yomitoku 0.4.1__py3-none-any.whl → 0.5.1__py3-none-any.whl

yomitoku/configs/cfg_text_detector_dbnet.py

@@ -30,7 +30,7 @@ class PostProcess:
     thresh: float = 0.2
     box_thresh: float = 0.5
     max_candidates: int = 1500
-    unclip_ratio: float = 2.0
+    unclip_ratio: float = 7.0
 
 
 @dataclass

yomitoku/data/functions.py

@@ -3,7 +3,7 @@ from pathlib import Path
 import cv2
 import numpy as np
 import torch
-from pdf2image import convert_from_path
+import pypdfium2
 
 from ..constants import (
     MIN_IMAGE_SIZE,
@@ -70,6 +70,7 @@ def load_pdf(pdf_path: str, dpi=200) -> list[np.ndarray]:
     Returns:
         list[np.ndarray]: list of image data(BGR)
     """
+
     pdf_path = Path(pdf_path)
     if not pdf_path.exists():
         raise FileNotFoundError(f"File not found: {pdf_path}")
@@ -86,11 +87,19 @@ def load_pdf(pdf_path: str, dpi=200) -> list[np.ndarray]:
         )
 
     try:
-        images = convert_from_path(pdf_path, dpi=dpi)
+        doc = pypdfium2.PdfDocument(pdf_path)
+        renderer = doc.render(
+            pypdfium2.PdfBitmap.to_pil,
+            scale=dpi / 72,
+        )
+        images = list(renderer)
+        images = [np.array(image.convert("RGB"))[:, :, ::-1] for image in images]
+
+        doc.close()
     except Exception as e:
         raise ValueError(f"Failed to open the PDF file: {pdf_path}") from e
 
-    return [np.array(img)[:, :, ::-1] for img in images]
+    return images
 
 
 def resize_shortest_edge(
@@ -193,9 +202,7 @@ def validate_quads(img: np.ndarray, quads: list[list[list[int]]]):
         h, w = img.shape[:2]
 
         if x1 < 0 or x2 > w or y1 < 0 or y2 > h:
-            raise ValueError(
-                f"The vertices are out of the image. {quad.tolist()}"
-            )
+            raise ValueError(f"The vertices are out of the image. {quad.tolist()}")
 
     return True
 

yomitoku/document_analyzer.py

@@ -109,8 +109,8 @@ def extract_words_within_element(pred_words, element):
     if len(contained_words) == 0:
         return None, None, check_list
 
-    mean_width = word_sum_width / len(contained_words)
-    mean_height = word_sum_height / len(contained_words)
+    # mean_width = word_sum_width / len(contained_words)
+    # mean_height = word_sum_height / len(contained_words)
 
     word_direction = [word.direction for word in contained_words]
     cnt_horizontal = word_direction.count("horizontal")
@@ -120,18 +120,12 @@ def extract_words_within_element(pred_words, element):
     if element_direction == "horizontal":
         contained_words = sorted(
             contained_words,
-            key=lambda x: (
-                x.points[0][1] // int(mean_height),
-                x.points[0][0],
-            ),
+            key=lambda x: (sum([p[1] for p in x.points]) / 4),
         )
     else:
         contained_words = sorted(
             contained_words,
-            key=lambda x: (
-                x.points[1][0] // int(mean_width),
-                x.points[1][1],
-            ),
+            key=lambda x: (sum([p[0] for p in x.points]) / 4),
             reverse=True,
         )
 

yomitoku/layout_analyzer.py

@@ -27,8 +27,7 @@ class LayoutAnalyzer:
 
         if isinstance(configs, dict):
             assert (
-                "layout_parser" in configs
-                or "table_structure_recognizer" in configs
+                "layout_parser" in configs or "table_structure_recognizer" in configs
             ), "Invalid config key. Please check the config keys."
 
             if "layout_parser" in configs:
@@ -53,9 +52,7 @@ class LayoutAnalyzer:
     def __call__(self, img):
         layout_results, vis = self.layout_parser(img)
         table_boxes = [table.box for table in layout_results.tables]
-        table_results, vis = self.table_structure_recognizer(
-            img, table_boxes, vis=vis
-        )
+        table_results, vis = self.table_structure_recognizer(img, table_boxes, vis=vis)
 
         results = LayoutAnalyzerSchema(
             paragraphs=layout_results.paragraphs,

yomitoku/models/dbnet_plus.py

@@ -20,9 +20,7 @@ class BackboneBase(nn.Module):
             "layer4": "layer4",
         }
 
