From 9ee84b0557d1348211a2267e649db392e640dad0 Mon Sep 17 00:00:00 2001 From: Gustaf Rydholm Date: Thu, 30 Sep 2021 23:10:42 +0200 Subject: Add new notebooks --- notebooks/04-efficientnet-transformer.ipynb | 219 ++++++++++++++++++++++ notebooks/04-efficientnet.ipynb | 279 ++++++++++++++++++++++++++++ notebooks/04-vq-transformer.ipynb | 253 +++++++++++++++++++++++++ notebooks/04-vqvae.ipynb | 233 +++++++++++++++++++++++ 4 files changed, 984 insertions(+) create mode 100644 notebooks/04-efficientnet-transformer.ipynb create mode 100644 notebooks/04-efficientnet.ipynb create mode 100644 notebooks/04-vq-transformer.ipynb create mode 100644 notebooks/04-vqvae.ipynb (limited to 'notebooks') diff --git a/notebooks/04-efficientnet-transformer.ipynb b/notebooks/04-efficientnet-transformer.ipynb new file mode 100644 index 0000000..427c98c --- /dev/null +++ b/notebooks/04-efficientnet-transformer.ipynb @@ -0,0 +1,219 @@ +{ + "cells": [ + { + "cell_type": "code", + "execution_count": 2, + "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": 3, + "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": 4, + "id": "3cf50475-39f2-4642-a7d1-5bcbc0a036f7", + "metadata": {}, + "outputs": [], + "source": [ + "path = \"../training/conf/experiment/cnn_htr_char_lines.yaml\"" + ] + }, + { + "cell_type": "code", + "execution_count": 5, + "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": 6, + "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'}, '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': 8, 'num_workers': 12, 'train_fraction': 0.8, 'augment': False, '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': 2, '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}}" + ] + }, + "execution_count": 6, + "metadata": {}, + "output_type": "execute_result" + } + ], + "source": [ + "cfg" + ] + }, + { + "cell_type": "code", + "execution_count": 7, + "id": "aaeab329-aeb0-4a1b-aa35-5a2aab81b1d0", + "metadata": {}, + "outputs": [], + "source": [ + "net = instantiate(cfg.network)" + ] + }, + { + "cell_type": "code", + "execution_count": 9, + "id": "618b997c-e6a6-4487-b70c-9d260cb556d3", + "metadata": {}, + "outputs": [], + "source": [ + "from torchinfo import summary" + ] + }, + { + "cell_type": "code", + "execution_count": 10, + "id": "25759b7b-8deb-4163-b75d-a1357c9fe88f", + "metadata": {}, + "outputs": [ + { + "name": "stdout", + "output_type": "stream", + "text": [ + "torch.Size([4, 4, 89, 1024])\n", + "torch.Size([4, 4, 89, 1024])\n", + "torch.Size([4, 4, 89, 1024])\n", + "torch.Size([4, 4, 32, 1024])\n", + "torch.Size([4, 4, 89, 1024])\n", + "torch.Size([4, 4, 89, 1024])\n", + "torch.Size([4, 4, 89, 1024])\n", + "torch.Size([4, 4, 32, 1024])\n" + ] + }, + { + "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", + "├─EfficientNet: 1-1 [4, 1280, 1, 32] --\n", + "│ └─Sequential: 2-3 [4, 32, 28, 512] --\n", + "│ │ └─ZeroPad2d: 3-7 [4, 1, 57, 1025] --\n", + "│ │ └─Conv2d: 3-8 [4, 32, 28, 512] 288\n", + "│ │ └─BatchNorm2d: 3-9 [4, 32, 28, 512] 64\n", + "│ │ └─Mish: 3-10 [4, 32, 28, 512] --\n", + "│ └─ModuleList: 2-1 -- --\n", + "│ │ └─MBConvBlock: 3-11 [4, 16, 28, 512] 1,448\n", + "│ │ └─MBConvBlock: 3-12 [4, 24, 14, 256] 9,864\n", + "│ │ └─MBConvBlock: 3-13 [4, 24, 14, 256] 19,380\n", + "│ │ └─MBConvBlock: 3-14 [4, 40, 7, 128] 24,020\n", + "│ │ └─MBConvBlock: 3-15 [4, 40, 7, 128] 55,340\n", + "│ │ └─MBConvBlock: 3-16 [4, 80, 3, 64] 61,180\n", + "│ │ └─MBConvBlock: 3-17 [4, 80, 3, 64] 199,000\n", + "│ │ └─MBConvBlock: 3-18 [4, 80, 3, 64] 199,000\n", + "│ │ └─MBConvBlock: 3-19 [4, 112, 3, 64] 222,104\n", + "│ │ └─MBConvBlock: 3-20 [4, 112, 3, 64] 396,872\n", + "│ │ └─MBConvBlock: 3-21 [4, 112, 3, 64] 396,872\n", + "│ │ └─MBConvBlock: 3-22 [4, 192, 1, 32] 450,792\n", + "│ │ └─MBConvBlock: 3-23 [4, 192, 1, 32] 1,141,152\n", + "│ │ └─MBConvBlock: 3-24 [4, 192, 1, 32] 1,141,152\n", + "│ │ └─MBConvBlock: 3-25 [4, 192, 1, 32] 1,141,152\n", + "│ │ └─MBConvBlock: 3-26 [4, 320, 1, 32] 1,270,432\n", + "│ └─Sequential: 2-4 [4, 1280, 1, 32] --\n", + "│ │ └─Conv2d: 3-27 [4, 1280, 1, 32] 409,600\n", + "│ │ └─BatchNorm2d: 3-28 [4, 1280, 1, 32] 2,560\n", + "├─Sequential: 1-2 [4, 128, 32] --\n", + "│ └─Conv2d: 2-5 [4, 128, 1, 32] 163,968\n", + "│ └─PositionalEncoding2D: 2-6 [4, 128, 1, 32] --\n", + "│ └─Flatten: 2-7 [4, 128, 32] --\n", + "├─Embedding: 1-3 [4, 89, 128] 7,424\n", + "├─PositionalEncoding: 1-4 [4, 89, 128] --\n", + "│ └─Dropout: 2-8 [4, 89, 128] --\n", + "├─Decoder: 1-5 [4, 89, 128] --\n", + "├─Linear: 1-6 [4, 89, 58] 7,482\n", + "====================================================================================================\n", + "Total params: 16,107,322\n", + "Trainable params: 16,107,322\n", + "Non-trainable params: 0\n", + "Total mult-adds (G): 2.84\n", + "====================================================================================================\n", + "Input size (MB): 0.92\n", + "Forward/backward pass size (MB): 677.01\n", + "Params size (MB): 64.43\n", + "Estimated Total Size (MB): 742.36\n", + "====================================================================================================" + ] + }, + "execution_count": 10, + "metadata": {}, + "output_type": "execute_result" + } + ], + "source": [ + "summary(net, ((4, 1, 56, 1024), (4, 89)), device=\"cpu\")" + ] + } + ], + "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 +} diff --git a/notebooks/04-efficientnet.ipynb b/notebooks/04-efficientnet.ipynb new file mode 100644 index 0000000..4148e7d --- /dev/null +++ b/notebooks/04-efficientnet.ipynb @@ -0,0 +1,279 @@ +{ + "cells": [ + { + "cell_type": "code", + "execution_count": 2, + "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": 3, + "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": 4, + "id": "3cf50475-39f2-4642-a7d1-5bcbc0a036f7", + "metadata": {}, + "outputs": [], + "source": [ + "path = \"../training/conf/network/encoder/efficientnet.yaml\"" + ] + }, + { + "cell_type": "code", + "execution_count": 