from typing import Tuple, List, Sequence, Optional, Union
from torchvision import transforms
from torch import nn, Tensor
from PIL import Image
from pathlib import Path
from bs4 import BeautifulSoup as bs
import numpy as np
import numpy.typing as npt
from numpy import uint8
ImageType = npt.NDArray[uint8]
from transformers import AutoModelForObjectDetection
import torch
import matplotlib.pyplot as plt
import matplotlib.patches as patches
from matplotlib.patches import Patch
from unitable import UnitablePredictor
from doctrfiles import DoctrWordDetector,DoctrTextRecognizer
from utils import crop_an_Image,cropImageExtraMargin
from utils import denoisingAndSharpening
#based on this notebook:https://github.com/NielsRogge/Transformers-Tutorials/blob/master/Table%20Transformer/Inference_with_Table_Transformer_(TATR)_for_parsing_tables.ipynb
class MaxResize(object):
def __init__(self, max_size=800):
self.max_size = max_size
def __call__(self, image):
width, height = image.size
current_max_size = max(width, height)
scale = self.max_size / current_max_size
resized_image = image.resize((int(round(scale*width)), int(round(scale*height))))
return resized_image
html_table_template = (
lambda table: f"""
"""
)
class DetectionAndOcrTable1():
def __init__(self,englishFlag=True):
self.unitablePredictor = UnitablePredictor()
self.wordDetector = DoctrWordDetector(architecture="db_resnet50",
path_weights="doctrfiles/models/db_resnet50-79bd7d70.pt",
path_config_json ="doctrfiles/models/db_resnet50_config.json")
if englishFlag:
self.textRecognizer = DoctrTextRecognizer(architecture="master", path_weights="./doctrfiles/models/master-fde31e4a.pt",
path_config_json="./doctrfiles/models/master.json")
else:
self.textRecognizer = DoctrTextRecognizer(architecture="parseq", path_weights="./doctrfiles/models/doctr-multilingual-parseq.bin",
path_config_json="./doctrfiles/models/multilingual-parseq-config.json")
@staticmethod
def build_table_from_html_and_cell(
structure: List[str], content: List[str] = None
) -> List[str]:
"""Build table from html and cell token list"""
assert structure is not None
html_code = list()
# deal with empty table
if content is None:
content = ["placeholder"] * len(structure)
for tag in structure:
if tag in ("| [] | ", ">[]"):
if len(content) == 0:
continue
cell = content.pop(0)
html_code.append(tag.replace("[]", cell))
else:
html_code.append(tag)
return html_code
@staticmethod
def save_detection(detected_lines_images:List[ImageType], prefix = './res/test1/res_'):
i = 0
for img in detected_lines_images:
pilimg = Image.fromarray(img)
pilimg.save(prefix+str(i)+'.png')
i=i+1
@staticmethod
# for output bounding box post-processing
def box_cxcywh_to_xyxy(x):
x_c, y_c, w, h = x.unbind(-1)
b = [(x_c - 0.5 * w), (y_c - 0.5 * h), (x_c + 0.5 * w), (y_c + 0.5 * h)]
return torch.stack(b, dim=1)
@staticmethod
def rescale_bboxes(out_bbox, size):
img_w, img_h = size
b = DetectionAndOcrTable1.box_cxcywh_to_xyxy(out_bbox)
b = b * torch.tensor([img_w, img_h, img_w, img_h], dtype=torch.float32)
return b
@staticmethod
def outputs_to_objects(outputs, img_size, id2label):
m = outputs.logits.softmax(-1).max(-1)
pred_labels = list(m.indices.detach().cpu().numpy())[0]
pred_scores = list(m.values.detach().cpu().numpy())[0]
pred_bboxes = outputs['pred_boxes'].detach().cpu()[0]
pred_bboxes = [elem.tolist() for elem in DetectionAndOcrTable1.rescale_bboxes(pred_bboxes, img_size)]
objects = []
for label, score, bbox in zip(pred_labels, pred_scores, pred_bboxes):
class_label = id2label[int(label)]
if not class_label == 'no object':
objects.append({'label': class_label, 'score': float(score),
'bbox': [float(elem) for elem in bbox]})
return objects
@staticmethod
def fig2img(fig):
"""Convert a Matplotlib figure to a PIL Image and return it"""
import io
buf = io.BytesIO()
fig.savefig(buf)
buf.seek(0)
img = Image.open(buf)
return img
#For that, the TATR authors employ some padding to make sure the borders of the table are included.
