{ "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 }