5, + "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": 6, + "id": "f939aa37-7b1d-45cc-885c-323c4540bda1", + "metadata": {}, + "outputs": [ + { + "data": { + "text/plain": [ + "{'_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}" + ] + }, + "execution_count": 6, + "metadata": {}, + "output_type": "execute_result" + } + ], + "source": [ + "cfg" + ] + }, + { + "cell_type": "code", + "execution_count": 7, + "id": "aaeab329-aeb0-4a1b-aa35-5a2aab81b1d0", + "metadata": {}, + "outputs": [], + "source": [ + "net = instantiate(cfg)" + ] + }, + { + "cell_type": "code", + "execution_count": 8, + "id": "618b997c-e6a6-4487-b70c-9d260cb556d3", + "metadata": {}, + "outputs": [], + "source": [ + "from torchinfo import summary" + ] + }, + { + "cell_type": "code", + "execution_count": 9, + "id": "25759b7b-8deb-4163-b75d-a1357c9fe88f", + "metadata": {}, + "outputs": [ + { + "data": { + "text/plain": [ + "==========================================================================================\n", + "Layer (type:depth-idx) Output Shape Param #\n", + "==========================================================================================\n", + "EfficientNet -- --\n", + "├─ModuleList: 1-1 -- --\n", + "├─Sequential: 1-2 [2, 32, 288, 320] --\n", + "│ └─ZeroPad2d: 2-1 [2, 1, 577, 641] --\n", + "│ └─Conv2d: 2-2 [2, 32, 288, 320] 288\n", + "│ └─BatchNorm2d: 2-3 [2, 32, 288, 320] 64\n", + "│ └─Mish: 2-4 [2, 32, 288, 320] --\n", + "├─ModuleList: 1-1 -- --\n", + "│ └─MBConvBlock: 2-5 [2, 16, 288, 320] --\n", + "│ │ └─Sequential: 3-1 [2, 32, 288, 320] 352\n", + "│ │ └─Sequential: 3-2 [2, 32, 288, 320] 552\n", + "│ │ └─Sequential: 3-3 [2, 16, 288, 320] 544\n", + "│ └─MBConvBlock: 2-6 [2, 24, 144, 160] --\n", + "│ │ └─Sequential: 3-4 [2, 96, 288, 320] 1,728\n", + "│ │ └─Sequential: 3-5 [2, 96, 144, 160] 1,056\n", + "│ │ └─Sequential: 3-6 [2, 96, 144, 160] 4,728\n", + "│ │ └─Sequential: 3-7 [2, 24, 144, 160] 2,352\n", + "│ └─MBConvBlock: 2-7 [2, 24, 144, 160] --\n", + "│ │ └─Sequential: 3-8 [2, 144, 144, 160] 3,744\n", + "│ │ └─Sequential: 3-9 [2, 144, 144, 160] 1,584\n", + "│ │ └─Sequential: 3-10 [2, 144, 144, 160] 10,548\n", + "│ │ └─Sequential: 3-11 [2, 24, 144, 160] 3,504\n", + "│ └─MBConvBlock: 2-8 [2, 40, 72, 80] --\n", + "│ │ └─Sequential: 3-12 [2, 144, 144, 160] 3,744\n", + "│ │ └─Sequential: 3-13 [2, 144, 72, 80] 3,888\n", + "│ │ └─Sequential: 3-14 [2, 144, 72, 80] 10,548\n", + "│ │ └─Sequential: 3-15 [2, 40, 72, 80] 5,840\n", + "│ └─MBConvBlock: 2-9 [2, 40, 72, 80] --\n", + "│ │ └─Sequential: 3-16 [2, 240, 72, 80] 10,080\n", + "│ │ └─Sequential: 3-17 [2, 240, 72, 80] 6,480\n", + "│ │ └─Sequential: 3-18 [2, 240, 72, 80] 29,100\n", + "│ │ └─Sequential: 3-19 [2, 40, 72, 80] 9,680\n", + "│ └─MBConvBlock: 2-10 [2, 80, 36, 40] --\n", + "│ │ └─Sequential: 3-20 [2, 240, 72, 80] 10,080\n", + "│ │ └─Sequential: 3-21 [2, 240, 36, 40] 2,640\n", + "│ │ └─Sequential: 3-22 [2, 240, 36, 40] 29,100\n", + "│ │ └─Sequential: 3-23 [2, 80, 36, 