diff options
| -rw-r--r-- | notebooks/04-convnext.ipynb | 181 |
1 files changed, 111 insertions, 70 deletions
diff --git a/notebooks/04-convnext.ipynb b/notebooks/04-convnext.ipynb index 6d70884..1779d13 100644 --- a/notebooks/04-convnext.ipynb +++ b/notebooks/04-convnext.ipynb @@ -2,19 +2,10 @@ "cells": [ { "cell_type": "code", - "execution_count": 61, + "execution_count": 1, "id": "7c02ae76-b540-4b16-9492-e9210b3b9249", "metadata": {}, - "outputs": [ - { - "name": "stdout", - "output_type": "stream", - "text": [ - "The autoreload extension is already loaded. To reload it, use:\n", - " %reload_ext autoreload\n" - ] - } - ], + "outputs": [], "source": [ "import os\n", "os.environ['CUDA_VISIBLE_DEVICE'] = ''\n", @@ -37,7 +28,7 @@ }, { "cell_type": "code", - "execution_count": 62, + "execution_count": 2, "id": "ccdb6dde-47e5-429a-88f2-0764fb7e259a", "metadata": {}, "outputs": [], @@ -49,7 +40,7 @@ }, { "cell_type": "code", - "execution_count": 63, + "execution_count": 3, "id": "3cf50475-39f2-4642-a7d1-5bcbc0a036f7", "metadata": {}, "outputs": [], @@ -59,7 +50,7 @@ }, { "cell_type": "code", - "execution_count": 64, + "execution_count": 4, "id": "e52ecb01-c975-4e55-925d-1182c7aea473", "metadata": {}, "outputs": [], @@ -70,17 +61,17 @@ }, { "cell_type": "code", - "execution_count": 65, + "execution_count": 5, "id": "f939aa37-7b1d-45cc-885c-323c4540bda1", "metadata": {}, "outputs": [ { "data": { "text/plain": [ - "{'_target_': 'text_recognizer.networks.convnext.ConvNext', 'dim': 16, 'dim_mults': [2, 4, 8], 'depths': [3, 3, 6], 'downsampling_factors': [[2, 2], [2, 2], [2, 2]]}" + "{'_target_': 'text_recognizer.networks.convnext.ConvNext', 'dim': 16, 'dim_mults': [2, 4, 8], 'depths': [3, 3, 6], 'downsampling_factors': [[2, 2], [2, 2], [2, 2]], 'attn': {'_target_': 'text_recognizer.networks.convnext.TransformerBlock', 'attn': {'_target_': 'text_recognizer.networks.convnext.Attention', 'dim': 128, 'heads': 4, 'dim_head': 64, 'scale': 8}, 'ff': {'_target_': 'text_recognizer.networks.convnext.FeedForward', 'dim': 128, 'mult': 4}}}" ] }, - "execution_count": 65, + "execution_count": 5, "metadata": {}, "output_type": "execute_result" } @@ -91,8 +82,20 @@ }, { "cell_type": "code", - "execution_count": 66, - "id": "aaeab329-aeb0-4a1b-aa35-5a2aab81b1d0", + "execution_count": 10, + "id": "a2b420c1", + "metadata": {}, + "outputs": [], + "source": [ + "cfg.dim_mults = [2, 4, 8, 8, 8]\n", + "cfg.depths = [3, 3, 3, 3, 6]\n", + "cfg.downsampling_factors = [[2, 2], [2, 2], [2, 2], [2, 1], [2, 1]]" + ] + }, + { + "cell_type": "code", + "execution_count": 11, + "id": "c9589350", "metadata": {}, "outputs": [], "source": [ @@ -101,7 +104,7 @@ }, { "cell_type": "code", - "execution_count": 67, + "execution_count": 12, "id": "618b997c-e6a6-4487-b70c-9d260cb556d3", "metadata": {}, "outputs": [], @@ -111,7 +114,7 @@ }, { "cell_type": "code", - "execution_count": 68, + "execution_count": 13, "id": "25759b7b-8deb-4163-b75d-a1357c9fe88f", "metadata": { "scrolled": false @@ -123,67 +126,105 @@ "====================================================================================================\n", "Layer (type:depth-idx) Output Shape Param #\n", "====================================================================================================\n", - "ConvNext [2, 128, 7, 128] --\n", - "├─Conv2d: 1-1 [2, 16, 56, 1024] 