{ "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", "import torch\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/vit_lines.yaml\"" ] }, { "cell_type": "code", "execution_count": 38, "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": 39, "id": "f939aa37-7b1d-45cc-885c-323c4540bda1", "metadata": {}, "outputs": [ { "data": { "text/plain": [ "{'_target_': 'text_recognizer.network.vit.VisionTransformer', 'image_height': 56, 'image_width': 1024, 'patch_height': 28, 'patch_width': 32, 'dim': 256, 'num_classes': 57, 'encoder': {'_target_': 'text_recognizer.network.transformer.encoder.Encoder', 'dim': 256, 'inner_dim': 1024, 'heads': 8, 'dim_head': 64, 'depth': 6, 'dropout_rate': 0.0}, 'decoder': {'_target_': 'text_recognizer.network.transformer.decoder.Decoder', 'dim': 256, 'inner_dim': 1024, 'heads': 8, 'dim_head': 64, 'depth': 6, 'dropout_rate': 0.0}, 'token_embedding': {'_target_': 'text_recognizer.network.transformer.embedding.token.TokenEmbedding', 'num_tokens': 57, 'dim': 256, 'use_l2': True}, 'pos_embedding': {'_target_': 'text_recognizer.network.transformer.embedding.absolute.AbsolutePositionalEmbedding', 'dim': 256, 'max_length': 89, 'use_l2': True}, 'tie_embeddings': True, 'pad_index': 3}" ] }, "execution_count": 39, "metadata": {}, "output_type": "execute_result" } ], "source": [ "cfg" ] }, { "cell_type": "code", "execution_count": 40, "id": "aaeab329-aeb0-4a1b-aa35-5a2aab81b1d0", "metadata": {}, "outputs": [], "source": [ "net = instantiate(cfg)" ] }, { "cell_type": "code", "execution_count": 41, "id": "618b997c-e6a6-4487-b70c-9d260cb556d3", "metadata": {}, "outputs": [], "source": [ "from torchinfo import summary" ] }, { "cell_type": "code", "execution_count": 43, "id": "7daf1f49", "metadata": {}, "outputs": [ { "data": { "text/plain": [ "====================================================================================================\n", "Layer (type:depth-idx) Output Shape Param #\n", "====================================================================================================\n", "VisionTransformer [1, 57, 89] --\n", "├─Sequential: 1-1 [1, 64, 256] --\n", "│ └─Rearrange: 2-1 [1, 64, 896] --\n", "│ └─LayerNorm: 2-2 [1, 64, 896] 1,792\n", "│ └─Linear: 2-3 [1, 64, 256] 229,632\n", "│ └─LayerNorm: 2-4 [1, 64, 256] 512\n", "├─Encoder: 1-2 [1, 64, 256] --\n", "│ └─ModuleList: 2-5 -- --\n", "│ │ └─ModuleList: 3-1 -- --\n", "│ │ │ └─Attention: 4-1 [1, 64, 256] 525,824\n", "│ │ │ └─FeedForward: 4-2 [1, 64, 256] 526,080\n", "│ │ └─ModuleList: 3-2 -- --\n", "│ │ │ └─Attention: 4-3 [1, 64, 256] 525,824\n", "│ │ │ └─FeedForward: 4-4 [1, 64, 256] 526,080\n", "│ │ └─ModuleList: 3-3 -- --\n", "│ │ │ └─Attention: 4-5 [1, 64, 256] 525,824\n", "│ │ │ └─FeedForward: 4-6 [1, 64, 256] 526,080\n", "│ │ └─ModuleList: 3-4 -- --\n", "│ │ │ └─Attention: 4-7 [1, 64, 256] 525,824\n", "│ │ │ └─FeedForward: 4-8 [1, 64, 256] 526,080\n", "│ │ └─ModuleList: 3-5 -- --\n", "│ │ │ └─Attention: 4-9 [1, 64, 256] 525,824\n", "│ │ │ └─FeedForward: 4-10 [1, 64, 256] 526,080\n", "│ │ └─ModuleList: 3-6 -- --\n", "│ │ │ └─Attention: 4-11 [1, 64, 256] 525,824\n", "│ │ │ └─FeedForward: 4-12 [1, 64, 256] 526,080\n", "│ └─LayerNorm: 2-6 [1, 64, 256] 512\n", "├─TokenEmbedding: 1-3 [1, 89, 256] --\n", "│ └─Embedding: 2-7 [1, 89, 256] 14,592\n", "├─AbsolutePositionalEmbedding: 1-4 [89, 