40] 19,360\n", + "│ └─MBConvBlock: 2-11 [2, 80, 36, 40] --\n", + "│ │ └─Sequential: 3-24 [2, 480, 36, 40] 39,360\n", + "│ │ └─Sequential: 3-25 [2, 480, 36, 40] 5,280\n", + "│ │ └─Sequential: 3-26 [2, 480, 36, 40] 115,800\n", + "│ │ └─Sequential: 3-27 [2, 80, 36, 40] 38,560\n", + "│ └─MBConvBlock: 2-12 [2, 80, 36, 40] --\n", + "│ │ └─Sequential: 3-28 [2, 480, 36, 40] 39,360\n", + "│ │ └─Sequential: 3-29 [2, 480, 36, 40] 5,280\n", + "│ │ └─Sequential: 3-30 [2, 480, 36, 40] 115,800\n", + "│ │ └─Sequential: 3-31 [2, 80, 36, 40] 38,560\n", + "│ └─MBConvBlock: 2-13 [2, 112, 36, 40] --\n", + "│ │ └─Sequential: 3-32 [2, 480, 36, 40] 39,360\n", + "│ │ └─Sequential: 3-33 [2, 480, 36, 40] 12,960\n", + "│ │ └─Sequential: 3-34 [2, 480, 36, 40] 115,800\n", + "│ │ └─Sequential: 3-35 [2, 112, 36, 40] 53,984\n", + "│ └─MBConvBlock: 2-14 [2, 112, 36, 40] --\n", + "│ │ └─Sequential: 3-36 [2, 672, 36, 40] 76,608\n", + "│ │ └─Sequential: 3-37 [2, 672, 36, 40] 18,144\n", + "│ │ └─Sequential: 3-38 [2, 672, 36, 40] 226,632\n", + "│ │ └─Sequential: 3-39 [2, 112, 36, 40] 75,488\n", + "│ └─MBConvBlock: 2-15 [2, 112, 36, 40] --\n", + "│ │ └─Sequential: 3-40 [2, 672, 36, 40] 76,608\n", + "│ │ └─Sequential: 3-41 [2, 672, 36, 40] 18,144\n", + "│ │ └─Sequential: 3-42 [2, 672, 36, 40] 226,632\n", + "│ │ └─Sequential: 3-43 [2, 112, 36, 40] 75,488\n", + "│ └─MBConvBlock: 2-16 [2, 192, 18, 20] --\n", + "│ │ └─Sequential: 3-44 [2, 672, 36, 40] 76,608\n", + "│ │ └─Sequential: 3-45 [2, 672, 18, 20] 18,144\n", + "│ │ └─Sequential: 3-46 [2, 672, 18, 20] 226,632\n", + "│ │ └─Sequential: 3-47 [2, 192, 18, 20] 129,408\n", + "│ └─MBConvBlock: 2-17 [2, 192, 18, 20] --\n", + "│ │ └─Sequential: 3-48 [2, 1152, 18, 20] 223,488\n", + "│ │ └─Sequential: 3-49 [2, 1152, 18, 20] 31,104\n", + "│ │ └─Sequential: 3-50 [2, 1152, 18, 20] 664,992\n", + "│ │ └─Sequential: 3-51 [2, 192, 18, 20] 221,568\n", + "│ └─MBConvBlock: 2-18 [2, 192, 18, 20] --\n", + "│ │ └─Sequential: 3-52 [2, 1152, 18, 20] 223,488\n", + "│ │ └─Sequential: 3-53 [2, 1152, 18, 20] 31,104\n", + "│ │ └─Sequential: 3-54 [2, 1152, 18, 20] 664,992\n", + "│ │ └─Sequential: 3-55 [2, 192, 18, 20] 221,568\n", + "│ └─MBConvBlock: 2-19 [2, 192, 18, 20] --\n", + "│ │ └─Sequential: 3-56 [2, 1152, 18, 20] 223,488\n", + "│ │ └─Sequential: 3-57 [2, 1152, 18, 20] 31,104\n", + "│ │ └─Sequential: 3-58 [2, 1152, 18, 20] 664,992\n", + "│ │ └─Sequential: 3-59 [2, 192, 18, 20] 221,568\n", + "│ └─MBConvBlock: 2-20 [2, 320, 18, 20] --\n", + "│ │ └─Sequential: 3-60 [2, 1152, 18, 20] 223,488\n", + "│ │ └─Sequential: 3-61 [2, 1152, 18, 20] 12,672\n", + "│ │ └─Sequential: 3-62 [2, 1152, 18, 20] 664,992\n", + "│ │ └─Sequential: 3-63 [2, 320, 18, 20] 369,280\n", + "├─Sequential: 1-3 [2, 1280, 18, 20] --\n", + "│ └─Conv2d: 2-21 [2, 1280, 18, 20] 409,600\n", + "│ └─BatchNorm2d: 2-22 [2, 1280, 18, 20] 2,560\n", + "==========================================================================================\n", + "Total params: 7,142,272\n", + "Trainable