800\n", - "├─ModuleList: 1-2 -- --\n", - "│ └─ModuleList: 2-1 -- --\n", - "│ │ └─ConvNextBlock: 3-1 [2, 16, 56, 1024] --\n", - "│ │ │ └─Conv2d: 4-1 [2, 16, 56, 1024] 800\n", - "│ │ │ └─Sequential: 4-2 [2, 16, 56, 1024] 9,280\n", - "│ │ │ └─Identity: 4-3 [2, 16, 56, 1024] --\n", - "│ │ └─ModuleList: 3-2 -- --\n", - "│ │ │ └─ConvNextBlock: 4-4 [2, 16, 56, 1024] 10,080\n", - "│ │ │ └─ConvNextBlock: 4-5 [2, 16, 56, 1024] 10,080\n", - "│ │ │ └─ConvNextBlock: 4-6 [2, 16, 56, 1024] 10,080\n", - "│ │ └─Downsample: 3-3 [2, 32, 28, 512] --\n", - "│ │ │ └─Sequential: 4-7 [2, 32, 28, 512] 2,080\n", - "│ └─ModuleList: 2-2 -- --\n", - "│ │ └─ConvNextBlock: 3-4 [2, 32, 28, 512] --\n", - "│ │ │ └─Conv2d: 4-8 [2, 32, 28, 512] 1,600\n", - "│ │ │ └─Sequential: 4-9 [2, 32, 28, 512] 36,992\n", - "│ │ │ └─Identity: 4-10 [2, 32, 28, 512] --\n", - "│ │ └─ModuleList: 3-5 -- --\n", - "│ │ │ └─ConvNextBlock: 4-11 [2, 32, 28, 512] 38,592\n", - "│ │ │ └─ConvNextBlock: 4-12 [2, 32, 28, 512] 38,592\n", - "│ │ │ └─ConvNextBlock: 4-13 [2, 32, 28, 512] 38,592\n", - "│ │ └─Downsample: 3-6 [2, 64, 14, 256] --\n", - "│ │ │ └─Sequential: 4-14 [2, 64, 14, 256] 8,256\n", - "│ └─ModuleList: 2-3 -- --\n", - "│ │ └─ConvNextBlock: 3-7 [2, 64, 14, 256] --\n", - "│ │ │ └─Conv2d: 4-15 [2, 64, 14, 256] 3,200\n", - "│ │ │ └─Sequential: 4-16 [2, 64, 14, 256] 147,712\n", - "│ │ │ └─Identity: 4-17 [2, 64, 14, 256] --\n", - "│ │ └─ModuleList: 3-8 -- --\n", - "│ │ │ └─ConvNextBlock: 4-18 [2, 64, 14, 256] 150,912\n", - "│ │ │ └─ConvNextBlock: 4-19 [2, 64, 14, 256] 150,912\n", - "│ │ │ └─ConvNextBlock: 4-20 [2, 64, 14, 256] 150,912\n", - "│ │ │ └─ConvNextBlock: 4-21 [2, 64, 14, 256] 150,912\n", - "│ │ │ └─ConvNextBlock: 4-22 [2, 64, 14, 256] 150,912\n", - "│ │ │ └─ConvNextBlock: 4-23 [2, 64, 14, 256] 150,912\n", - "│ │ └─Downsample: 3-9 [2, 128, 7, 128] --\n", - "│ │ │ └─Sequential: 4-24 [2, 128, 7, 128] 32,896\n", - "├─LayerNorm: 1-3 [2, 128, 7, 128] 128\n", + "ConvNext [2, 128, 18, 80] 5,969,376\n", + "├─Conv2d: 1-1 [2, 16, 576, 640] 800\n", + "├─ModuleList: 1 -- --\n", + "│ └─ModuleList: 2 -- --\n", + "│ │ └─ConvNextBlock: 3-1 [2, 16, 576, 640] --\n", + "│ │ │ └─Conv2d: 4-1 [2, 16, 576, 640] 800\n", + "│ │ │ └─Sequential: 4-2 [2, 16, 576, 640] 9,280\n", + "│ │ │ └─Identity: 4-3 [2, 16, 576, 640] --\n", + "│ │ └─ModuleList: 3 -- --\n", + "│ │ │ └─ConvNextBlock: 4-4 [2, 16, 576, 640] 10,080\n", + "│ │ │ └─ConvNextBlock: 4-5 [2, 16, 576, 640] 10,080\n", + "│ │ │ └─ConvNextBlock: 4-6 [2, 16, 576, 640] 10,080\n", + "│ │ └─Downsample: 3-2 [2, 32, 288, 320] --\n", + "│ │ │ └─Sequential: 4-7 [2, 32, 288, 320] 2,080\n", + "│ └─ModuleList: 2 -- --\n", + "│ │ └─ConvNextBlock: 3-3 [2, 32, 288, 320] --\n", + "│ │ │ └─Conv2d: 4-8 [2, 32, 288, 320] 1,600\n", + "│ │ │ └─Sequential: 4-9 [2, 32, 288, 320] 36,992\n", + "│ │ │ └─Identity: 4-10 [2, 32, 288, 320] --\n", + "│ │ └─ModuleList: 3 -- --\n", + "│ │ │ └─ConvNextBlock: 4-11 [2, 32, 288, 320] 38,592\n", + "│ │ │ └─ConvNextBlock: 4-12 [2, 32, 288, 320] 38,592\n", + "│ │ │ └─ConvNextBlock: 4-13 [2, 32, 288, 320] 38,592\n", + "│ │ └─Downsample: 3-4 [2, 64, 144, 160] --\n", + "│ │ │ └─Sequential: 4-14 [2, 64, 144, 160] 8,256\n", + "│ └─ModuleList: 2 -- --\n", + "│ │ └─ConvNextBlock: 3-5 [2, 64, 144, 160] --\n", + "│ │ │ └─Conv2d: 4-15 [2, 64, 144, 160] 3,200\n", + "│ │ │ └─Sequential: 4-16 [2, 64, 144, 160] 147,712\n", + "│ │ │ └─Identity: 4-17 [2, 64, 144, 160] --\n", + "│ │ └─ModuleList: 3 -- --\n", + "│ │ │ └─ConvNextBlock: 4-18 [2, 64, 144, 160] 150,912\n", + "│ │ │ └─ConvNextBlock: 4-19 [2, 64, 144, 160] 150,912\n", + "│ │ │ └─ConvNextBlock: 4-20 [2, 64, 144, 160] 150,912\n", + "│ │ └─Downsample: 3-6 [2, 128, 72, 80] --\n", + "│ │ │ └─Sequential: 4-21 [2, 128, 72, 80] 32,896\n", + "│ └─ModuleList: 2 -- --\n", + "│ │ └─ConvNextBlock: 3-7 [2, 128, 72, 80] --\n", + "│ │ │ └─Conv2d: 4-22 [2, 128, 72, 80] 6,400\n", + "│ │ │ └─Sequential: 4-23 [2, 128, 72, 80] 590,336\n", + "│ │ │ └─Identity: 4-24 [2, 128, 72, 80] --\n", + "│ │ └─ModuleList: 3 -- --\n", + "│ │ │ └─ConvNextBlock: 4-25 [2, 128, 72, 80] 596,736\n", + "│ │ │ └─ConvNextBlock: 4-26 [2, 128, 72, 80] 596,736\n", + "│ │ │ └─ConvNextBlock: 4-27 [2, 128, 72, 80] 596,736\n", + "│ │ └─Downsample: 3-8 [2, 128, 36, 80] --\n", + "│ │ │ └─Sequential: 4-28 [2, 128, 36, 80] 32,896\n", + "│ └─ModuleList: 2 -- --\n", + "│ │ └─ConvNextBlock: 3-9 [2, 128, 36, 80] --\n", + "│ │ │ └─Conv2d: 4-29 [2, 128, 36, 80] 6,400\n", + "│ │ │ └─Sequential: 4-30 [2, 128, 36, 80] 590,336\n", + "│ │ │ └─Identity: 4-31 [2, 128, 36, 80] --\n", + "│ │ └─ModuleList: 3 -- --\n", + "│ │ │ └─ConvNextBlock: 4-32 [2, 128, 36, 80] 596,736\n", + "│ │ │ └─ConvNextBlock: 4-33 [2, 128, 36, 80] 596,736\n", + "│ │ │ └─ConvNextBlock: 4-34 [2, 128, 36, 80] 596,736\n", + "│ │ │ └─ConvNextBlock: 4-35 [2, 128, 36, 80] 596,736\n", + "│ │ │ └─ConvNextBlock: 4-36 [2, 128, 36, 80] 596,736\n", + "│ │ │ └─ConvNextBlock: 4-37 [2, 128, 36, 80] 596,736\n", + "│ │ └─Downsample: 3-10 [2, 128, 18, 80] --\n", + "│ │ │ └─Sequential: 4-38 [2, 128, 18, 80] 32,896\n", + "├─TransformerBlock: 1-2 [2, 128, 18, 80] --\n", + "│ └─Attention: 2-1 [2, 128, 18, 80] --\n", + "│ │ └─LayerNorm: 3-11 [2, 128, 18, 80] 128\n", + "│ │ └─Conv2d: 3-12 [2, 768, 18, 80] 98,304\n", + "│ │ └─Conv2d: 3-13 [2, 128, 18, 80] 32,768\n", + "│ └─FeedForward: 2-2 [2, 128, 18, 80] --\n", + "│ │ └─Residual: 3-14 [2, 128, 18, 80] --\n", + "│ │ │ └─Sequential: 4-39 [2, 128, 18, 80] 131,712\n", + "├─LayerNorm: 1-3 [2, 128, 18, 80] 128\n", "====================================================================================================\n", - "Total params: 1,295,232\n", - "Trainable params: 1,295,232\n", + "Total params: 7,735,296\n", + "Trainable params: 7,735,296\n", "Non-trainable params: 0\n", - "Total mult-adds (G): 16.88\n", + "Total mult-adds (G): 140.23\n", "====================================================================================================\n", - "Input size (MB): 0.46\n", - "Forward/backward pass size (MB): 598.21\n", - "Params size (MB): 5.18\n", - "Estimated Total Size (MB): 603.85\n", + "Input size (MB): 2.95\n", + "Forward/backward pass size (MB): 3987.21\n", + "Params size (MB): 30.94\n", + "Estimated Total Size (MB): 4021.10\n", "====================================================================================================" ] }, - "execution_count": 68, + "execution_count": 13, "metadata": {}, "output_type": "execute_result" } ], "source": [ - "summary(net, (2, 1, 56, 1024), device=\"cpu\", depth=4)" + "summary(net, (2, 1, 576, 640), device=\"cpu\", depth=4)" ] + }, + { + "cell_type": "code", + "execution_count": null, + "id": "7a7e27f4", + "metadata": {}, + "outputs": [], + "source": [] } ], "metadata": { |