Spaces:
Running
Running
File size: 6,435 Bytes
ac8579b | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 | {
"cells": [
{
"cell_type": "markdown",
"metadata": {},
"source": [
"## Annotate Comics/Manga\n",
"Download comictextdetector.pt and put it into data directory.\n",
"Run next block to generate following annotations for data\\examples\\AisazuNihaIrarenai-003.jpg:\n",
"- AisazuNihaIrarenai-003.txt: yolo format bounding boxes of english&japanese text block bounding boxes. 0 is eng.\n",
"- mask-AisazuNihaIrarenai-003.png\n",
"- line-AisazuNihaIrarenai-003.txt: icdar format bboxes of text lines."
]
},
{
"cell_type": "code",
"execution_count": 1,
"metadata": {},
"outputs": [
{
"name": "stderr",
"output_type": "stream",
"text": [
"100%|██████████| 1/1 [00:04<00:00, 4.78s/it]\n"
]
}
],
"source": [
"from inference import model2annotations\n",
"\n",
"img_dir = r'data/examples'\n",
"model_path = r'data/comictextdetector.pt'\n",
"img_dir = r'data/examples' # can be dir list\n",
"save_dir = r'data/examples/annotations'\n",
"model2annotations(model_path, img_dir, save_dir, save_json=False)"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"## Generate synthetic data\n",
"- current rendering script won't handle characters missing from fonts.\n",
"- Please use no-text images."
]
},
{
"cell_type": "code",
"execution_count": 1,
"metadata": {},
"outputs": [
{
"name": "stderr",
"output_type": "stream",
"text": [
"100%|██████████| 10/10 [00:12<00:00, 1.23s/it]\n"
]
}
],
"source": [
"from text_rendering import ComicTextSampler, render_comictext, ALIGN_LEFT, ALIGN_CENTER\n",
"import copy\n",
"\n",
"ja_sampler_dict = {\n",
" 'num_txtblk': 20,\n",
" 'font': {\n",
" 'font_dir': 'data/examples/fonts', # font file directory\n",
" 'font_statics': 'data/font_statics_en.csv', # Just a font list file, please create your own list and ignore the last two cols.\n",
" 'num': 1200, # first 500 of the fontlist will be used \n",
"\n",
" # params to mimic comic/manga text style\n",
" 'size': {'value': [0.02, 0.03, 0.15],\n",
" 'prob': [1, 0.4, 0.15]},\n",
" 'stroke_width': {'value': [0, 0.1, 0.15],\n",
" 'prob': [1, 0.5, 0.2]},\n",
" 'color': {'value': ['black', 'white', 'random'],\n",
" 'prob': [1, 1, 0.4]},\n",
" },\n",
" 'text': {\n",
" 'lang': 'ja', # render japanese, 'en' for english\n",
" 'orientation': {'value': [1, 0], # 1 is vertical text.\n",
" 'prob': [1, 0.3]},\n",
" 'rotation': {'value': [0, 30, 60],\n",
" 'prob': [1, 0.3, 0.1]},\n",
" 'num_lines': {'value': [0.15],\n",
" 'prob': [1]}, \n",
" 'length': {'value': [0.3],\n",
" 'prob': [1]},\n",
" 'min_num_lines': 1,\n",
" 'min_length': 3,\n",
" 'alignment': {'value': [ALIGN_LEFT, ALIGN_CENTER],\n",
" 'prob': [0.3, 1]}\n",
" }\n",
" }\n",
"\n",
"jp_cts = ComicTextSampler((845, 1280), ja_sampler_dict, seed=0)\n",
"eng_dict = copy.deepcopy(ja_sampler_dict)\n",
"eng_dict['text']['lang'] = 'en'\n",
"eng_dict['text']['orientation'] = {'value': [1, 0],\n",
" 'prob': [0, 1]}\n",
"eng_cts = ComicTextSampler((845, 1280), eng_dict, seed=0)\n",
"\n",
"img_dir = r'data/examples'\n",
"save_dir = r'data/examples/annotations'\n",
" \n",
"render_comictext([eng_cts, jp_cts], img_dir, save_dir=save_dir, save_prefix=None, render_num=10, label_dir=None, show=False)"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"## Training\n",
"### Train Text Block Detector\n",
"Train yolov5s using official repo of yolov5, assume the trained model is 'yolov5sblk.pt', go to the root directory of yolov5 and run following code.\n",
"\n",
"``` python\n",
"import torch\n",
"m = torch.load('yolov5sblk.pt')['model']\n",
"save_dict = {\n",
" 'cfg': m.yaml,\n",
" 'weights': m.state_dict()\n",
"}\n",
"torch.save(save_dict, 'yolov5sblk.ckpt')\n",
"```\n",
"### Train Text Segmentation Head\n",
"1. Put yolov5sblk.ckpt into data. \n",
"2. Refer to train_seg.py for further details. \n",
"\n",
"### Train DBHead\n",
"Please refer to train_db.py.\n"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"## Concat weights & export as onnx"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
"from utils.export import *\n",
"concate_models('data/yolov5sblk.ckpt', 'data/unet_best.ckpt', 'data/db_best.ckpt', 'data/textdetector.pt')\n",
"\n",
"batch_size, imgsz = 1, 1024\n",
"cuda = torch.cuda.is_available()\n",
"device = 'cpu'\n",
"im = torch.zeros(batch_size, 3, imgsz, imgsz).to(device)\n",
"model_path = r'data/textdetector.pt'\n",
"model = TextDetBase(model_path, device=device).to(device)\n",
"export_onnx(model, im, model_path, 11)"
]
}
],
"metadata": {
"interpreter": {
"hash": "545b34d9a5e72e2b90b819a16ec22002dd3dc9d66aaf1029c3177c6408a5603b"
},
"kernelspec": {
"display_name": "Python 3.9.7 64-bit",
"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"
},
"orig_nbformat": 4
},
"nbformat": 4,
"nbformat_minor": 2
}
|