-        self.body = IntermediateLayerGetter(
-            backbone, return_layers=return_layers
-        )
+        self.body = IntermediateLayerGetter(backbone, return_layers=return_layers)
 
     def forward(self, tensor):
         xs = self.body(tensor)
@@ -57,18 +55,10 @@ class DBNetDecoder(nn.Module):
         self.training = True
         self.input_proj = nn.ModuleDict(
             {
-                "layer1": nn.Conv2d(
-                    in_channels[0], self.d_model, 1, bias=False
-                ),
-                "layer2": nn.Conv2d(
-                    in_channels[1], self.d_model, 1, bias=False
-                ),
-                "layer3": nn.Conv2d(
-                    in_channels[2], self.d_model, 1, bias=False
-                ),
-                "layer4": nn.Conv2d(
-                    in_channels[3], self.d_model, 1, bias=False
-                ),
+                "layer1": nn.Conv2d(in_channels[0], self.d_model, 1, bias=False),
+                "layer2": nn.Conv2d(in_channels[1], self.d_model, 1, bias=False),
+                "layer3": nn.Conv2d(in_channels[2], self.d_model, 1, bias=False),
+                "layer4": nn.Conv2d(in_channels[3], self.d_model, 1, bias=False),
             }
         )
 
@@ -89,9 +79,7 @@ class DBNetDecoder(nn.Module):
                         padding=1,
                         bias=False,
                     ),
-                    nn.Upsample(
-                        scale_factor=2, mode="bilinear", align_corners=False
-                    ),
+                    nn.Upsample(scale_factor=2, mode="bilinear", align_corners=False),
                 ),
                 "layer3": nn.Sequential(
                     nn.Conv2d(
@@ -101,9 +89,7 @@ class DBNetDecoder(nn.Module):
                         padding=1,
                         bias=False,
                     ),
-                    nn.Upsample(
-                        scale_factor=4, mode="bilinear", align_corners=False
-                    ),
+                    nn.Upsample(scale_factor=4, mode="bilinear", align_corners=False),
                 ),
                 "layer4": nn.Sequential(
                     nn.Conv2d(
@@ -113,17 +99,13 @@ class DBNetDecoder(nn.Module):
                         padding=1,
                         bias=False,
                     ),
-                    nn.Upsample(
-                        scale_factor=4, mode="bilinear", align_corners=False
-                    ),
+                    nn.Upsample(scale_factor=4, mode="bilinear", align_corners=False),
                 ),
             }
         )
 
         self.binarize = nn.Sequential(
-            nn.Conv2d(
-                self.d_model, self.d_model // 4, 3, padding=1, bias=False
-            ),
+            nn.Conv2d(self.d_model, self.d_model // 4, 3, padding=1, bias=False),
             nn.BatchNorm2d(self.d_model // 4),
             nn.ReLU(inplace=True),
             nn.ConvTranspose2d(self.d_model // 4, self.d_model // 4, 2, 2),
@@ -166,16 +148,12 @@ class DBNetDecoder(nn.Module):
             m.weight.data.fill_(1.0)
             m.bias.data.fill_(1e-4)
 
-    def _init_thresh(
-        self, inner_channels, serial=False, smooth=False, bias=False
-    ):
+    def _init_thresh(self, inner_channels, serial=False, smooth=False, bias=False):
         in_channels = inner_channels
         if serial:
             in_channels += 1
         self.thresh = nn.Sequential(
-            nn.Conv2d(
-                in_channels, inner_channels // 4, 3, padding=1, bias=bias
-            ),
+            nn.Conv2d(in_channels, inner_channels // 4, 3, padding=1, bias=bias),
             nn.BatchNorm2d(inner_channels // 4),
             nn.ReLU(inplace=True),
             self._init_upsample(
@@ -186,16 +164,12 @@ class DBNetDecoder(nn.Module):
             ),
             nn.BatchNorm2d(inner_channels // 4),
             nn.ReLU(inplace=True),
-            self._init_upsample(
-                inner_channels // 4, 1, smooth=smooth, bias=bias
-            ),
+            self._init_upsample(inner_channels // 4, 1, smooth=smooth, bias=bias),
             nn.Sigmoid(),
         )
         return self.thresh
 
-    def _init_upsample(
-        self, in_channels, out_channels, smooth=False, bias=False
-    ):
+    def _init_upsample(self, in_channels, out_channels, smooth=False, bias=False):
         if smooth:
             inter_out_channels = out_channels
             if out_channels == 1:

yomitoku/models/layers/rtdetr_backbone.py

@@ -1,5 +1,4 @@
-"""Copyright(c) 2023 lyuwenyu. All Rights Reserved.
-"""
+"""Copyright(c) 2023 lyuwenyu. All Rights Reserved."""
 
 from collections import OrderedDict
 
@@ -48,9 +47,7 @@ class ConvNormLayer(nn.Module):
 class BasicBlock(nn.Module):
     expansion = 1
 
-    def __init__(
-        self, ch_in, ch_out, stride, shortcut, act="relu", variant="b"
-    ):
+    def __init__(self, ch_in, ch_out, stride, shortcut, act="relu", variant="b"):
         super().__init__()
 
         self.shortcut = shortcut
@@ -89,9 +86,7 @@ class BasicBlock(nn.Module):
 class BottleNeck(nn.Module):
     expansion = 4
 
