Create app.py
Browse files
app.py
ADDED
@@ -0,0 +1,56 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import argparse
|
2 |
+
|
3 |
+
import cv2
|
4 |
+
from ditod import add_vit_config
|
5 |
+
from detectron2.config import get_cfg
|
6 |
+
from detectron2.utils.visualizer import ColorMode, Visualizer
|
7 |
+
from detectron2.data import MetadataCatalog
|
8 |
+
from detectron2.engine import DefaultPredictor
|
9 |
+
|
10 |
+
|
11 |
+
# Step 1: instantiate config
|
12 |
+
cfg = get_cfg()
|
13 |
+
add_vit_config(cfg)
|
14 |
+
cfg.merge_from_file("cascade_dit_base.yaml")
|
15 |
+
|
16 |
+
# Step 2: add model weights URL to config
|
17 |
+
cfg.MODEL.WEIGHTS = https://layoutlm.blob.core.windows.net/dit/dit-fts/publaynet_dit-b_mrcnn.pth
|
18 |
+
|
19 |
+
# Step 3: set device
|
20 |
+
# TODO also support GPU
|
21 |
+
cfg.MODEL.DEVICE='cpu'
|
22 |
+
|
23 |
+
# Step 4: define model
|
24 |
+
predictor = DefaultPredictor(cfg)
|
25 |
+
|
26 |
+
|
27 |
+
def analyze_image(img):
|
28 |
+
md = MetadataCatalog.get(cfg.DATASETS.TEST[0])
|
29 |
+
if cfg.DATASETS.TEST[0]=='icdar2019_test':
|
30 |
+
md.set(thing_classes=["table"])
|
31 |
+
else:
|
32 |
+
md.set(thing_classes=["text","title","list","table","figure"])
|
33 |
+
|
34 |
+
output = predictor(img)["instances"]
|
35 |
+
v = Visualizer(img[:, :, ::-1],
|
36 |
+
md,
|
37 |
+
scale=1.0,
|
38 |
+
instance_mode=ColorMode.SEGMENTATION)
|
39 |
+
result = v.draw_instance_predictions(output.to("cpu"))
|
40 |
+
result_image = result.get_image()[:, :, ::-1]
|
41 |
+
|
42 |
+
return result_image
|
43 |
+
|
44 |
+
title = "Interactive demo: Document Layout Analysis with DiT"
|
45 |
+
description = "This is a demo for Microsoft's Document Image Transformer (DiT)."
|
46 |
+
examples =[['document.png']]
|
47 |
+
|
48 |
+
iface = gr.Interface(fn=analyze_image,
|
49 |
+
inputs=gr.inputs.Image(type="numpy"),
|
50 |
+
outputs=gr.outputs.Image(type="numpy", label="analyzed image"),
|
51 |
+
title=title,
|
52 |
+
description=description,
|
53 |
+
article=article,
|
54 |
+
examples=examples,
|
55 |
+
enable_queue=True)
|
56 |
+
iface.launch(debug=True)
|