Update notebooks

1 files changed, 249 insertions, 0 deletions

diff --git a/notebooks/04-conv-transformer.ipynb b/notebooks/04-conv-transformer.ipynb
new file mode 100644
index 0000000..3303b63
--- /dev/null
+++ b/notebooks/04-conv-transformer.ipynb
@@ -0,0 +1,249 @@
+{
+ "cells": [
+  {
+   "cell_type": "code",
+   "execution_count": 1,
+   "id": "7c02ae76-b540-4b16-9492-e9210b3b9249",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "import os\n",
+    "os.environ['CUDA_VISIBLE_DEVICE'] = ''\n",
+    "import random\n",
+    "\n",
+    "%matplotlib inline\n",
+    "import matplotlib.pyplot as plt\n",
+    "\n",
+    "import numpy as np\n",
+    "from omegaconf import OmegaConf\n",
+    "\n",
+    "%load_ext autoreload\n",
+    "%autoreload 2\n",
+    "\n",
+    "from importlib.util import find_spec\n",
+    "if find_spec(\"text_recognizer\") is None:\n",
+    "    import sys\n",
+    "    sys.path.append('..')"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 2,
+   "id": "ccdb6dde-47e5-429a-88f2-0764fb7e259a",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "from hydra import compose, initialize\n",
+    "from omegaconf import OmegaConf\n",
+    "from hydra.utils import instantiate"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 3,
+   "id": "3cf50475-39f2-4642-a7d1-5bcbc0a036f7",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "path = \"../training/conf/network/conv_transformer.yaml\""
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 9,
+   "id": "e52ecb01-c975-4e55-925d-1182c7aea473",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "with open(path, \"rb\") as f:\n",
+    "    cfg = OmegaConf.load(f)"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 10,
+   "id": "f939aa37-7b1d-45cc-885c-323c4540bda1",
+   "metadata": {},
+   "outputs": [
+    {
+     "data": {
+      "text/plain": [
+       "{'_target_': 'text_recognizer.networks.ConvTransformer', 'input_dims': [1, 1, 576, 640], 'hidden_dim': 128, 'num_classes': 58, 'pad_index': 3, 'encoder': {'_target_': 'text_recognizer.networks.EfficientNet', 'arch': 'b0', 'stochastic_dropout_rate': 0.2, 'bn_momentum': 0.99, 'bn_eps': 0.001, 'depth': 5, 'out_channels': 128, 'stride': [2, 1]}, 'decoder': {'_target_': 'text_recognizer.networks.transformer.Decoder', 'depth': 6, 'block': {'_target_': 'text_recognizer.networks.transformer.DecoderBlock', 'self_attn': {'_target_': 'text_recognizer.networks.transformer.Attention', 'dim': 128, 'num_heads': 8, 'dim_head': 64, 'dropout_rate': 0.4, 'causal': True, 'rotary_embedding': {'_target_': 'text_recognizer.networks.transformer.RotaryEmbedding', 'dim': 64}}, 'cross_attn': {'_target_': 'text_recognizer.networks.transformer.Attention', 'dim': 128, 'num_heads': 8, 'dim_head': 64, 'dropout_rate': 0.4, 'causal': False}, 'norm': {'_target_': 'text_recognizer.networks.transformer.RMSNorm', 'dim': 128}, 'ff': {'_target_': 'text_recognizer.networks.transformer.FeedForward', 'dim': 128, 'dim_out': None, 'expansion_factor': 2, 'glu': True, 'dropout_rate': 0.4}}}, 'pixel_embedding': {'_target_': 'text_recognizer.networks.transformer.AxialPositionalEmbedding', 'dim': 128, 'shape': [18, 80]}}"
+      ]
+     },
+     "execution_count": 10,
+     "metadata": {},
+     "output_type": "execute_result"
+    }
+   ],
+   "source": [
+    "cfg"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 11,
+   "id": "aaeab329-aeb0-4a1b-aa35-5a2aab81b1d0",
+   "metadata": {
+    "scrolled": false
+   },
+   "outputs": [],
+   "source": [
+    "net = instantiate(cfg)"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 12,
+   "id": "618b997c-e6a6-4487-b70c-9d260cb556d3",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "from torchinfo import summary"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 13,
+   "id": "25759b7b-8deb-4163-b75d-a1357c9fe88f",
+   "metadata": {
+    "scrolled": false
+   },
+   "outputs": [
+    {
+     "data": {
+      "text/plain": [
+       "==============================================================================================================\n",
+       "Layer (type:depth-idx)                                       Output Shape              Param #\n",
+       "==============================================================================================================\n",
+       "ConvTransformer                                              [1, 58, 682]              --\n",
+       "├─EfficientNet: 1-1                                          [1, 128, 18, 80]          850,044\n",
+       "│    └─Sequential: 2-1                                       [1, 16, 288, 320]         --\n",
+       "│    │    └─ZeroPad2d: 3-1                                   [1, 1, 577, 641]          --\n",
