nc=numclasses)\n", "print ('Trg. Data Shape: ', td.shape)\n", "print ('Val. Data Shape: ', vd.shape)\n", "print ('Test Data Shape: ', xd.shape)" ] }, { "cell_type": "code", "execution_count": 6, "metadata": {}, "outputs": [ { "name": "stderr", "output_type": "stream", "text": [ "I0000 00:00:1755331162.538913 72924 gpu_device.cc:2022] Created device /job:localhost/replica:0/task:0/device:GPU:0 with 1730 MB memory: -> device: 0, name: NVIDIA GeForce MX250, pci bus id: 0000:02:00.0, compute capability: 6.1\n" ] }, { "data": { "text/html": [ "Model: \"sequential\"\n",
"\n"
],
"text/plain": [
"\u001b[1mModel: \"sequential\"\u001b[0m\n"
]
},
"metadata": {},
"output_type": "display_data"
},
{
"data": {
"text/html": [
"┏━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━┓\n",
"┃ Layer (type) ┃ Output Shape ┃ Param # ┃\n",
"┡━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━┩\n",
"│ conv2d (Conv2D) │ (None, 28, 28, 36) │ 936 │\n",
"├─────────────────────────────────┼────────────────────────┼───────────────┤\n",
"│ conv2d_1 (Conv2D) │ (None, 26, 26, 12) │ 3,900 │\n",
"├─────────────────────────────────┼────────────────────────┼───────────────┤\n",
"│ conv2d_2 (Conv2D) │ (None, 24, 24, 24) │ 2,616 │\n",
"├─────────────────────────────────┼────────────────────────┼───────────────┤\n",
"│ max_pooling2d (MaxPooling2D) │ (None, 6, 6, 24) │ 0 │\n",
"├─────────────────────────────────┼────────────────────────┼───────────────┤\n",
"│ flatten (Flatten) │ (None, 864) │ 0 │\n",
"├─────────────────────────────────┼────────────────────────┼───────────────┤\n",
"│ dense (Dense) │ (None, 64) │ 55,360 │\n",
"├─────────────────────────────────┼────────────────────────┼───────────────┤\n",
"│ dense_1 (Dense) │ (None, 512) │ 33,280 │\n",
"├─────────────────────────────────┼────────────────────────┼───────────────┤\n",
"│ dense_2 (Dense) │ (None, 508) │ 260,604 │\n",
"└─────────────────────────────────┴────────────────────────┴───────────────┘\n",
"\n"
],
"text/plain": [
"┏━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━┓\n",
"┃\u001b[1m \u001b[0m\u001b[1mLayer (type) \u001b[0m\u001b[1m \u001b[0m┃\u001b[1m \u001b[0m\u001b[1mOutput Shape \u001b[0m\u001b[1m \u001b[0m┃\u001b[1m \u001b[0m\u001b[1m Param #\u001b[0m\u001b[1m \u001b[0m┃\n",
"┡━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━┩\n",
"│ conv2d (\u001b[38;5;33mConv2D\u001b[0m) │ (\u001b[38;5;45mNone\u001b[0m, \u001b[38;5;34m28\u001b[0m, \u001b[38;5;34m28\u001b[0m, \u001b[38;5;34m36\u001b[0m) │ \u001b[38;5;34m936\u001b[0m │\n",
"├─────────────────────────────────┼────────────────────────┼───────────────┤\n",
"│ conv2d_1 (\u001b[38;5;33mConv2D\u001b[0m) │ (\u001b[38;5;45mNone\u001b[0m, \u001b[38;5;34m26\u001b[0m, \u001b[38;5;34m26\u001b[0m, \u001b[38;5;34m12\u001b[0m) │ \u001b[38;5;34m3,900\u001b[0m │\n",
"├─────────────────────────────────┼────────────────────────┼───────────────┤\n",
"│ conv2d_2 (\u001b[38;5;33mConv2D\u001b[0m) │ (\u001b[38;5;45mNone\u001b[0m, \u001b[38;5;34m24\u001b[0m, \u001b[38;5;34m24\u001b[0m, \u001b[38;5;34m24\u001b[0m) │ \u001b[38;5;34m2,616\u001b[0m │\n",
"├─────────────────────────────────┼────────────────────────┼───────────────┤\n",
"│ max_pooling2d (\u001b[38;5;33mMaxPooling2D\u001b[0m) │ (\u001b[38;5;45mNone\u001b[0m, \u001b[38;5;34m6\u001b[0m, \u001b[38;5;34m6\u001b[0m, \u001b[38;5;34m24\u001b[0m) │ \u001b[38;5;34m0\u001b[0m │\n",
"├─────────────────────────────────┼────────────────────────┼───────────────┤\n",