@staticmethod
def objects_to_crops(img, tokens, objects, class_thresholds, padding=10):
"""
Process the bounding boxes produced by the table detection model into
cropped table images and cropped tokens.
"""
table_crops = []
for obj in objects:
# abit unecessary here cause i crop them anywyas
if obj['score'] < class_thresholds[obj['label']]:
continue
cropped_table = {}
bbox = obj['bbox']
bbox = [bbox[0]-padding, bbox[1]-padding, bbox[2]+padding, bbox[3]+padding]
cropped_img = img.crop(bbox)
# Add padding to the cropped image
padded_width = cropped_img.width + 40
padded_height = cropped_img.height +40
new_img_np = np.full((padded_height, padded_width, 3), fill_value=255, dtype=np.uint8)
y_offset = (padded_height - cropped_img.height) // 2
x_offset = (padded_width - cropped_img.width) // 2
new_img_np[y_offset:y_offset + cropped_img.height, x_offset:x_offset+cropped_img.width] = np.array(cropped_img)
padded_img = Image.fromarray(new_img_np,'RGB')
table_tokens = [token for token in tokens if iob(token['bbox'], bbox) >= 0.5]
for token in table_tokens:
token['bbox'] = [token['bbox'][0]-bbox[0] + padding,
token['bbox'][1]-bbox[1] + padding,
token['bbox'][2]-bbox[0] + padding,
token['bbox'][3]-bbox[1] + padding]
# If table is predicted to be rotated, rotate cropped image and tokens/words:
if obj['label'] == 'table rotated':
padded_img = padded_img.rotate(270, expand=True)
for token in table_tokens:
bbox = token['bbox']
bbox = [padded_img.size[0]-bbox[3]-1,
bbox[0],
padded_img.size[0]-bbox[1]-1,
bbox[2]]
token['bbox'] = bbox
cropped_table['image'] = padded_img
cropped_table['tokens'] = table_tokens
table_crops.append(cropped_table)
return table_crops
@staticmethod
def visualize_detected_tables(img, det_tables, out_path=None):
plt.imshow(img, interpolation="lanczos")
fig = plt.gcf()
fig.set_size_inches(20, 20)
ax = plt.gca()
for det_table in det_tables:
bbox = det_table['bbox']
if det_table['label'] == 'table':
facecolor = (1, 0, 0.45)
edgecolor = (1, 0, 0.45)
alpha = 0.3
linewidth = 2
hatch='//////'
elif det_table['label'] == 'table rotated':
facecolor = (0.95, 0.6, 0.1)
edgecolor = (0.95, 0.6, 0.1)
alpha = 0.3
linewidth = 2
hatch='//////'
else:
continue
rect = patches.Rectangle(bbox[:2], bbox[2]-bbox[0], bbox[3]-bbox[1], linewidth=linewidth,
edgecolor='none',facecolor=facecolor, alpha=0.1)
ax.add_patch(rect)
rect = patches.Rectangle(bbox[:2], bbox[2]-bbox[0], bbox[3]-bbox[1], linewidth=linewidth,
edgecolor=edgecolor,facecolor='none',linestyle='-', alpha=alpha)
ax.add_patch(rect)
rect = patches.Rectangle(bbox[:2], bbox[2]-bbox[0], bbox[3]-bbox[1], linewidth=0,
edgecolor=edgecolor,facecolor='none',linestyle='-', hatch=hatch, alpha=0.2)
ax.add_patch(rect)
plt.xticks([], [])
plt.yticks([], [])
legend_elements = [Patch(facecolor=(1, 0, 0.45), edgecolor=(1, 0, 0.45),
label='Table', hatch='//////', alpha=0.3),
Patch(facecolor=(0.95, 0.6, 0.1), edgecolor=(0.95, 0.6, 0.1),
label='Table (rotated)', hatch='//////', alpha=0.3)]
plt.legend(handles=legend_elements, bbox_to_anchor=(0.5, -0.02), loc='upper center', borderaxespad=0,
fontsize=10, ncol=2)
plt.gcf().set_size_inches(10, 10)
plt.axis('off')
if out_path is not None:
plt.savefig(out_path, bbox_inches='tight', dpi=150)
return fig
def predict(self,image:Image.Image,debugfolder_filename_page_name,denoise=False):
"""
0. Locate the table using Table detection
1. Unitable
"""
print("Running table transformer + Unitable Hybrid Model")
# Step 0 : Locate the table using Table detection TODO
#First we load a Table Transformer pre-trained for table detection. We use the "no_timm" version here to load the checkpoint with a Transformers-native backbone.
