Various tools for computer vision

Codes for label format conversion.

• toCOCO
  

  I have achieved the conversion to coco annotations including: IC15, HRSC2016, UCAS-AOS, NWPU VHR 10. and VOC. Besides, it also supports labelme annotations converted to COCO format.

• toDOTA

The codes in this part support the conversion of data annotations from IC15, HRSC2016, UCAS-AOS, NWPU VHR 10 into polygonal form of DOTA annotations. And it also supports the preprocessing code of the corresponding datasets and the .json files generation codes.

• toTXT

  Convert to the label format required for mAP calculation, the calculation code refers to my implementation here (for rbox).

• toYOLO

  This part will no longer be maintained and it supports both xml and ICDAR formats.

• MSRA-TD500
  Support evaluation on MSRA-TD500 with IC15 scripts.

• DOTA_devkit

  Toolkits for DOTA datasets, with some annotations and modification attached.

• dataset_partition

  Dataset partition for train, val, test part.
  Note: x2 means train + val, x3 means train + val + test.
  Remember to enlarge val & test set if your dataset is tiny. (such as 6:2:2)

• subdataset_generation

  Division of subset from total dataset, used for hyperparameter adjust.
  (you can regard it as x1 dataset_partition)

• generate_imageset

  Generate image absolute path for easy training. Supported data sets include: IC13, IC15, HRSC2016, DOTA, UCAS_AOD, NWPU_VHR, and VOC.

• operate_on_datasets
  Operations on dataset , such as copy, label matching.

Codes for rotated object detection. Supports functions such as cuda rnms, cuda riou, python rnms, pytorch riou calculations.

I have implemented some data augmentations such as Affine, HSV transform, Blur, Grayscale, Gamma, RandomNoise, Sharpen, Contrast, RandomFlip. On this basis, imbalanced datasets can be automatically augmented via simple sampling strategy here.

Useful tool for drawing bbox through providied points. The only customed part is your points obtaining function.

Support rbox evaluation and mAP calculation for object detection.

Provided a template for plotting 2D and 3D figure.

Simple examples for excel files processing via pandas.

Visualization for training process.

Easy implemention for crawling info from website.

Feature visualization tools.

• asff-fpn
• bam-attention

PRs are welcomed, if you have any questions, you can open an issue or contact me via mq_chaser@126.com.