Upload app.py
Browse files
app.py
ADDED
@@ -0,0 +1,159 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
# -*- coding: utf-8 -*-
|
2 |
+
"""LiLT For Deployment
|
3 |
+
|
4 |
+
Automatically generated by Colaboratory.
|
5 |
+
|
6 |
+
Original file is located at
|
7 |
+
https://colab.research.google.com/drive/1ol6RWyff15SF6ZJPf47X5380hBTEDiUH
|
8 |
+
"""
|
9 |
+
|
10 |
+
# ## Installing the dependencies (might take some time)
|
11 |
+
|
12 |
+
# !pip install -q pytesseract
|
13 |
+
# !sudo apt install -q tesseract-ocr
|
14 |
+
# !pip install -q transformers
|
15 |
+
# !pip install -q pytorch-lightning
|
16 |
+
# !pip install -q einops
|
17 |
+
# !pip install -q tqdm
|
18 |
+
# !pip install -q gradio
|
19 |
+
# !pip install -q Pillow==7.1.2
|
20 |
+
# !pip install -q wandb
|
21 |
+
# !pip install -q gdown
|
22 |
+
# !pip install -q torchmetrics
|
23 |
+
|
24 |
+
## Requirements.txt
|
25 |
+
import os
|
26 |
+
os.system('pip install pyyaml==5.1')
|
27 |
+
## install PyTesseract
|
28 |
+
os.system('pip install -q pytesseract')
|
29 |
+
os.environ["TOKENIZERS_PARALLELISM"] = "false"
|
30 |
+
|
31 |
+
import pandas as pd
|
32 |
+
import os
|
33 |
+
from PIL import Image
|
34 |
+
from transformers import RobertaTokenizer
|
35 |
+
import torch
|
36 |
+
from torch.utils.data import Dataset, DataLoader
|
37 |
+
import torch.nn as nn
|
38 |
+
import pytorch_lightning as pl
|
39 |
+
|
40 |
+
from dataset import create_features
|
41 |
+
from modeling import LiLT
|
42 |
+
from utils import LiLTPL
|
43 |
+
|
44 |
+
import gdown
|
45 |
+
import gradio as gr
|
46 |
+
|
47 |
+
seed = 42
|
48 |
+
|
49 |
+
## One can change this configuration and try out new combination
|
50 |
+
config = {
|
51 |
+
"hidden_dropout_prob": 0.1,
|
52 |
+
"hidden_size_t": 768,
|
53 |
+
"hidden_size" : 768,
|
54 |
+
"hidden_size_l": 768 // 6,
|
55 |
+
"intermediate_ff_size_factor": 4,
|
56 |
+
"max_2d_position_embeddings": 1001,
|
57 |
+
"max_seq_len_l": 512,
|
58 |
+
"max_seq_len_t" : 512,
|
59 |
+
"num_attention_heads": 12,
|
60 |
+
"num_hidden_layers": 12,
|
61 |
+
'dim_head' : 64,
|
62 |
+
"shape_size": 96,
|
63 |
+
"vocab_size": 50265,
|
64 |
+
"eps": 1e-12,
|
65 |
+
"fine_tune" : True
|
66 |
+
}
|
67 |
+
|
68 |
+
id2label = ['scientific_report',
|
69 |
+
'resume',
|
70 |
+
'memo',
|
71 |
+
'file_folder',
|
72 |
+
'specification',
|
73 |
+
'news_article',
|
74 |
+
'letter',
|
75 |
+
'form',
|
76 |
+
'budget',
|
77 |
+
'handwritten',
|
78 |
+
'email',
|
79 |
+
'invoice',
|
80 |
+
'presentation',
|
81 |
+
'scientific_publication',
|
82 |
+
'questionnaire',
|
83 |
+
'advertisement']
|
84 |
+
|
85 |
+
## Defining tokenizer
|
86 |
+
tokenizer = RobertaTokenizer.from_pretrained('roberta-base')
|
87 |
+
|
88 |
+
url = 'https://drive.google.com/uc?id=1eRV4fS_LFwI5MHqcRwLUNQZgewxI6Se_'
