![](https://camo.githubusercontent.com/2a730c8c63f1423d2cba625c62827f58a69d92fc69a851db9418f5d80cc992b6/68747470733a2f2f666f7274686562616467652e636f6d2f696d616765732f6261646765732f6d616b65732d70656f706c652d736d696c652e737667)
Make awesome things that matter.
- Modify some configuration in config.yaml
- For evaluation metric, please refer to MegReader repository
# iou-based Pascal
make ioueval
# overlap-based DetEval
make deteval
History (on TotalText dataset)
![](./assets/train_history.png)
![](./assets/test_history.png)
![](./assets/test_img_history_01.png)
![](./assets/test_img_history_02.png)
Heatmap |
Polygon |
Rotated rectangle |
![](./assets/tt_heatmap_01.jpg) |
![](./assets/tt_poly_01.jpg) |
![](./assets/tt_rect_01.jpg) |
![](./assets/tt_heatmap_02.jpg) |
![](./assets/tt_poly_02.jpg) |
![](./assets/tt_rect_02.jpg) |
![](./assets/tt_heatmap_03.jpg) |
![](./assets/tt_poly_03.jpg) |
![](./assets/tt_rect_03.jpg) |
Text-line detection (the model trained on CTW1500 dataset)
Image origin |
Text-line detected |
![](./assets/ctw_gt_01.jpg) |
![](./assets/ctw_result_01.jpg) |
![](./assets/ctw_gt_02.jpg) |
![](./assets/ctw_result_02.jpg) |
![](./assets/ocr_01.jpg)
![](./assets/ocr_02.jpg)
![](./assets/ocr_03.jpg)
Metric evaluation (DetEval - P/R/HMean)
# for TotalText dataset
make deteval
Method |
image size |
init lr |
b-thresh |
p-thresh |
unclip ratio |
Precision |
Recall |
F-measure |
TotalText-resnet18-fcn (word-level) |
640 |
0.005 |
0.25 |
0.50 |
1.50 |
0.70 |
0.64 |
0.67 |
CTW1500-resnet18-fcn (line-level) |
640 |
0.005 |
0.25 |
0.50 |
1.50 |
0.83 |
0.66 |
0.74 |
I got a lot of code from DBNet.pytorch, thanks to @WenmuZhou