params: 7,142,272\n", + "Non-trainable params: 0\n", + "Total mult-adds (G): 11.27\n", + "==========================================================================================\n", + "Input size (MB): 2.95\n", + "Forward/backward pass size (MB): 1922.96\n", + "Params size (MB): 28.57\n", + "Estimated Total Size (MB): 1954.48\n", + "==========================================================================================" + ] + }, + "execution_count": 9, + "metadata": {}, + "output_type": "execute_result" + } + ], + "source": [ + "summary(net, (2, 1, 576, 640), device=\"cpu\")" + ] + }, + { + "cell_type": "code", + "execution_count": 11, + "id": "3ef95a63-7044-45bf-a085-faf5ea0c03ec", + "metadata": {}, + "outputs": [ + { + "ename": "TypeError", + "evalue": "'EfficientNet' object is not subscriptable", + "output_type": "error", + "traceback": [ + "\u001b[0;31m---------------------------------------------------------------------------\u001b[0m", + "\u001b[0;31mTypeError\u001b[0m Traceback (most recent call last)", + "\u001b[0;32m/tmp/ipykernel_2800/4064962505.py\u001b[0m in \u001b[0;36m\u001b[0;34m\u001b[0m\n\u001b[0;32m----> 1\u001b[0;31m \u001b[0mnet\u001b[0m\u001b[0;34m[\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m-\u001b[0m\u001b[0;36m2\u001b[0m\u001b[0;34m]\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m", + "\u001b[0;31mTypeError\u001b[0m: 'EfficientNet' object is not subscriptable" + ] + } + ], + "source": [ + "net[:-2]" + ] + }, + { + "cell_type": "code", + "execution_count": null, + "id": "62ca0d97-625c-474b-8d6c-d0caba79e198", + "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 +} diff --git a/notebooks/04-vq-transformer.ipynb b/notebooks/04-vq-transformer.ipynb new file mode 100644 index 0000000..69d2688 --- /dev/null +++ b/notebooks/04-vq-transformer.ipynb @@ -0,0 +1,253 @@ +{ + "cells": [ + { + "cell_type": "code", + "execution_count": null, + "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": null, + "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": null, + "id": "3cf50475-39f2-4642-a7d1-5bcbc0a036f7", + "metadata": {}, + "outputs": [], + "source": [ + "path = \"../training/conf/experiment/vqgan_htr_char_iam_lines.yaml\"" + ] + }, + { + "cell_type": "code", + "execution_count": null, + "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": null, + "id": "f939aa37-7b1d-45cc-885c-323c4540bda1", + "metadata": {}, + "outputs": [], + "source": [ + "cfg" + ] + }, + { + "cell_type": "code", + "execution_count": null, + "id": "aaeab329-aeb0-4a1b-aa35-5a2aab81b1d0", + "metadata": {}, + "outputs": [], + "source": [ + "net = instantiate(cfg.network)" + ] + }, + { + "cell_type": "code", + "execution_count": null, + "id": "a564ac7a-b67f-4bc1-af36-0fe0a58c1bc9", + "metadata": {}, + "outputs": [], + "source": [ + "import torch" + ] + }, + { + "cell_type": "code", + "execution_count": null, + "id": "aeddcc5c-e48d-4d90-8efa-963011ef40bc", + "metadata": {}, + "outputs": [], + "source": [ + "x = torch.randn((16, 1, 16, 64))\n", + "y = torch.randint(0, 56, (16, 89))" + ] + }, + { + "cell_type": "code", + "execution_count": null, + "id": "0f0d78bc-7e0a-4d06-8e38-49b29ad25933", + "metadata": {}, + "outputs": [], + "source": [ + "y.shape" + ] + }, + { + "cell_type": "code", + "execution_count": null, + "id": "e9f4ee2a-c93f-4461-8d75-40c8c12d9d48", + "metadata": {}, + "outputs": [], + "source": [ + "yy = net(x, y)" + ] + }, + { + "cell_type": "code", + "execution_count": null, + "id": "7a7493a9-0e1d-46ef-8180-27605e18d082", + "metadata": {}, + "outputs": [], + "source": [ + "yy[0].shape" + ] + }, + { + "cell_type": "code", + "execution_count": null, + "id": "75bc9695-2afd-455c-a4fb-2e182456ccbd", + "metadata": {}, + "outputs": [], + "source": [ + "z = torch.randn((16, 8, 32))" + ] + }, + { + "cell_type": "code", + "execution_count": null, + "id": "3df6f9a0-6e66-4f46-a5b7-c0bb71b16b9b", + "metadata": {}, + "outputs": [], + "source": [ + "z, _ = net.encode(x)" + ] + }, + { + "cell_type": "code", + "execution_count": null, + "id": "6d6e9dd1-c56e-4169-8216-bcc84ea980e3", + "metadata": {}, + "outputs": [], + "source": [ + "z.shape" + ] + }, + { + "cell_type": "code", + "execution_count": null, + "id": "8f1539cb-b9b2-40b7-a843-d7479ddbddd7", + "metadata": {}, + "outputs": [], + "source": [ + "yy = net.decode(z, y[:, :2])" + ] + }, + { + "cell_type": "code", + "execution_count": null, + "id": "5cdba0a9-da7d-4e33-b209-7f360d1a38e5", + "metadata": {}, + "outputs": [], + "source": [ + "yy.shape" + ] + }, + { + "cell_type": "code", + "execution_count": null, + "id": "6da8065f-f93f-4aec-a60e-408712a28c3b", + "metadata": {}, + "outputs": [], + "source": [ + "torch.argmax(yy,dim=-2).shape" + ] + }, + { + "cell_type": "code", + "execution_count": null, + "id": "beabbda7-6a1f-4294-8f01-f9d866ffe088", + "metadata": {}, + "outputs": [], + "source": [ + "yy[0].shape" + ] + }, + { + "cell_type": "code", + "execution_count": null, + "id": "618b997c-e6a6-4487-b70c-9d260cb556d3", + "metadata": {}, + "outputs": [], + "source": [ + "from torchinfo import summary" + ] + }, + { + "cell_type": "code", + "execution_count": null, + "id": "25759b7b-8deb-4163-b75d-a1357c9fe88f", + "metadata": {}, + "outputs": [], + "source": [ + "summary(net, (1, 1, 576, 640), device=\"cpu\")" + ] + }, + { + "cell_type": "code", + "execution_count": null, + "id": "62ca0d97-625c-474b-8d6c-d0caba79e198", + "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 +} diff --git a/notebooks/04-vqvae.ipynb b/notebooks/04-vqvae.ipynb new file mode 100644 index 0000000..1b31671 --- /dev/null +++ b/notebooks/04-vqvae.ipynb @@ -0,0 +1,233 @@ +{ + "cells": [ + { + "cell_type": "code", + "execution_count": 7, + "id": "136a80f5-10e1-40c4-973a-a7eb7939bb1f", + "metadata": {}, + "outputs": [ + { + "name": "stdout", + "output_type": "stream", + "text": [ + "The autoreload extension is already loaded. To reload it, use:\n", + " %reload_ext