|
89 |
+
output = 'lilt_ckpt.ckpt'
|
90 |
+
gdown.download(url, output, quiet=False)
|
91 |
+
|
92 |
+
class RVLCDIPData(Dataset):
|
93 |
+
|
94 |
+
def __init__(self, image_list, label_list, tokenizer, max_len = 512, size = 1000):
|
95 |
+
|
96 |
+
self.image_list = image_list
|
97 |
+
self.label_list = label_list
|
98 |
+
self.tokenizer = tokenizer
|
99 |
+
self.max_seq_length = max_len
|
100 |
+
self.size = size
|
101 |
+
|
102 |
+
def __len__(self):
|
103 |
+
return len(self.image_list)
|
104 |
+
|
105 |
+
def __getitem__(self, idx):
|
106 |
+
img_path = self.image_list[idx]
|
107 |
+
label = self.label_list[idx]
|
108 |
+
|
109 |
+
boxes, words, normal_box = create_features(
|
110 |
+
img_path = img_path,
|
111 |
+
tokenizer = self.tokenizer,
|
112 |
+
max_seq_length = self.max_seq_length,
|
113 |
+
size = self.size,
|
114 |
+
use_ocr = True,
|
115 |
+
)
|
116 |
+
|
117 |
+
final_encoding = {'input_boxes': boxes, 'input_words': words}
|
118 |
+
final_encoding['label'] = torch.as_tensor(label).long()
|
119 |
+
|
120 |
+
return final_encoding
|
121 |
+
|
122 |
+
lilt = LiLTPL(config)
|
123 |
+
# path_to_weights = 'drive/MyDrive/docformer_rvl_checkpoint/docformer_v1.ckpt'
|
124 |
+
lilt.load_from_checkpoint('lilt_ckpt.ckpt')
|
125 |
+
|
126 |
+
## Taken from LayoutLMV2 space
|
127 |
+
|
128 |
+
image = gr.inputs.Image(type="pil")
|
129 |
+
label = gr.outputs.Label(num_top_classes=5)
|
130 |
+
examples = [['00093726.png'], ['00866042.png']]
|
131 |
+
title = "Interactive demo: LiLT for Image Classification"
|
132 |
+
description = "Demo for classifying document images with LiLT model. To use it, \
|
133 |
+
simply upload an image or use the example images below and click 'submit' to let the model predict the 5 most probable Document classes. \
|
134 |
+
Results will show up in a few seconds."
|
135 |
+
|
136 |
+
def classify_image(image):
|
137 |
+
|
138 |
+
image.save('sample_img.png')
|
139 |
+
boxes, words, normal_box = create_features(
|
140 |
+
img_path = 'sample_img.png',
|
141 |
+
tokenizer = tokenizer,
|
142 |
+
max_seq_length = 512,
|
143 |
+
size = 1000,
|
144 |
+
use_ocr = True,
|
145 |
+
)
|
146 |
+
|
147 |
+
final_encoding = {'input_boxes': boxes.unsqueeze(0), 'input_words': words.unsqueeze(0)}
|
148 |
+
output = lilt.forward(final_encoding)
|
149 |
+
output = output[0].softmax(axis = -1)
|
150 |
+
|
151 |
+
final_pred = {}
|
152 |
+
for i, score in enumerate(output):
|
153 |
+
score = output[i]
|
154 |
+
final_pred[id2label[i]] = score.detach().cpu().tolist()
|
155 |
+
|
156 |
+
return final_pred
|
157 |
+
|
158 |
+
gr.Interface(fn=classify_image, inputs=image, outputs=label, title=title, description=description, examples=examples, enable_queue=True).launch(debug=True)
|
159 |
+
|