"│ flatten (\u001b[38;5;33mFlatten\u001b[0m) │ (\u001b[38;5;45mNone\u001b[0m, \u001b[38;5;34m864\u001b[0m) │ \u001b[38;5;34m0\u001b[0m │\n",
"├─────────────────────────────────┼────────────────────────┼───────────────┤\n",
"│ dense (\u001b[38;5;33mDense\u001b[0m) │ (\u001b[38;5;45mNone\u001b[0m, \u001b[38;5;34m64\u001b[0m) │ \u001b[38;5;34m55,360\u001b[0m │\n",
"├─────────────────────────────────┼────────────────────────┼───────────────┤\n",
"│ dense_1 (\u001b[38;5;33mDense\u001b[0m) │ (\u001b[38;5;45mNone\u001b[0m, \u001b[38;5;34m512\u001b[0m) │ \u001b[38;5;34m33,280\u001b[0m │\n",
"├─────────────────────────────────┼────────────────────────┼───────────────┤\n",
"│ dense_2 (\u001b[38;5;33mDense\u001b[0m) │ (\u001b[38;5;45mNone\u001b[0m, \u001b[38;5;34m508\u001b[0m) │ \u001b[38;5;34m260,604\u001b[0m │\n",
"└─────────────────────────────────┴────────────────────────┴───────────────┘\n"
]
},
"metadata": {},
"output_type": "display_data"
},
{
"data": {
"text/html": [
"Total params: 356,696 (1.36 MB)\n", "\n" ], "text/plain": [ "\u001b[1m Total params: \u001b[0m\u001b[38;5;34m356,696\u001b[0m (1.36 MB)\n" ] }, "metadata": {}, "output_type": "display_data" }, { "data": { "text/html": [ "
Trainable params: 356,696 (1.36 MB)\n", "\n" ], "text/plain": [ "\u001b[1m Trainable params: \u001b[0m\u001b[38;5;34m356,696\u001b[0m (1.36 MB)\n" ] }, "metadata": {}, "output_type": "display_data" }, { "data": { "text/html": [ "
Non-trainable params: 0 (0.00 B)\n", "\n" ], "text/plain": [ "\u001b[1m Non-trainable params: \u001b[0m\u001b[38;5;34m0\u001b[0m (0.00 B)\n" ] }, "metadata": {}, "output_type": "display_data" }, { "name": "stdout", "output_type": "stream", "text": [ "None\n" ] } ], "source": [ "# We build the deep network model here. The default model has\n", "# one input layer (2D with input image size)\n", "# two alternating convolution (3x3 filters) and max pooling layers\n", "# one fully-connected layer\n", "# one output layer (as many neurons as number of classes)\n", "##########################################################################\n", "lng_model = keras.Sequential()\n", "# km.build_default_model(lng_model, nrows, ncols, 1, numclasses)\n", "lng_model.add(layers.Conv2D(36, (5, 5), activation='relu', \n", " input_shape=(nrows, ncols, 1)))\n", "# lng_model.add(layers.MaxPooling2D(pool_size=(2, 2)))\n", "lng_model.add(layers.Conv2D(12, (3, 3), activation='tanh'))\n", "# lng_model.add(layers.MaxPooling2D(pool_size=(1, 1)))\n", "lng_model.add(layers.Conv2D(24, (3, 3), activation='relu'))\n", "lng_model.add(layers.MaxPooling2D(pool_size=(4, 4)))\n", "lng_model.add(layers.Flatten())\n", "lng_model.add(layers.Dense(64, activation='tanh'))\n", "lng_model.add(layers.Dense(512, activation='tanh'))\n", "lng_model.add(layers.Dense(numclasses, activation='softmax'))\n", "# End of Model definition\n", "\n", "print (lng_model.summary())" ] }, { "cell_type": "code", "execution_count": 7, "metadata": {}, "outputs": [ { "name": "stderr", "output_type": "stream", "text": [ "2025-08-16 13:29:29.946595: W external/local_xla/xla/tsl/framework/cpu_allocator_impl.cc:83] Allocation of 225280000 exceeds 10% of free system memory.\n", "2025-08-16 13:29:30.226236: W external/local_xla/xla/tsl/framework/cpu_allocator_impl.cc:83] Allocation of 225280000 exceeds 10% of free system memory.