model = AutoModelForObjectDetection.from_pretrained("microsoft/table-transformer-detection", revision="no_timm")
device = "cuda" if torch.cuda.is_available() else "cpu"
model.to(device)
#Preparing the image for the model
detection_transform = transforms.Compose([
MaxResize(800),
transforms.ToTensor(),
transforms.Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225])
])
pixel_values = detection_transform(image).unsqueeze(0)
pixel_values = pixel_values.to(device)
# Next, we forward the pixel values through the model.
# The model outputs logits of shape (batch_size, num_queries, num_labels + 1). The +1 is for the "no object" class.
with torch.no_grad():
outputs = model(pixel_values)
# update id2label to include "no object"
id2label = model.config.id2label
id2label[len(model.config.id2label)] = "no object"
#[{'label': 'table', 'score': 0.9999570846557617, 'bbox': [110.24547576904297, 73.31171417236328, 1024.609130859375, 308.7159423828125]}]
objects = DetectionAndOcrTable1.outputs_to_objects(outputs, image.size, id2label)
#Only do these for objects with score greater than 0.8
objects = [obj for obj in objects if obj['score'] > 0.95]
print("detected object from the table transformers are")
print(objects)
if objects:
#Next, we crop the table out of the image. For that, the TATR authors employ some padding to make sure the borders of the table are included.
tokens = []
detection_class_thresholds = {
"table": 0.95, #this is a bit double cause we do up there another filtering but didn't want to modify too much from original code
"table rotated": 0.95,
"no object": 10
}
crop_padding = 10
tables_crops = DetectionAndOcrTable1.objects_to_crops(image, tokens, objects, detection_class_thresholds, padding=crop_padding)
cropped_tables =[]
for i in range (len(tables_crops)):
cropped_table = tables_crops[i]['image'].convert("RGB")
cropped_table.save(debugfolder_filename_page_name+"cropped_table_"+str(i)+".png")
cropped_tables.append(cropped_table)
# Step 1: Unitable
#This take PIL Images as input
if denoise:
cropped_tables =denoisingAndSharpening(cropped_tables)
pred_htmls, pred_bboxs = self.unitablePredictor.predict(cropped_tables,debugfolder_filename_page_name)
table_codes = []
for k in range(len(cropped_tables)):
pred_html =pred_htmls[k]
pred_bbox = pred_bboxs[k]
# Some tabless have a lot of words in their header
# So for the headers, give doctr word ddetector doesn't work when the images aren't square
table_header_cells = 0
header_exists = False
for cell in pred_html:
if cell=='>[]' or cell == '[] | ':
table_header_cells += 1
if cell =='':
header_exists = True
break
if not header_exists:
table_header_cells = 0
pred_cell = []
cell_imgs_to_viz = []
cell_img_num=0
# Find what one line should be if there is a cell with a single line
one_line_height = 100000
for i in range(table_header_cells):
box = pred_bbox[i]
xmin, ymin, xmax, ymax = box
current_box_height = abs(ymax-ymin)
if current_box_height 0 and current_box_height>one_line_height+5:
cell_img= cropImageExtraMargin([fourbytwo],cropped_tables[k],margin=1.4)[0]
table_header_cells -= 1
#List of 4 x 2
detection_results = self.wordDetector.predict(cell_img,sort_vertical=True)
input_to_recog = []
if detection_results == []:
input_to_recog.append(cell_img)
else:
for wordbox in detection_results:
cropped_image= crop_an_Image(wordbox.box,cell_img)
if cropped_image.shape[0] >0 and cropped_image.shape[1]>0:
input_to_recog.append(cropped_image)
else:
print("Empty image")
else:
cell_img = crop_an_Image(fourbytwo,cropped_tables[k])
if table_header_cells>0:
table_header_cells -= 1
if cell_img.shape[0] >0 and cell_img.shape[1]>0:
input_to_recog =[cell_img]
cell_imgs_to_viz.append(cell_img)
if input_to_recog != []:
words = self.textRecognizer.predict_for_tables(input_to_recog)
cell_output = " ".join(words)
pred_cell.append(cell_output)
else:
#Don't lose empty cell
pred_cell.append("")
print(pred_cell)
#Step3 :
pred_code = self.build_table_from_html_and_cell(pred_html, pred_cell)
pred_code = "".join(pred_code)
pred_code = html_table_template(pred_code)
soup = bs(pred_code)
#formatted and indented) string representation of the HTML document
table_code = soup.prettify()
print(table_code)
# Append extracted table to table_codes
table_codes.append(table_code)
return table_codes