Description:
rpycocotools 0.0.7
rpycocotools
The rpycocotools package provides tools to load, manipulate, convert and visualize COCO format datasets. The documentation is available here.
Installation
The package is available on PyPI here, and can installed with pip:
pip install rpycocotools
You can also git clone this repo and build it yourself with:
pip install -r requirements/requirements-build.txt
pip install .
Usage example
Visualize image with a given id:
import rpycocotools
coco_dataset = rpycocotools.COCO("../data_samples/coco_25k/annotations.json", "../data_samples/coco_25k/images")
coco_dataset.visualize_img(174482)
import rpycocotools
coco_dataset = rpycocotools.COCO("../data_samples/coco_25k/annotations.json", "../data_samples/coco_25k/images")
anns = coco_dataset.get_img_anns(174482)
mask = rpycocotools.mask.decode(anns[0].segmentation)
The mask is a numpy array and can be visualized (for example with opencv):
Benchmarks
Details
There are a few benchmarking scripts to compare to pycocotools.
The results reported here are done on my own PC and presented only to get a general idea. I might run the benchmark on a more reproducible environment in the future.
Setup
Some of the benchmarks use the instances_train2017.json files from the 2017 COCO dataset.
Either place this file in the data_samples folder or only run the commands below with the -m "not coco2017" option.
pip install -r requirements/requirements-benchmarks.txt
pip install .
Load
Benchmark how much time it takes load a COCO dataset.
python -m pytest benchmarks/load.py -vv
Results:
Test Name
Mean time in s
rpycocotools on COCO instances_train2017.json
4.4
pycocotools on COCO instances_train2017.json
16.5
Area
Benchmark how much time it takes to compute the total number of mask pixels in a COCO dataset.
python -m pytest benchmarks/area.py -vv -m coco2017
Results:
Test Name
Mean time in ms
rpycocotools on COCO instances_train2017.json
880.6
pycocotools on COCO instances_train2017.json
19,302.9
Decode masks
Benchmark how much time it takes to decode all the masks in a COCO dataset.
python -m pytest benchmarks/decode.py -vv -m coco2017
Results:
Test Name
Mean time in s
rpycocotools on COCO instances_train2017.json
371
pycocotools on COCO instances_train2017.json
141
Results after converting all the segmentations to RLE before decoding (conversion time not included):
Test Name
Mean time in s
rpycocotools on COCO instances_train2017.json
300
pycocotools on COCO instances_train2017.json
120
License
For personal and professional use. You cannot resell or redistribute these repositories in their original state.
Customer Reviews
There are no reviews.