\n" ] }, { "name": "stdout", "output_type": "stream", "text": [ "Epoch 1/3\n" ] }, { "name": "stderr", "output_type": "stream", "text": [ "WARNING: All log messages before absl::InitializeLog() is called are written to STDERR\n", "I0000 00:00:1755331171.677625 73038 service.cc:148] XLA service 0x7fe38800e7d0 initialized for platform CUDA (this does not guarantee that XLA will be used). Devices:\n", "I0000 00:00:1755331171.677759 73038 service.cc:156] StreamExecutor device (0): NVIDIA GeForce MX250, Compute Capability 6.1\n", "2025-08-16 13:29:31.721710: I tensorflow/compiler/mlir/tensorflow/utils/dump_mlir_util.cc:268] disabling MLIR crash reproducer, set env var `MLIR_CRASH_REPRODUCER_DIRECTORY` to enable.\n", "I0000 00:00:1755331171.947755 73038 cuda_dnn.cc:529] Loaded cuDNN version 90800\n", "2025-08-16 13:29:33.388388: I external/local_xla/xla/service/gpu/autotuning/conv_algorithm_picker.cc:557] Omitted potentially buggy algorithm eng14{} for conv (f32[500,36,28,28]{3,2,1,0}, u8[0]{0}) custom-call(f32[500,1,32,32]{3,2,1,0}, f32[36,1,5,5]{3,2,1,0}, f32[36]{0}), window={size=5x5}, dim_labels=bf01_oi01->bf01, custom_call_target=\"__cudnn$convBiasActivationForward\", backend_config={\"cudnn_conv_backend_config\":{\"activation_mode\":\"kNone\",\"conv_result_scale\":1,\"leakyrelu_alpha\":0,\"side_input_scale\":0},\"force_earliest_schedule\":false,\"operation_queue_id\":\"0\",\"wait_on_operation_queues\":[]}\n", "2025-08-16 13:29:33.522587: I external/local_xla/xla/service/gpu/autotuning/conv_algorithm_picker.cc:557] Omitted potentially buggy algorithm eng14{} for conv (f32[500,12,26,26]{3,2,1,0}, u8[0]{0}) custom-call(f32[500,36,28,28]{3,2,1,0}, f32[12,36,3,3]{3,2,1,0}, f32[12]{0}), window={size=3x3}, dim_labels=bf01_oi01->bf01, custom_call_target=\"__cudnn$convBiasActivationForward\", backend_config={\"cudnn_conv_backend_config\":{\"activation_mode\":\"kNone\",\"conv_result_scale\":1,\"leakyrelu_alpha\":0,\"side_input_scale\":0},\"force_earliest_schedule\":false,\"operation_queue_id\":\"0\",\"wait_on_operation_queues\":[]}\n", "2025-08-16 13:29:35.179601: I external/local_xla/xla/service/gpu/autotuning/conv_algorithm_picker.cc:557] Omitted potentially buggy algorithm eng14{} for conv (f32[500,24,24,24]{3,2,1,0}, u8[0]{0}) custom-call(f32[500,12,26,26]{3,2,1,0}, f32[24,12,3,3]{3,2,1,0}, f32[24]{0}), window={size=3x3}, dim_labels=bf01_oi01->bf01, custom_call_target=\"__cudnn$convBiasActivationForward\", backend_config={\"cudnn_conv_backend_config\":{\"activation_mode\":\"kNone\",\"conv_result_scale\":1,\"leakyrelu_alpha\":0,\"side_input_scale\":0},\"force_earliest_schedule\":false,\"operation_queue_id\":\"0\",\"wait_on_operation_queues\":[]}\n" ] }, { "name": "stdout", "output_type": "stream", "text": [ "\u001b[1m 2/110\u001b[0m \u001b[37m━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[1m8s\u001b[0m 77ms/step - accuracy: 0.0205 - loss: 6.2136 " ] }, { "name": "stderr", "output_type": "stream", "text": [ "I0000 00:00:1755331184.523981 73038 device_compiler.h:188] Compiled cluster using XLA! This line is logged at most once for the lifetime of the process.\n" ] }, { "name": "stdout", "output_type": "stream", "text": [ "\u001b[1m110/110\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m22s\u001b[0m 76ms/step - accuracy: 0.2544 - loss: 4.0960\n", "Epoch 2/3\n", "\u001b[1m110/110\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m8s\u001b[0m 75ms/step - accuracy: 0.8779 - loss: 0.6065\n", "Epoch 3/3\n", "\u001b[1m110/110\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m8s\u001b[0m 76ms/step - accuracy: 0.9620 - loss: 0.2097\n" ] }, { "data": { "text/plain": [ "