-    def __init__(
-        self, ch_in, ch_out, stride, shortcut, act="relu", variant="b"
-    ):
+    def __init__(self, ch_in, ch_out, stride, shortcut, act="relu", variant="b"):
         super().__init__()
 
         if variant == "a":
@@ -114,17 +109,13 @@ class BottleNeck(nn.Module):
                             ("pool", nn.AvgPool2d(2, 2, 0, ceil_mode=True)),
                             (
                                 "conv",
-                                ConvNormLayer(
-                                    ch_in, ch_out * self.expansion, 1, 1
-                                ),
+                                ConvNormLayer(ch_in, ch_out * self.expansion, 1, 1),
                             ),
                         ]
                     )
                 )
             else:
-                self.short = ConvNormLayer(
-                    ch_in, ch_out * self.expansion, 1, stride
-                )
+                self.short = ConvNormLayer(ch_in, ch_out * self.expansion, 1, stride)
 
         self.act = nn.Identity() if act is None else get_activation(act)
 
@@ -145,9 +136,7 @@ class BottleNeck(nn.Module):
 
 
 class Blocks(nn.Module):
-    def __init__(
-        self, block, ch_in, ch_out, count, stage_num, act="relu", variant="b"
-    ):
+    def __init__(self, block, ch_in, ch_out, count, stage_num, act="relu", variant="b"):
         super().__init__()
 
         self.blocks = nn.ModuleList()

yomitoku/models/layers/rtdetr_hybrid_encoder.py

@@ -1,5 +1,4 @@
-"""Copyright(c) 2023 lyuwenyu. All Rights Reserved.
-"""
+"""Copyright(c) 2023 lyuwenyu. All Rights Reserved."""
 
 import copy
 from collections import OrderedDict
@@ -241,9 +240,7 @@ class HybridEncoder(nn.Module):
         for in_channel in in_channels:
             if version == "v1":
                 proj = nn.Sequential(
-                    nn.Conv2d(
-                        in_channel, hidden_dim, kernel_size=1, bias=False
-                    ),
+                    nn.Conv2d(in_channel, hidden_dim, kernel_size=1, bias=False),
                     nn.BatchNorm2d(hidden_dim),
                 )
             elif version == "v2":
@@ -279,9 +276,7 @@ class HybridEncoder(nn.Module):
 
         self.encoder = nn.ModuleList(
             [
-                TransformerEncoder(
-                    copy.deepcopy(encoder_layer), num_encoder_layers
-                )
+                TransformerEncoder(copy.deepcopy(encoder_layer), num_encoder_layers)
                 for _ in range(len(use_encoder_idx))
             ]
         )
@@ -336,9 +331,7 @@ class HybridEncoder(nn.Module):
                 # self.register_buffer(f'pos_embed{idx}', pos_embed)
 
     @staticmethod
-    def build_2d_sincos_position_embedding(
-        w, h, embed_dim=256, temperature=10000.0
-    ):
+    def build_2d_sincos_position_embedding(w, h, embed_dim=256, temperature=10000.0):
         """ """
         grid_w = torch.arange(int(w), dtype=torch.float32)
         grid_h = torch.arange(int(h), dtype=torch.float32)
@@ -376,9 +369,7 @@ class HybridEncoder(nn.Module):
                         src_flatten.device
                     )
 
-                memory: torch.Tensor = self.encoder[i](
-                    src_flatten, pos_embed=pos_embed
-                )
+                memory: torch.Tensor = self.encoder[i](src_flatten, pos_embed=pos_embed)
                 proj_feats[enc_ind] = (
                     memory.permute(0, 2, 1)
                     .reshape(-1, self.hidden_dim, h, w)
@@ -390,13 +381,9 @@ class HybridEncoder(nn.Module):
         for idx in range(len(self.in_channels) - 1, 0, -1):
             feat_heigh = inner_outs[0]
             feat_low = proj_feats[idx - 1]
-            feat_heigh = self.lateral_convs[len(self.in_channels) - 1 - idx](
-                feat_heigh
-            )
+            feat_heigh = self.lateral_convs[len(self.in_channels) - 1 - idx](feat_heigh)
             inner_outs[0] = feat_heigh
-            upsample_feat = F.interpolate(
-                feat_heigh, scale_factor=2.0, mode="nearest"
-            )
+            upsample_feat = F.interpolate(feat_heigh, scale_factor=2.0, mode="nearest")
             inner_out = self.fpn_blocks[len(self.in_channels) - 1 - idx](
                 torch.concat([upsample_feat, feat_low], dim=1)
             )

yomitoku/models/parseq.py

@@ -26,9 +26,7 @@ from ..postprocessor import ParseqTokenizer as Tokenizer
 from .layers.parseq_transformer import Decoder, Encoder, TokenEmbedding
 
