polimorfo 0.9.54

a dataset loader and converter for object detection segmentation and classification

polimòrfo

Polimòrfo (πολύμορϕος, comp. di πολυ- «poli-» e μορϕή «forma») is a dataset loader and converter library for object detection segmentation and classification. The goal of the project is to create a library able to process dataset in format:

• COCO: Common Objects in Context

• Pascal VOC: Visual Object Classes Challenge

• Google Open Images: Object Detection and Segmentation dataset released by Google

and transform these dataset into a common format (COCO).

Moreover, the library offers utilies to handle (load, convert, store and transform) the various type of annotations. This is important when you need to: - convert mask to polygons - store mask in a efficient format - convert mask/poygons into bounding boxes

• Free software: Apache Software License 2.0

• Documentation: https://polimorfo.readthedocs.io.

Features

TODO

• [X] Coco dataset

• [X] download coco datasets for train and val

• [X] add annotations loader and converter

• [X] add the ability to create dataet from scratch

• [ ] add voc dataset format

• [ ]

Credits

This package was created with Cookiecutter and the audreyr/cookiecutter-pypackage project template.

History

0.2.0 (2020-02-18)

• Add support to process coco dataset

0.2.1 (2020-02-28)

• add support to download files and archives from the web and google drive

0.3.0 (2020-10-04)

• addedd support for removing categories and other utilities

0.4.0 (2020-10-05)

• addedd support to create a dataset from scratch

0.5.0 (2020-10-06)

• added support to visualize images and annotations

• make image removing optional during annotations and categories deletion

0.6.0 (2020-10-12)

• added copy dataset

• added split dataset

0.6.1 (2020-10-12)

• fixed a bug in colors generation for show images

0.6.2 (2020-10-12)

• update signature for function def update_images_path(self, func):

0.7.0 (2020-10-19)

• add method to dump dataset in format segmentation map

0.8.0 (2020-10-23)

• fixed bug in maskutils.mask_to_polygons

• add class to transform the predictions from instance and semantic segmentation in coco format

• fixed bug in add_image, add_annotation, add_category

• make load_image and load_images load random images sampled from the dataset

0.8.1 (2020-10-23)

• fixed bug for tqdm when removing a category and its annotations from the dataset

0.8.2 (2020-10-23)

• removed the prefix jpg when saving masks

• update draw instance to draw only bounding boxes

0.8.3 (2020-10-24)

• fixed bug in enum for draw instances

0.8.4 (2020-10-24)

• add show bounding boxes

0.8.5 (2020-10-24)

• changed representation for masks from [width, height, labels] to [labels, width, height]

0.8.6 (2020-10-24)

• added method to crop images

• added method to move annotations with respect a bounding box

0.8.7 (2020-10-24)

• support fully creation o a new dataset

0.8.8-11 (2020-10-26)

• fixed vairous bugs

0.8.12 (2020-10-26)

• fixed bug when the size of the segments is equal to 4

0.8.13 (2020-10-26)

• fixed bug in json dump to serialize numpy array

0.8.14 (2020-10-26)

• fixed bug in json dump to serialize numpy types

0.9.1 (2020-10-28)

• fixed various bugs

• add index for speedup lookup operations

0.9.2 (2020-10-28)

• add new feature to compute mean average precision and recall per class and global

0.9.3 (2020-10-28)

• add computation of mean average precision and mean average recall per image

0.9.4 (2020-10-28)

• fixed bug in score computation

0.9.36

• fixed bug in mask generation

• feature that allows us to add a single mask per component when saving segmentation results

0.9.38

• add min confidence when displaying prediction from a segmentation mask model

• now semantic coco accepts only logits to create annotations

0.9.39

• add new method to remap category idxs

0.9.48

• add new feature to save images and masks to a folder and filter out images and mask with less than k annotations

0.9.52

• the method get_segmentation_mask return also the avg score of the image annotations

• the method save_mask_images save also a weight files with the avg score for the image