+       "│    │    └─Conv2d: 3-2                                      [1, 16, 288, 320]         144\n",
+       "│    │    └─BatchNorm2d: 3-3                                 [1, 16, 288, 320]         32\n",
+       "│    │    └─Mish: 3-4                                        [1, 16, 288, 320]         --\n",
+       "│    └─ModuleList: 2                                         --                        --\n",
+       "│    │    └─MBConvBlock: 3-5                                 [1, 16, 288, 320]         --\n",
+       "│    │    │    └─Depthwise: 4-1                              [1, 16, 288, 320]         176\n",
+       "│    │    │    └─SqueezeAndExcite: 4-2                       [1, 16, 288, 320]         148\n",
+       "│    │    │    └─Pointwise: 4-3                              [1, 16, 288, 320]         288\n",
+       "│    │    └─MBConvBlock: 3-6                                 [1, 24, 144, 160]         --\n",
+       "│    │    │    └─InvertedBottleneck: 4-4                     [1, 96, 288, 320]         1,728\n",
+       "│    │    │    └─Depthwise: 4-5                              [1, 96, 144, 160]         1,056\n",
+       "│    │    │    └─SqueezeAndExcite: 4-6                       [1, 96, 144, 160]         868\n",
+       "│    │    │    └─Pointwise: 4-7                              [1, 24, 144, 160]         2,352\n",
+       "│    │    └─MBConvBlock: 3-7                                 [1, 24, 144, 160]         --\n",
+       "│    │    │    └─InvertedBottleneck: 4-8                     [1, 144, 144, 160]        3,744\n",
+       "│    │    │    └─Depthwise: 4-9                              [1, 144, 144, 160]        1,584\n",
+       "│    │    │    └─SqueezeAndExcite: 4-10                      [1, 144, 144, 160]        1,878\n",
+       "│    │    │    └─Pointwise: 4-11                             [1, 24, 144, 160]         3,504\n",
+       "│    │    └─MBConvBlock: 3-8                                 [1, 40, 72, 80]           --\n",
+       "│    │    │    └─InvertedBottleneck: 4-12                    [1, 144, 144, 160]        3,744\n",
+       "│    │    │    └─Depthwise: 4-13                             [1, 144, 72, 80]          3,888\n",
+       "│    │    │    └─SqueezeAndExcite: 4-14                      [1, 144, 72, 80]          1,878\n",
+       "│    │    │    └─Pointwise: 4-15                             [1, 40, 72, 80]           5,840\n",
+       "│    │    └─MBConvBlock: 3-9                                 [1, 40, 72, 80]           --\n",
+       "│    │    │    └─InvertedBottleneck: 4-16                    [1, 240, 72, 80]          10,080\n",
+       "│    │    │    └─Depthwise: 4-17                             [1, 240, 72, 80]          6,480\n",
+       "│    │    │    └─SqueezeAndExcite: 4-18                      [1, 240, 72, 80]          5,050\n",
+       "│    │    │    └─Pointwise: 4-19                             [1, 40, 72, 80]           9,680\n",
+       "│    │    └─MBConvBlock: 3-10                                [1, 80, 36, 80]           --\n",
+       "│    │    │    └─InvertedBottleneck: 4-20                    [1, 240, 72, 80]          10,080\n",
+       "│    │    │    └─Depthwise: 4-21                             [1, 240, 36, 80]          2,640\n",
+       "│    │    │    └─SqueezeAndExcite: 4-22                      [1, 240, 36, 80]          5,050\n",
+       "│    │    │    └─Pointwise: 4-23                             [1, 80, 36, 80]           19,360\n",
+       "│    │    └─MBConvBlock: 3-11                                [1, 80, 36, 80]           --\n",
+       "│    │    │    └─InvertedBottleneck: 4-24                    [1, 480, 36, 80]          39,360\n",
+       "│    │    │    └─Depthwise: 4-25                             [1, 480, 36, 80]          5,280\n",
+       "│    │    │    └─SqueezeAndExcite: 4-26                      [1, 480, 36, 80]          19,700\n",
+       "│    │    │    └─Pointwise: 4-27                             [1, 80, 36, 80]           38,560\n",
+       "│    │    └─MBConvBlock: 3-12                                [1, 80, 36, 80]           --\n",
+       "│    │    │    └─InvertedBottleneck: 4-28                    [1, 480, 36, 80]          39,360\n",
+       "│    │    │    └─Depthwise: 4-29                             [1, 480, 36, 80]          5,280\n",
+       "│    │    │    └─SqueezeAndExcite: 4-30                      [1, 480, 36, 80]          19,700\n",
+       "│    │    │    └─Pointwise: 4-31                             [1, 80, 36, 80]           38,560\n",
+       "│    │    └─MBConvBlock: 3-13                                [1, 112, 18, 80]          --\n",
