All kinds of utils for format convertion or something else (described in readme files)
Each python file matches some samples inside.
More details are marked in the python files as annotation or README.md
file inside the folder.
Some conversion files included here,I'll update it if necessary.
- txt2voc
The tool inside helps you to conver txt in certain format to voc format.
Provided txt labels ground truth in form ofNWPU VHR-10
, it's not a hard job to change related code for your owm labels.
NWPU VHR-10's label is showed as follow:
2,(270,296),(358,685)
2,(366,278),(453,661)
4,(541,633),(727,698)
-
voc2txt
User-friendly work for extracting imformation from xml files. And an example is attached for better understanding. -
labelme2COCO
It's a bit hard to finish this work , cause the widespread wrong version about it , I hate plagiarism :)
This tool helps you to convert json file created by labelme to COCO style for training.
Attentions:
1.If you want to label the segmentation mask , there isno need
to label bbox!(Or bugs arised)
2.When using labelme , pls named mask according to COCO format,such asvehicle_car_1
.(supercategory,category,instance id) -
voc2coco
Just use it. -
voc2yolo
Yolo format: class_id + Normalized xywh (id markded from 0) for each row.
- augmentation
Various kinds of data augmentation implementions as well as some demos are concluded inside .
- drawbox
Useful tool for drawing bbox through providied points.The only customed part is your points obtaining function.
Provided a template for plotting 2D and 3D figure.
-
data_partition
Helpful of partition for dataset. -
img_format_trans
A simple tool for conversion of image format. -
visdom-train-example
A example for training while monitoring on loss and ac curve.More details and attention have been attached to file inside. -
python-cmd
Linux command execution through python file , provided as a easy but sufficient demo.
Beside,exec
commond also helps for many cases ,take good use of it.(such as drawing pictures) -
shuffle
Randomly generate small subdataset from the dataset. -
match_check
Verify if images and annotations do not correspond and output the associated file names. -
crop_bbox_and_save
Crop bbox area from raw image and save for other usage. -
generate_imageset
Generate trainval setting files.
(Two mode included:yolo and voc) -
K-means
K-means implement for box clustering.