{ "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/experiment/cnn_htr_char_lines.yaml\"" ] }, { "cell_type": "code", "execution_count": 7, "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": 8, "id": "f939aa37-7b1d-45cc-885c-323c4540bda1", "metadata": {}, "outputs": [ { "data": { "text/plain": [ "{'defaults': [{'override /mapping': None}, {'override /criterion': None}, {'override /datamodule': None}, {'override /network': None}, {'override /model': None}, {'override /lr_schedulers': None}, {'override /optimizers': None}], 'criterion': {'_target_': 'torch.nn.CrossEntropyLoss', 'ignore_index': 3}, 'mapping': {'_target_': 'text_recognizer.data.emnist_mapping.EmnistMapping'}, 'callbacks': {'stochastic_weight_averaging': {'_target_': 'pytorch_lightning.callbacks.StochasticWeightAveraging', 'swa_epoch_start': 0.8, 'swa_lrs': 0.05, 'annealing_epochs': 10, 'annealing_strategy': 'cos', 'device': None}}, 'optimizers': {'madgrad': {'_target_': 'madgrad.MADGRAD', 'lr': 0.0001, 'momentum': 0.9, 'weight_decay': 0, 'eps': 1e-06, 'parameters': 'network'}}, 'lr_schedulers': {'network': {'_target_': 'torch.optim.lr_scheduler.CosineAnnealingLR', 'T_max': 1024, 'eta_min': 4.5e-06, 'last_epoch': -1, 'interval': 'epoch', 'monitor': 'val/loss'}}, 'datamodule': {'_target_': 'text_recognizer.data.iam_lines.IAMLines', 'batch_size': 24, 'num_workers': 12, 'train_fraction': 0.8, 'augment': True, 'pin_memory': False}, 'network': {'_target_': 'text_recognizer.networks.conv_transformer.ConvTransformer', 'input_dims': [1, 56, 1024], 'hidden_dim': 128, 'encoder_dim': 1280, 'dropout_rate': 0.2, 'num_classes': 58, 'pad_index': 3, 'encoder': {'_target_': 'text_recognizer.networks.encoders.efficientnet.EfficientNet', 'arch': 'b0', 'out_channels': 1280, 'stochastic_dropout_rate': 0.2, 'bn_momentum': 0.99, 'bn_eps': 0.001}, 'decoder': {'_target_': 'text_recognizer.networks.transformer.Decoder', 'dim': 128, 'depth': 3, 'num_heads': 4, 'attn_fn': 'text_recognizer.networks.transformer.attention.Attention', 'attn_kwargs': {'dim_head': 32, 'dropout_rate': 0.2}, 'norm_fn': 'torch.nn.LayerNorm', 'ff_fn': 'text_recognizer.networks.transformer.mlp.FeedForward', 'ff_kwargs': {'dim_out': None, 'expansion_factor': 4, 'glu': True, 'dropout_rate': 0.2}, 'cross_attend': True, 'pre_norm': True, 'rotary_emb': None}}, 'model': {'_target_': 'text_recognizer.models.transformer.TransformerLitModel', 'max_output_len': 89, 'start_token': '', 'end_token': '', 'pad_token': '

'}, 'trainer': {'_target_': 'pytorch_lightning.Trainer', 'stochastic_weight_avg': False, 'auto_scale_batch_size': 'binsearch', 'auto_lr_find': False, 'gradient_clip_val': 0, 'fast_dev_run': False, 'gpus': 1, 'precision': 16, 'max_epochs': 1024, 'terminate_on_nan': True, 'weights_summary': 'top', 'limit_train_batches': 1.0, 'limit_val_batches': 1.0, 'limit_test_batches': 1.0, 'resume_from_checkpoint': None, 'accumulate_grad_batches': 4, 'overfit_batches': 0.0}, 'summary': [[1, 1, 56, 1024], [1, 89]]}" ] }, "execution_count": 8, "metadata": {}, "output_type": "execute_result" } ], "source": [ "cfg" ] }, { "cell_type": "code", "execution_count": 23, "id": "2fdd47e6-aac0-47e9-b224-f46c1f015f17", "metadata": {}, "outputs": [ { "data": { "text/plain": [ "[[1, 1, 56, 1024], [1, 89]]" ] }, "execution_count": 23, "metadata": {}, "output_type": "execute_result" } ], "source": [ "list(map(lambda x: list(x), cfg.summary))" ] }, { "cell_type": "code", "execution_count": 17, "id": "92405411-f944-4519-9a6d-13b0761c2de6", "metadata": {}, "outputs": [ { "data": { "text/plain": [ "list" ] }, "execution_count": 17, "metadata": {}, "output_type": "execute_result" } ], "source": [ "type(list(cfg.summary))" ] }, { "cell_type": "code", "execution_count": 11, "id": "aaeab329-aeb0-4a1b-aa35-5a2aab81b1d0", "metadata": {}, "outputs": [], "source": [ "net = instantiate(cfg.network)" ] }, { "cell_type": "code", "execution_count": 12, "id": "618b997c-e6a6-4487-b70c-9d260cb556d3", "metadata": {}, "outputs": [], "source": [ "from torchinfo import summary" ] }, { "cell_type": "code", "execution_count": 24, "id": "25759b7b-8deb-4163-b75d-a1357c9fe88f", "metadata": {}, "outputs": [ { "data": { "text/plain": [ "====================================================================================================\n", "Layer (type:depth-idx) Output Shape Param #\n", "====================================================================================================\n", "ConvTransformer -- --\n", "├─EfficientNet: 1 -- --\n", "│ └─ModuleList: 2-1 -- --\n", "├─Decoder: 1 -- --\n", "│ └─ModuleList: 2-2 -- --\n", "│ │ └─ModuleList: 3-1 -- 2,097,536\n", "│ │ └─ModuleList: 3-2 -- 2,097,536\n", "│ │ └─ModuleList: 3-3 -- 198,016\n", "│ │ └─ModuleList: 3-4 -- 2,097,536\n", "│ │ └─ModuleList: 3-5 -- 2,097,536\n", "│ │ └─ModuleList: 3-6 -- 198,016\n", "│ │ └─ModuleList: 3-7 -- 2,097,536\n", "│ │ └─ModuleList: 3-8 -- 2,097,536\n", "│ │ └─ModuleList: 3-9 -- 198,016\n", "├─EfficientNet: 1-1 [1, 1280, 1, 32] --\n", "│ └─Sequential: 2-3 [1, 32, 28, 512] --\n", "│ │ └─ZeroPad2d: 3-10 [1, 1, 57, 1025] --\n", "│ │ └─Conv2d: 3-11 [1, 32, 28, 512] 288\n", "│ │ └─BatchNorm2d: 3-12 [1, 32, 28, 512] 64\n", "│ │ └─Mish: 3-13 [1, 32, 28, 512] --\n", "│ └─ModuleList: 2-1 -- --\n", "│ │ └─MBConvBlock: 3-14 [1, 16, 28, 512] 1,448\n", "│ │ └─MBConvBlock: 3-15 [1, 24, 14, 256] 9,864\n", "│ │ └─MBConvBlock: 3-16 [1, 24, 14, 256] 19,380\n", "│ │ └─MBConvBlock: 3-17 [1, 40, 7, 128] 24,020\n", "│ │ └─MBConvBlock: 3-18 [1, 40, 7, 128] 55,340\n", "│ │ └─MBConvBlock: 3-19 [1, 80, 3, 64] 61,180\n", "│ │ └─MBConvBlock: 3-20 [1, 80, 3, 64] 199,000\n", "│ │ └─MBConvBlock: 3-21 [1, 80, 3, 64] 199,000\n", "│ │ └─MBConvBlock: 3-22 [1, 112, 3, 64] 222,104\n", "│ │ └─MBConvBlock: 3-23 [1, 112, 3, 64] 396,872\n", "│ │ └─MBConvBlock: 3-24 [1, 112, 3, 64] 396,872\n", "│ │ └─MBConvBlock: 3-25 [1, 192, 1, 32] 450,792\n", "│ │ └─MBConvBlock: 3-26 [1, 192, 1, 32] 1,141,152\n", "│ │ └─MBConvBlock: 3-27 [1, 192, 1, 32] 1,141,152\n", "│ │ └─MBConvBlock: 3-28 [1, 192, 1, 32] 1,141,152\n", "│ │ └─MBConvBlock: 3-29 [1, 320, 1, 32] 1,270,432\n", "│ └─Sequential: 2-4 [1, 1280, 1, 32] --\n", "│ │ └─Conv2d: 3-30 [1, 1280, 1, 32] 409,600\n", "│ │ └─BatchNorm2d: 3-31 [1, 1280, 1, 32] 2,560\n", "├─Sequential: 1-2 [1, 128, 32] --\n", "│ └─Conv2d: 2-5 [1, 128, 1, 32] 163,968\n", "│ └─PositionalEncoding2D: 2-6 [1, 128, 1, 32] --\n", "│ └─Flatten: 2-7 [1, 128, 32] --\n", "├─Embedding: 1-3 [1, 89, 128] 7,424\n", "├─PositionalEncoding: 1-4 [1, 89, 128] --\n", "│ └─Dropout: 2-8 [1, 89, 128] --\n", "├─Decoder: 1-5 [1, 89, 128] --\n", "├─Linear: 1-6 [1, 89, 58] 7,482\n", "====================================================================================================\n", "Total params: 20,500,410\n", "Trainable params: 20,500,410\n", "Non-trainable params: 0\n", "Total mult-adds (M): 714.87\n", "====================================================================================================\n", "Input size (MB): 0.23\n", "Forward/backward pass size (MB): 184.29\n", "Params size (MB): 82.00\n", "Estimated Total Size (MB): 266.52\n", "====================================================================================================" ] }, "execution_count": 24, "metadata": {}, "output_type": "execute_result" } ], "source": [ "summary(net, list(map(lambda x: list(x), cfg.summary)), device=\"cpu\")" ] }, { "cell_type": "code", "execution_count": null, "id": "bf9d8d67-d7d2-4cf7-b166-377a79d5fd70", "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": "3.9.7" } }, "nbformat": 4, "nbformat_minor": 5 }