search

waspinator / deep-learning-explorer

Quickly explore how different deep learning systems work with your data
Apache License 2.0
42 stars 15 forks source link

Your coco-style shape datasets can be evaluated by cocoapi？ #1

Open gaqiness opened 6 years ago

gaqiness commented 6 years ago

Hi waspinator @waspinator You write the /libraries/mextra/utils.py for model evaluating in your shape datasets. right? But, I think the your shape datasets is coco-style. Can I use coco-API to evaluate the model in your shape dataset? https://github.com/nightrome/cocoapi https://github.com/nightrome/cocoapi/blob/master/PythonAPI/pycocoEvalDemo.ipynb How can I do this? Could you tell me more detail? many thanks!

waspinator commented 6 years ago

I haven't used cocoeval.py, but it looks like you need both the ground truth and the results to be in coco-style. I don't convert the output back into coco-style, but that's a good idea for improvement.

I'm not sure when I can get to it, but if you wanted to try and implement mask-rcnn output to coco, you could try using https://github.com/waspinator/pycococreator as a reference.

gaqiness commented 6 years ago

many thanks, I will try this