autoreload\n" + ] + } + ], + "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", + "from hydra import compose, initialize\n", + "from omegaconf import OmegaConf\n", + "from hydra.utils import instantiate\n", + "from torchinfo import summary\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": "1a0fb9ca-1886-4fd4-839f-dc111a450cfd", + "metadata": {}, + "outputs": [], + "source": [ + "path = \"../training/conf/network/vqvae.yaml\"" + ] + }, + { + "cell_type": "code", + "execution_count": 5, + "id": "0182a614-5781-44a6-b659-008e7c584fa7", + "metadata": {}, + "outputs": [ + { + "name": "stdout", + "output_type": "stream", + "text": [ + "encoder:\n", + " _target_: text_recognizer.networks.vqvae.encoder.Encoder\n", + " in_channels: 1\n", + " hidden_dim: 32\n", + " channels_multipliers:\n", + " - 1\n", + " - 2\n", + " - 4\n", + " dropout_rate: 0.0\n", + " activation: mish\n", + " use_norm: true\n", + " num_residuals: 4\n", + " residual_channels: 32\n", + "decoder:\n", + " _target_: text_recognizer.networks.vqvae.decoder.Decoder\n", + " out_channels: 1\n", + " hidden_dim: 32\n", + " channels_multipliers:\n", + " - 4\n", + " - 2\n", + " - 1\n", + " dropout_rate: 0.0\n", + " activation: mish\n", + " use_norm: true\n", + " num_residuals: 4\n", + " residual_channels: 32\n", + "_target_: text_recognizer.networks.vqvae.vqvae.VQVAE\n", + "hidden_dim: 128\n", + "embedding_dim: 32\n", + "num_embeddings: 8192\n", + "decay: 0.99\n", + "\n" + ] + }, + { + "name": "stderr", + "output_type": "stream", + "text": [ + "/home/aktersnurra/.cache/pypoetry/virtualenvs/text-recognizer-ejNaVa9M-py3.9/lib/python3.9/site-packages/hydra/_internal/defaults_list.py:251: UserWarning: In 'vqvae': Defaults list is missing `_self_`. See https://hydra.cc/docs/upgrades/1.0_to_1.1/default_composition_order for more information\n", + " warnings.warn(msg, UserWarning)\n" + ] + } + ], + "source": [ + "with initialize(config_path=\"../training/conf/network/\", job_name=\"test_app\"):\n", + " cfg = compose(config_name=\"vqvae\")\n", + " print(OmegaConf.to_yaml(cfg))" + ] + }, + { + "cell_type": "code", + "execution_count": 6, + "id": "a500f94c-7dae-477e-a3fb-2a2d62ee7b72", + "metadata": {}, + "outputs": [], + "source": [ + "net = instantiate(cfg)" + ] + }, + { + "cell_type": "code", + "execution_count": 9, + "id": "7f3b3559-5e23-485e-bf57-9405568a1fbf", + "metadata": {}, + "outputs": [ + { + "data": { + "text/plain": [ + "====================================================================================================\n", + "Layer (type:depth-idx) Output Shape Param #\n", + "====================================================================================================\n", + "VQVAE -- --\n", + "├─Encoder: 1-1 [1, 128, 72, 80] --\n", + "│ └─Sequential: 2-1 [1, 128, 72, 80] --\n", + "│ │ └─Conv2d: 3-1 [1, 32, 576, 640] 320\n", + "│ │ └─Normalize: 3-2 [1, 32, 576, 640] 64\n", + "│ │ └─Mish: 3-3 [1, 32, 576, 640] --\n", + "│ │ └─Mish: 3-4 [1, 32, 576, 640] --\n", + "│ │ └─Mish: 