/instance_segmetation_tools

实例分割小工具合集，包括mask2labelme format data 等

Primary LanguagePython

instance_segmetation_tools

Requirement

labelme==3.8.0

Support

  • trans labelme to mask (0 for background)
  • trans labelme to COCO format data
  • trans labelme to VOC format data
  • MAP for detection
  • MIoU for instance segmentation
  • Simple copy and paste for instance segmentation
  • mask png to labelme format json data

MIoU for instance segmentation

Pred mask & Label mask

[ 0, 0, 1, ....
  1, 1, 1, ....
  0, 2, 2, ....
  0, 0 ,0, ....
  ...
  ...]

0 for backdound, >= 1 for classes Info.txt

jt
qp

Results

-----------Acc of each classes-----------
back_ground          : 97.387168 %	
jt          : 97.197456 %	
qp          : 0.000141 %	
mean Pixel Acc: 64.86
-----------IoU of each classes-----------
back_ground          : 93.342430 %	
jt          : 95.128109 %	
qp          : 0.000140 %	
mIOU: 62.82

MAP for detection

copy from https://github.com/Cartucho/mAP

What's New

  • We add precision and recall statistics in output.
# <a name="6">mAP of all classes</a><a style="float:right;text-decoration:none;" href="#index">[Top]</a>
mAP = 74.78%
jt: tp 106, fp 18, fn 7, percision 0.855 recall 0.938
qp: tp 51, fp 25, fn 29, percision 0.671 recall 0.637
all classes: percision 0.785, recall 0.813
# <a name="7">Number of ground-truth objects per class</a><a style="float:right;text-decoration:none;" href="#index">[Top]</a>
jt: 113
qp: 80

Create the ground-truth files

```
<class_name> <left> <top> <right> <bottom> 1 or 0
```
  • The difficult parameter is 1 otherwise 0, use it if you want the calculation to ignore a specific detection.
  • E.g. "image_1.txt": 
    tvmonitor 2 10 173 238 0
book 439 157 556 241 0
book 437 246 518 351 1
pottedplant 272 190 316 259 0

Create the detection-results files

  • Create a separate detection-results text file for each image.
  • Use matching names for the files (e.g. image: "image_1.jpg", detection-results: "image_1.txt").
  • In these files, each line should be in the following format: 
    <class_name> <confidence> <left> <top> <right> <bottom>
  • E.g. "image_1.txt": 
    tvmonitor 0.471781 0 13 174 244
cup 0.414941 274 226 301 265
book 0.460851 429 219 528 247
chair 0.292345 0 199 88 436
book 0.269833 433 260 506 336

color

- green -> TP: True Positives (object detected and matches ground-truth)
- red -> FP: False Positives (object detected but does not match ground-truth)
- pink -> FN: False Negatives (object not detected but present in the ground-truth)