+       "│    │    │    └─InvertedBottleneck: 4-32                    [1, 480, 36, 80]          39,360\n",
+       "│    │    │    └─Depthwise: 4-33                             [1, 480, 18, 80]          12,960\n",
+       "│    │    │    └─SqueezeAndExcite: 4-34                      [1, 480, 18, 80]          19,700\n",
+       "│    │    │    └─Pointwise: 4-35                             [1, 112, 18, 80]          53,984\n",
+       "│    │    └─MBConvBlock: 3-14                                [1, 112, 18, 80]          --\n",
+       "│    │    │    └─InvertedBottleneck: 4-36                    [1, 672, 18, 80]          76,608\n",
+       "│    │    │    └─Depthwise: 4-37                             [1, 672, 18, 80]          18,144\n",
+       "│    │    │    └─SqueezeAndExcite: 4-38                      [1, 672, 18, 80]          38,332\n",
+       "│    │    │    └─Pointwise: 4-39                             [1, 112, 18, 80]          75,488\n",
+       "│    │    └─MBConvBlock: 3-15                                [1, 112, 18, 80]          --\n",
+       "│    │    │    └─InvertedBottleneck: 4-40                    [1, 672, 18, 80]          76,608\n",
+       "│    │    │    └─Depthwise: 4-41                             [1, 672, 18, 80]          18,144\n",
+       "│    │    │    └─SqueezeAndExcite: 4-42                      [1, 672, 18, 80]          38,332\n",
+       "│    │    │    └─Pointwise: 4-43                             [1, 112, 18, 80]          75,488\n",
+       "│    └─Sequential: 2-2                                       [1, 128, 18, 80]          --\n",
+       "│    │    └─Conv2d: 3-16                                     [1, 128, 18, 80]          14,336\n",
+       "│    │    └─BatchNorm2d: 3-17                                [1, 128, 18, 80]          256\n",
+       "│    │    └─Dropout: 3-18                                    [1, 128, 18, 80]          --\n",
+       "├─Conv2d: 1-2                                                [1, 128, 18, 80]          16,512\n",
+       "├─AxialPositionalEmbedding: 1-3                              [1, 128, 18, 80]          12,544\n",
+       "├─Embedding: 1-4                                             [1, 682, 128]             7,424\n",
+       "├─Decoder: 1-5                                               [1, 682, 128]             --\n",
+       "│    └─ModuleList: 2                                         --                        --\n",
+       "│    │    └─DecoderBlock: 3-19                               [1, 682, 128]             --\n",
+       "│    │    └─DecoderBlock: 3-20                               [1, 682, 128]             --\n",
+       "│    │    └─DecoderBlock: 3-21                               [1, 682, 128]             --\n",
+       "│    │    └─DecoderBlock: 3-22                               [1, 682, 128]             --\n",
+       "│    │    └─DecoderBlock: 3-23                               [1, 682, 128]             --\n",
+       "│    │    └─DecoderBlock: 3-24                               [1, 682, 128]             --\n",
+       "├─Linear: 1-6                                                [1, 682, 58]              7,482\n",
+       "==============================================================================================================\n",
+       "Total params: 4,652,006\n",
+       "Trainable params: 4,652,006\n",
+       "Non-trainable params: 0\n",
+       "Total mult-adds (G): 2.44\n",
+       "==============================================================================================================\n",
+       "Input size (MB): 1.48\n",
+       "Forward/backward pass size (MB): 1041.70\n",
+       "Params size (MB): 18.61\n",
+       "Estimated Total Size (MB): 1061.78\n",
+       "=============================================================================================================="
+      ]
+     },
+     "execution_count": 13,
+     "metadata": {},
+     "output_type": "execute_result"
+    }
+   ],
+   "source": [
+    "summary(net, ((1, 1, 576, 640), (1, 682)), device=\"cpu\", depth=4)"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "506f01a3",
+   "metadata": {},
+   "outputs": [],
+   "source": []
+  }
+ ],
+ "metadata": {
+  "kernelspec": {
+   "display_name": "Python 3 (ipykernel)",
+   "language": "python",
+   "name": "python3"
+  },
+  "language_info": {
+   "codemirror_mode": {
+    "name": "ipython",
+    "version": 3
+   },
+   "file_extension": ".py",
+   "mimetype": "text/x-python",
+   "name": "python",
+   "nbconvert_exporter": "python",
+   "pygments_lexer": "ipython3",
+   "version": "3.9.4"
+  }
+ },
+ "nbformat": 4,
+ "nbformat_minor": 5
+}