 
-def init_weights(
-    module: nn.Module, name: str = "", exclude: Sequence[str] = ()
-):
+def init_weights(module: nn.Module, name: str = "", exclude: Sequence[str] = ()):
     """Initialize the weights using the typical initialization schemes used in SOTA models."""
     if any(map(name.startswith, exclude)):
         return
@@ -41,9 +39,7 @@ def init_weights(
         if module.padding_idx is not None:
             module.weight.data[module.padding_idx].zero_()
     elif isinstance(module, nn.Conv2d):
-        nn.init.kaiming_normal_(
-            module.weight, mode="fan_out", nonlinearity="relu"
-        )
+        nn.init.kaiming_normal_(module.weight, mode="fan_out", nonlinearity="relu")
         if module.bias is not None:
             nn.init.zeros_(module.bias)
     elif isinstance(module, (nn.LayerNorm, nn.BatchNorm2d, nn.GroupNorm)):
@@ -93,9 +89,7 @@ class PARSeq(nn.Module, PyTorchModelHubMixin):
     @torch.jit.ignore
     def no_weight_decay(self):
         param_names = {"text_embed.embedding.weight", "pos_queries"}
-        enc_param_names = {
-            "encoder." + n for n in self.encoder.no_weight_decay()
-        }
+        enc_param_names = {"encoder." + n for n in self.encoder.no_weight_decay()}
         return param_names.union(enc_param_names)
 
     def encode(self, img: torch.Tensor):
@@ -149,9 +143,7 @@ class PARSeq(nn.Module, PyTorchModelHubMixin):
 
         # Special case for the forward permutation. Faster than using `generate_attn_masks()`
         tgt_mask = query_mask = torch.triu(
-            torch.ones(
-                (num_steps, num_steps), dtype=torch.bool, device=self._device
-            ),
+            torch.ones((num_steps, num_steps), dtype=torch.bool, device=self._device),
             1,
         )
 
@@ -185,10 +177,7 @@ class PARSeq(nn.Module, PyTorchModelHubMixin):
                     # greedy decode. add the next token index to the target input
                     tgt_in[:, j] = p_i.squeeze().argmax(-1)
                     # Efficient batch decoding: If all output words have at least one EOS token, end decoding.
-                    if (
-                        testing
-                        and (tgt_in == tokenizer.eos_id).any(dim=-1).all()
-                    ):
+                    if testing and (tgt_in == tokenizer.eos_id).any(dim=-1).all():
                         break
 
             logits = torch.cat(logits, dim=1)
@@ -227,9 +216,7 @@ class PARSeq(nn.Module, PyTorchModelHubMixin):
                 # Prior context is the previous output.
                 tgt_in = torch.cat([bos, logits[:, :-1].argmax(-1)], dim=1)
                 # Mask tokens beyond the first EOS token.
-                tgt_padding_mask = (tgt_in == tokenizer.eos_id).int().cumsum(
-                    -1
-                ) > 0
+                tgt_padding_mask = (tgt_in == tokenizer.eos_id).int().cumsum(-1) > 0
                 tgt_out = self.decode(
                     tgt_in,
                     memory,

yomitoku/postprocessor/dbnet_postporcessor.py

@@ -1,13 +1,12 @@
 import cv2
+import math
 import numpy as np
 import pyclipper
 from shapely.geometry import Polygon
 
 
 class DBnetPostProcessor:
-    def __init__(
-        self, min_size, thresh, box_thresh, max_candidates, unclip_ratio
-    ):
+    def __init__(self, min_size, thresh, box_thresh, max_candidates, unclip_ratio):
         self.min_size = min_size
         self.thresh = thresh
         self.box_thresh = box_thresh
@@ -24,9 +23,7 @@ class DBnetPostProcessor:
         pred = preds["binary"][0]
         segmentation = self.binarize(pred)[0]
         height, width = image_size
-        quads, scores = self.boxes_from_bitmap(
-            pred, segmentation, width, height
-        )
+        quads, scores = self.boxes_from_bitmap(pred, segmentation, width, height)
         return quads, scores
 
     def binarize(self, pred):
@@ -65,9 +62,7 @@ class DBnetPostProcessor:
             if self.box_thresh > score:
                 continue
 
-            box = self.unclip(points, unclip_ratio=self.unclip_ratio).reshape(
-                -1, 1, 2
-            )
+            box = self.unclip(points, unclip_ratio=self.unclip_ratio).reshape(-1, 1, 2)
             box, sside = self.get_mini_boxes(box)
             if sside < self.min_size + 2:
                 continue
@@ -76,9 +71,7 @@ class DBnetPostProcessor:
                 dest_width = dest_width.item()
                 dest_height = dest_height.item()
 
-            box[:, 0] = np.clip(
-                np.round(box[:, 0] / width * dest_width), 0, dest_width
-            )
+            box[:, 0] = np.clip(np.round(box[:, 0] / width * dest_width), 0, dest_width)
             box[:, 1] = np.clip(
                 np.round(box[:, 1] / height * dest_height), 0, dest_height
             )
@@ -88,9 +81,17 @@ class DBnetPostProcessor:
 