3-5 [1, 32, 576, 640] --\n", + "│ │ └─Conv2d: 3-6 [1, 32, 288, 320] 16,416\n", + "│ │ └─Normalize: 3-7 [1, 32, 288, 320] 64\n", + "│ │ └─Mish: 3-8 [1, 32, 288, 320] --\n", + "│ │ └─Mish: 3-9 [1, 32, 288, 320] --\n", + "│ │ └─Mish: 3-10 [1, 32, 288, 320] --\n", + "│ │ └─Conv2d: 3-11 [1, 64, 144, 160] 32,832\n", + "│ │ └─Normalize: 3-12 [1, 64, 144, 160] 128\n", + "│ │ └─Mish: 3-13 [1, 64, 144, 160] --\n", + "│ │ └─Mish: 3-14 [1, 64, 144, 160] --\n", + "│ │ └─Mish: 3-15 [1, 64, 144, 160] --\n", + "│ │ └─Conv2d: 3-16 [1, 128, 72, 80] 131,200\n", + "│ │ └─Residual: 3-17 [1, 128, 72, 80] 41,280\n", + "│ │ └─Residual: 3-18 [1, 128, 72, 80] 41,280\n", + "│ │ └─Residual: 3-19 [1, 128, 72, 80] 41,280\n", + "│ │ └─Residual: 3-20 [1, 128, 72, 80] 41,280\n", + "├─Conv2d: 1-2 [1, 32, 72, 80] 4,128\n", + "├─VectorQuantizer: 1-3 [1, 32, 72, 80] --\n", + "├─Conv2d: 1-4 [1, 128, 72, 80] 4,224\n", + "├─Decoder: 1-5 [1, 1, 576, 640] --\n", + "│ └─Sequential: 2-2 [1, 1, 576, 640] --\n", + "│ │ └─Residual: 3-21 [1, 128, 72, 80] 41,280\n", + "│ │ └─Residual: 3-22 [1, 128, 72, 80] 41,280\n", + "│ │ └─Residual: 3-23 [1, 128, 72, 80] 41,280\n", + "│ │ └─Residual: 3-24 [1, 128, 72, 80] 41,280\n", + "│ │ └─Normalize: 3-25 [1, 128, 72, 80] 256\n", + "│ │ └─Mish: 3-26 [1, 128, 72, 80] --\n", + "│ │ └─Mish: 3-27 [1, 128, 72, 80] --\n", + "│ │ └─Mish: 3-28 [1, 128, 72, 80] --\n", + "│ │ └─ConvTranspose2d: 3-29 [1, 64, 144, 160] 131,136\n", + "│ │ └─Normalize: 3-30 [1, 64, 144, 160] 128\n", + "│ │ └─Mish: 3-31 [1, 64, 144, 160] --\n", + "│ │ └─Mish: 3-32 [1, 64, 144, 160] --\n", + "│ │ └─Mish: 3-33 [1, 64, 144, 160] --\n", + "│ │ └─ConvTranspose2d: 3-34 [1, 32, 288, 320] 32,800\n", + "│ │ └─Normalize: 3-35 [1, 32, 288, 320] 64\n", + "│ │ └─Mish: 3-36 [1, 32, 288, 320] --\n", + "│ │ └─Mish: 3-37 [1, 32, 288, 320] --\n", + "│ │ └─Mish: 3-38 [1, 32, 288, 320] --\n", + "│ │ └─ConvTranspose2d: 3-39 [1, 32, 576, 640] 16,416\n", + "│ │ └─Normalize: 3-40 [1, 32, 576, 640] 64\n", + "│ │ └─Conv2d: 3-41 [1, 1, 576, 640] 289\n", + "====================================================================================================\n", + "Total params: 700,769\n", + "Trainable params: 700,769\n", + "Non-trainable params: 0\n", + "Total mult-adds (G): 17.28\n", + "====================================================================================================\n", + "Input size (MB): 1.47\n", + "Forward/backward pass size (MB): 659.13\n", + "Params size (MB): 2.80\n", + "Estimated Total Size (MB): 663.41\n", + "====================================================================================================" + ] + }, + "execution_count": 9, + "metadata": {}, + "output_type": "execute_result" + } + ], + "source": [ + "summary(net, (1, 1, 576, 640), device=\"cpu\")" + ] + }, + { + "cell_type": "code", + "execution_count": null, + "id": "9f880b03-d641-4640-acd3-aa5666ca5184", + "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 +} -- cgit v1.2.3-70-g09d2