256] --\n", "│ └─Embedding: 2-8 [89, 256] 22,784\n", "├─Decoder: 1-5 [1, 89, 256] --\n", "│ └─ModuleList: 2-9 -- --\n", "│ │ └─ModuleList: 3-7 -- --\n", "│ │ │ └─Attention: 4-13 [1, 89, 256] 525,824\n", "│ │ │ └─FeedForward: 4-14 [1, 89, 256] 526,080\n", "│ │ │ └─Attention: 4-15 [1, 89, 256] 525,824\n", "│ │ └─ModuleList: 3-8 -- --\n", "│ │ │ └─Attention: 4-16 [1, 89, 256] 525,824\n", "│ │ │ └─FeedForward: 4-17 [1, 89, 256] 526,080\n", "│ │ │ └─Attention: 4-18 [1, 89, 256] 525,824\n", "│ │ └─ModuleList: 3-9 -- --\n", "│ │ │ └─Attention: 4-19 [1, 89, 256] 525,824\n", "│ │ │ └─FeedForward: 4-20 [1, 89, 256] 526,080\n", "│ │ │ └─Attention: 4-21 [1, 89, 256] 525,824\n", "│ │ └─ModuleList: 3-10 -- --\n", "│ │ │ └─Attention: 4-22 [1, 89, 256] 525,824\n", "│ │ │ └─FeedForward: 4-23 [1, 89, 256] 526,080\n", "│ │ │ └─Attention: 4-24 [1, 89, 256] 525,824\n", "│ │ └─ModuleList: 3-11 -- --\n", "│ │ │ └─Attention: 4-25 [1, 89, 256] 525,824\n", "│ │ │ └─FeedForward: 4-26 [1, 89, 256] 526,080\n", "│ │ │ └─Attention: 4-27 [1, 89, 256] 525,824\n", "│ │ └─ModuleList: 3-12 -- --\n", "│ │ │ └─Attention: 4-28 [1, 89, 256] 525,824\n", "│ │ │ └─FeedForward: 4-29 [1, 89, 256] 526,080\n", "│ │ │ └─Attention: 4-30 [1, 89, 256] 525,824\n", "│ └─LayerNorm: 2-10 [1, 89, 256] 512\n", "====================================================================================================\n", "Total params: 16,048,128\n", "Trainable params: 16,048,128\n", "Non-trainable params: 0\n", "Total mult-adds (M): 18.03\n", "====================================================================================================\n", "Input size (MB): 0.23\n", "Forward/backward pass size (MB): 46.52\n", "Params size (MB): 64.16\n", "Estimated Total Size (MB): 110.91\n", "====================================================================================================" ] }, "execution_count": 43, "metadata": {}, "output_type": "execute_result" } ], "source": [ "summary(net, ((1, 1, 56, 1024), (1, 89)), device=\"cpu\", depth=4)" ] }, { "cell_type": "code", "execution_count": 20, "id": "1b1a8ac0-bd05-4076-90c2-2de6b740490d", "metadata": { "tags": [] }, "outputs": [], "source": [ "import torch" ] }, { "cell_type": "code", "execution_count": 24, "id": "248a0cb1", "metadata": {}, "outputs": [], "source": [ "t = net(torch.randn(1, 1, 56, 1024), torch.randint(1, 4, (1, 4)))" ] }, { "cell_type": "code", "execution_count": 25, "id": "c251a954-00ac-4680-87e4-f27b6ce06023", "metadata": { "tags": [] }, "outputs": [ { "data": { "text/plain": [ "torch.Size([1, 58, 4])" ] }, "execution_count": 25, "metadata": {}, "output_type": "execute_result" } ], "source": [ "t.shape" ] }, { "cell_type": "code", "execution_count": 17, "id": "02d82c5e-4e67-4f87-a539-393e4cf59b6e", "metadata": { "tags": [] }, "outputs": [], "source": [ "loss = torch.nn.CrossEntropyLoss()" ] }, { "cell_type": "code", "execution_count": 19, "id": "dc836993-a5d8-43b2-b41c-158a17990075", "metadata": { "tags": [] }, "outputs": [ { "data": { "text/plain": [ "tensor(4.0604, grad_fn=)" ] }, "execution_count": 19, "metadata": {}, "output_type": "execute_result" } ], "source": [ "loss(t.permute(0, 2, 1), torch.randint(0, 58, (1, 89)))" ] }, { "cell_type": "code", "execution_count": null, "id": "166bf656-aba6-4654-a530-dfce12666297", "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 }