         return boxes, scores
 
-    def unclip(self, box, unclip_ratio=1.5):
+    def unclip(self, box, unclip_ratio=7):
+        # 小さい文字が見切れやすい、大きい文字のマージンが過度に大きくなる等の課題がある
+        # 対応として、文字の大きさに応じて、拡大パラメータを動的に変更する
+        # Note: こののルールはヒューリスティックで理論的根拠はない
         poly = Polygon(box)
-        distance = poly.area * unclip_ratio / poly.length
+        width = box[:, 0].max() - box[:, 0].min()
+        height = box[:, 1].max() - box[:, 1].min()
+        box_dist = min(width, height)
+        ratio = unclip_ratio / math.sqrt(box_dist)
+
+        distance = poly.area * ratio / poly.length
         offset = pyclipper.PyclipperOffset()
         offset.AddPath(box, pyclipper.JT_ROUND, pyclipper.ET_CLOSEDPOLYGON)
         expanded = np.array(offset.Execute(distance))

yomitoku/postprocessor/parseq_tokenizer.py

@@ -122,7 +122,5 @@ class ParseqTokenizer(BaseTokenizer):
             eos_idx = len(ids)  # Nothing to truncate.
         # Truncate after EOS
         ids = ids[:eos_idx]
-        probs = probs[
-            : eos_idx + 1
-        ]  # but include prob. for EOS (if it exists)
+        probs = probs[: eos_idx + 1]  # but include prob. for EOS (if it exists)
         return probs, ids

yomitoku/utils/misc.py

@@ -1,5 +1,5 @@
 def load_charset(charset_path):
-    with open(charset_path, "r") as f:
+    with open(charset_path, "r", encoding="utf-8") as f:
         charset = f.read()
     return charset
 

yomitoku/utils/visualizer.py

@@ -1,8 +1,10 @@
 import cv2
 import numpy as np
-from PIL import Image, ImageDraw, ImageFont
-
+from PIL import Image, ImageDraw, ImageFont, features
 from ..constants import PALETTE
+from .logger import set_logger
+
+logger = set_logger(__name__, "INFO")
 
 
 def _reading_order_visualizer(img, elements, line_color, tip_size):
@@ -148,13 +150,18 @@ def rec_visualizer(
     out = img.copy()
     pillow_img = Image.fromarray(out)
     draw = ImageDraw.Draw(pillow_img)
+    has_raqm = features.check_feature(feature="raqm")
+    if not has_raqm:
+        logger.warning(
+            "libraqm is not installed. Vertical text rendering is not supported. Rendering horizontally instead."
+        )
 
     for pred, quad, direction in zip(
         outputs.contents, outputs.points, outputs.directions
     ):
         quad = np.array(quad).astype(np.int32)
         font = ImageFont.truetype(font_path, font_size)
-        if direction == "horizontal":
+        if direction == "horizontal" or not has_raqm:
             x_offset = 0
             y_offset = -font_size
 

{yomitoku-0.4.1.dist-info → yomitoku-0.5.1.dist-info}/METADATA

@@ -1,7 +1,7 @@
 Metadata-Version: 2.3
 Name: yomitoku
-Version: 0.4.1
-Summary: Yomitoku is a document image analysis package powered by AI technology for the Japanese language.
+Version: 0.5.1
+Summary: Yomitoku is an AI-powered document image analysis package designed specifically for the Japanese language.
 Author-email: Kotaro Kinoshita <kotaro.kinoshita@mlism.com>
 License: CC BY-NC-SA 4.0
 Keywords: Deep Learning,Japanese,OCR
@@ -10,24 +10,25 @@ Requires-Dist: huggingface-hub>=0.26.1
 Requires-Dist: lxml>=5.3.0
 Requires-Dist: omegaconf>=2.3.0
 Requires-Dist: opencv-python>=4.10.0.84
-Requires-Dist: pdf2image>=1.17.0
 Requires-Dist: pyclipper>=1.3.0.post6
 Requires-Dist: pydantic>=2.9.2
+Requires-Dist: pypdfium2>=4.30.0
 Requires-Dist: shapely>=2.0.6
 Requires-Dist: timm>=1.0.11
-Requires-Dist: torch>=2.5.0
+Requires-Dist: torch==2.5.0
 Requires-Dist: torchvision>=0.20.0
 Description-Content-Type: text/markdown
 
-# YomiToku
+日本語版 | [English](README_EN.md)
+
+<img src="static/logo/horizontal.png" width="800px">
 
 ![Python](https://img.shields.io/badge/Python-3.9|3.10|3.11|3.12-F9DC3E.svg?logo=python&logoColor=&style=flat)
 ![Pytorch](https://img.shields.io/badge/Pytorch-2.5-EE4C2C.svg?logo=Pytorch&style=fla)
-![OS](https://img.shields.io/badge/OS-Linux|MacOS-1793D1.svg?&style=fla)
+![CUDA](https://img.shields.io/badge/CUDA->=11.8-76B900.svg?logo=NVIDIA&style=fla)
+![OS](https://img.shields.io/badge/OS-Linux|Mac|Win-1793D1.svg?&style=fla)
 [![Document](https://img.shields.io/badge/docs-live-brightgreen)](https://kotaro-kinoshita.github.io/yomitoku-dev/)
 
-<img src="static/logo/horizontal.png" width="800px">
-
 ## 🌟 概要
 
 YomiToku は日本語に特化した AI 文章画像解析エンジン(Document AI)です。画像内の文字の全文 OCR およびレイアウト解析機能を有しており、画像内の文字情報や図表を認識、抽出、変換します。
@@ -60,31 +61,16 @@ Markdown でエクスポートした結果は関してはリポジトリ内の[s
 
 ## 📣 リリース情報
 
-- 2024 年 12 月 XX YomiToku vX.X.X を公開
+- 2024 年 11 月 26 YomiToku v0.5.1 (beta) を公開
 
 ## 💡 インストールの方法
 
 ```
-pip install git+https://github.com/kotaro-kinoshita/yomitoku-dev.git@main
+pip install yomitoku
 ```
 
-- pytorch がご自身の GPU の環境にあったものをインストールしてください
-
-### 依存ライブラリ
-
-pdf ファイルの解析を行うためには、別途、[poppler](https://poppler.freedesktop.org/)のインストールが必要です。
-
-**Mac**
-
-```
-brew install poppler
-```
-
-**Linux**
-
-```
-apt install poppler-utils -y
-```
+- pytorch はご自身の CUDAのバージョンにあったものをインストールしてください。デフォルトではCUDA12.4以上に対応したものがインストールされます。
+- pytorch は2.5以上のバージョンに対応しています。その関係でCUDA11.8以上のバージョンが必要になります。対応できない場合は、リポジトリ内のDockerfileを利用してください。
 
 ## 🚀 実行方法
 
@@ -98,8 +84,8 @@ yomitoku ${path_data} -f md -o results -v --figure
 - `-v`, `--vis` を指定すると解析結果を可視化した画像を出力します。
 - `-d`, `--device` モデルを実行するためのデバイスを指定します。gpu が利用できない場合は cpu で推論が実行されます。(デフォルト: cuda)
 - `--ignore_line_break` 画像の改行位置を無視して、段落内の文章を連結して返します。（デフォルト：画像通りの改行位置位置で改行します。）
-- `figure_letter` 検出した図表に含まれる文字も出力ファイルにエクスポートします。
-- `figure` 検出した図、画像を出力ファイルにエクスポートします。(html と markdown のみ)
+- `--figure_letter` 検出した図表に含まれる文字も出力ファイルにエクスポートします。
+- `--figure` 検出した図、画像を出力ファイルにエクスポートします。(html と markdown のみ)
 
 その他のオプションに関しては、ヘルプを参照
 
@@ -107,11 +93,10 @@ yomitoku ${path_data} -f md -o results -v --figure
 yomitoku --help
 ```
 
-### Note
-
-- CPU を用いての推論向けに最適化されておらず、処理時間が長くなりますので、GPU での実行を推奨します。
+**NOTE**
+- GPU での実行を推奨します。CPU を用いての推論向けに最適化されておらず、処理時間が長くなります。
 - 活字のみ識別をサポートしております。手書き文字に関しては、読み取れる場合もありますが、公式にはサポートしておりません。
-- OCR は文書 OCR と情景 OCR(看板など紙以外にプリントされた文字)に大別されますが、Yomitoku は文書 OCR 向けに最適化されています。
+- Yomitoku は文書 OCR 向けに最適化されており、情景 OCR(看板など紙以外にプリントされた文字の読み取り)向けには最適化されていません。
 - AI-OCR の識別精度を高めるために、入力画像の解像度が重要です。低解像度画像では識別精度が低下します。最低でも画像の短辺を 720px 以上の画像で推論することをお勧めします。
 
 ## 📝 ドキュメント
@@ -120,8 +105,8 @@ yomitoku --help
 
 ## LICENSE
 
-本リポジトリ内に格納されているリソースのライセンスは YomiToku は CC BY-NC-SA 4.0 に従います。
-非商用での個人利用、研究目的での利用は自由に利用できます。
+本リポジトリ内に格納されているソースコードおよび本プロジェクトに関連する HuggingFaceHub 上のモデルの重みファイルのライセンスは CC BY-NC-SA 4.0 に従います。
+非商用での個人利用、研究目的での利用はご自由にお使いください。
 商用目的での利用に関しては、別途、商用ライセンスを提供しますので、開発者にお問い合わせください。
 
-YomiToku © 2024 by MLism Inc. is licensed under CC BY-NC-SA 4.0. To view a copy of this license, visit https://creativecommons.org/licenses/by-nc-sa/4.0/
+YomiToku © 2024 by Kotaro Kinoshita is licensed under CC BY-NC-SA 4.0. To view a copy of this license, visit https://creativecommons.org/licenses/by-nc-sa/4.0/

{yomitoku-0.4.1.dist-info → yomitoku-0.5.1.dist-info}/RECORD

@@ -1,8 +1,8 @@
 yomitoku/__init__.py,sha256=kXOM8RbpwwLABG3p3vPT3dJWBk4JX2MFGrOeBEW0hKM,543
 yomitoku/base.py,sha256=lzR_V8t87aRasmFdFwD-8KAeSahSTI3AZaEn6g8sOv8,3871
 yomitoku/constants.py,sha256=zlW5QRc_u_F3C2RAgBFWyHJZexBnJT5N15GC-9d3iLo,686
-yomitoku/document_analyzer.py,sha256=0dTH6YrCvp2EZXufPnSN4DdF95DZ0_z1TIDML744oX0,10029
-yomitoku/layout_analyzer.py,sha256=WIP8PjuayoM7VNtmrbb1b1r4joHYuSyIHg91GZ3F46s,2071
+yomitoku/document_analyzer.py,sha256=HIg-nVzDhJIP-h-tn4uU86KakgHdlAhosEqK_i-SWe4,9906
+yomitoku/layout_analyzer.py,sha256=QTeRcVd8aySz8u6dg2ikET77ar3sqlukRLBwYfTyMPM,2033
 yomitoku/layout_parser.py,sha256=V2jCNHE61jNp8ytYdKwPV34V5qEK7y-7-Mq7-AkoQhU,5898
 yomitoku/ocr.py,sha256=Rcojw0aGA6yDF2RjqfK23_rMw-xm61KGd8JmTCTOOVU,2516
 yomitoku/reading_order.py,sha256=OfhOS9ttPDoPSuHrIRKyOzG19GGeRufbuSKDqhsohh4,6404
@@ -14,39 +14,39 @@ yomitoku/cli/main.py,sha256=MBD0S4sXgquJ8P2egkZjJcglXvCke5Uw46C28SDtr8g,6252
 yomitoku/configs/__init__.py,sha256=KBhb9S7xt22HZaIcoWSgZHfscXXj9YlimOwLH5z9CRo,454
 yomitoku/configs/cfg_layout_parser_rtdtrv2.py,sha256=8PRxB2Ar9UF7-DLtbgSokhrzdXb0veWI6Wc-X8qigRw,2329
 yomitoku/configs/cfg_table_structure_recognizer_rtdtrv2.py,sha256=o70GMHD8k-zeBeJtuhPS8x7vVB-ffucnJXeSyn-0AXo,2116
-yomitoku/configs/cfg_text_detector_dbnet.py,sha256=AUl9aStR6z7SEPldIDd7GNQVPQx0eUlyn6ui3B4RVjA,1153
+yomitoku/configs/cfg_text_detector_dbnet.py,sha256=U9k48PON7haoOaytiELhbZRpv9RMiUm6nnfHmdxIa9Q,1153
 yomitoku/configs/cfg_text_recognizer_parseq.py,sha256=hpFs3nKqh4XdU3BZMTultegtLEGahEsCaZdjfKC_MO8,1247
 yomitoku/data/__init__.py,sha256=KAofFc9rk9ZdTKBjemu9RM8Vj9XnKbWC2MPZ2RWtOdE,82
 yomitoku/data/dataset.py,sha256=-I4f-FDtgsPnJ2MnXB7FtwihMW3koDaSI1OEoqKneIg,1014
-yomitoku/data/functions.py,sha256=2rJz4Gfd3UzlTq2bzXyFhcwtxJoUjNsnnNMJfk5-i4o,7361
+yomitoku/data/functions.py,sha256=eOyxo8S6EoAf1xGSPLWQFb9-t5Rg52NggD9MFIrOSpY,7506
 yomitoku/export/__init__.py,sha256=aANEfuovH2aevFjb2pGrBLFP-4iRzEzD9wcriCR-M7I,229
 yomitoku/export/export_csv.py,sha256=-n8eYPIzDQuiixeqpTbWaN9aQ5oFyl7XRfpv51oKPTI,1979
 yomitoku/export/export_html.py,sha256=X3H_orkS1BRlQo8Z1NzgrFwsIboDzRAx9etmqj90k2Y,4866
 yomitoku/export/export_json.py,sha256=1ChvCAHfCmMQvCfcAb1p3fSpr4elNAs3xBSIbpfn3bc,998
 yomitoku/export/export_markdown.py,sha256=mCcsXUWBLrYc1NcRSBFfBT28d6eCddAF1oHp0qdBEnE,3986
 yomitoku/models/__init__.py,sha256=Enxq9sjJWusZuxecTori8IQa8NEYKaiiptDluHX1avg,144
-yomitoku/models/dbnet_plus.py,sha256=VsE9anGOL1OzCivLilWpJ__32JHnSBEJOwdk_fpHE_o,8428
-yomitoku/models/parseq.py,sha256=OfN3ts1Z6f5T27amoRKvnL8qCma-wf0veIbWWoG4GuU,8801
+yomitoku/models/dbnet_plus.py,sha256=jeWJZm0ihbxoJeAXBFK7uVIwoosx2IUNk7Ut5wRH0vA,7998
+yomitoku/models/parseq.py,sha256=7QT-q5_oWqXTDXobRk1R6Lpap_AxdC4AzkSsOgXjOwM,8611
 yomitoku/models/rtdetr.py,sha256=oJsr8RHz3frslhLfXdVJve47lUsrmqLjfdTrZ41tlQ0,687
 yomitoku/models/layers/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 yomitoku/models/layers/activate.py,sha256=HUw0q-76RNjZF-o9O3fowfJcw0t1H5o0pbyioGdqUvU,668
 yomitoku/models/layers/dbnet_feature_attention.py,sha256=Vpp_PiLVuI7Zs30TTg4RNRn16KTb81ewonADpUHd4aE,6060
 yomitoku/models/layers/parseq_transformer.py,sha256=33eroJf8rmgIptP-NpZLJMhG7XOTwV4rXsq674VrKnU,6704
-yomitoku/models/layers/rtdetr_backbone.py,sha256=8-57bh8IjUCL94qM5mTpOXUTYPih1Xek5E8xs5pMGBE,9537
-yomitoku/models/layers/rtdetr_hybrid_encoder.py,sha256=i19sqZAIfPVotvpWBuxpxbepi9xhnGlRpaiL9XMp_Cc,13804
+yomitoku/models/layers/rtdetr_backbone.py,sha256=QjfLW-3qn2My3Jbg6yLORX8A-D2sph9J9u3r5nNnDLo,9386
+yomitoku/models/layers/rtdetr_hybrid_encoder.py,sha256=D3dK37k7_0jPqV39-6Se8kBzF_SyZttNlbLleyNFiJU,13607
 yomitoku/models/layers/rtdetrv2_decoder.py,sha256=5bVYPLFYCy3PcjyHTPFHNLWqg3bctrk-dKVG4kayhaw,27517
 yomitoku/postprocessor/__init__.py,sha256=W4vUuqBaFtH5dlSBIYgyaCroGLMjpV6RrNGIBQ8NFVw,243
-yomitoku/postprocessor/dbnet_postporcessor.py,sha256=iEPOWbGaJ8YIYCQJOpBadbf7uUGEAPjmeNDcNAvY8yc,4523
-yomitoku/postprocessor/parseq_tokenizer.py,sha256=eXIHIazEkByjyXKegYEzQ3CE0ReAJYIC2VpQJjnNQjU,4337
+yomitoku/postprocessor/dbnet_postporcessor.py,sha256=o_y8b5REd2dFEdIpRcr6o-XBfOCHo9rBYGwokP_uhTc,4948
+yomitoku/postprocessor/parseq_tokenizer.py,sha256=e89_g_bc4Au3SchuxoJfJNATJTxFmVYetzXyAzPWm28,4315
 yomitoku/postprocessor/rtdetr_postprocessor.py,sha256=f52wfRKrxqSXy_LeidKDR9XAta_qPjto-oYEdO0XL8A,3386
 yomitoku/resource/MPLUS1p-Medium.ttf,sha256=KLL1KkCumIBkgQtx1n4SffdaFuCNffThktEAbkB1OU8,1758908
 yomitoku/resource/charset.txt,sha256=sU91kSi-9Wk4733bCXy4j_UDmvcsj96sHOq1ppUJlOY,21672
 yomitoku/utils/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 yomitoku/utils/graph.py,sha256=LKNB8ZhSQwOZMfeAimPMF5UCVVr2ZaUWoGDkz8z-uGU,456
 yomitoku/utils/logger.py,sha256=uOmtQDr0A0JD7wyFshedL08BiNrQorHnpktRXba8bjU,424
-yomitoku/utils/misc.py,sha256=dC1w3DmsoU_ECqngbAs14vPOFCbcecZSGmbztgwx4XU,2479
-yomitoku/utils/visualizer.py,sha256=EEDo4bts61FX6mJecgJiHtzY2vLH6sJOQgOVr9yVsF0,4912
-yomitoku-0.4.1.dist-info/METADATA,sha256=mTZjZU6_zGTcnYgGR7bu5nGNOnys7DFH_NkAbq3FQrc,7553
-yomitoku-0.4.1.dist-info/WHEEL,sha256=1yFddiXMmvYK7QYTqtRNtX66WJ0Mz8PYEiEUoOUUxRY,87
-yomitoku-0.4.1.dist-info/entry_points.txt,sha256=nFV3S11zgBNW0Qq_D0XQNg2R4lNXU_9XUFr6rdJoyF8,52
-yomitoku-0.4.1.dist-info/RECORD,,
+yomitoku/utils/misc.py,sha256=2Eyy7-9K_h4Mal1zGXq6OlxubfNzhS0mEYwn_xt7xl8,2497
+yomitoku/utils/visualizer.py,sha256=2pSmbhUPylzVVJ0bXtGDoNmMdArAByab4Py7Xavvs_A,5230
+yomitoku-0.5.1.dist-info/METADATA,sha256=-8bUVnN26cxYlZO0ZQH3liki_xMfhUX47ruHLl-2BGM,7817
+yomitoku-0.5.1.dist-info/WHEEL,sha256=1yFddiXMmvYK7QYTqtRNtX66WJ0Mz8PYEiEUoOUUxRY,87
+yomitoku-0.5.1.dist-info/entry_points.txt,sha256=nFV3S11zgBNW0Qq_D0XQNg2R4lNXU_9XUFr6rdJoyF8,52
+yomitoku-0.5.1.dist-info/RECORD,,