A General Toolbox for Identifying Object Detection Errors
βββββββββββββββββββ ββββββββ
ββββββββββββββββββββββββββββ
βββ ββββββ βββββββββ
βββ ββββββ βββββββββ
βββ βββββββββββββββββββ
βββ ββββββββββ ββββββββ
An easy-to-use, general toolbox to compute and evaluate the effect of object detection and instance segmentation on overall performance. This is the code for our paper: TIDE: A General Toolbox for Identifying Object Detection Errors (ArXiv) [ECCV2020 Spotlight].
Check out our ECCV 2020 short video for an explanation of what TIDE can do:
Installation
TIDE is available as a python package for python 3.6+ as tidecv. To install, simply install it with pip:
pip3 install tidecvThe current version is v1.0.1 (changelog).
Usage
TIDE is meant as a drop-in replacement for the COCO Evaluation toolkit, and getting started is easy:
from tidecv import TIDE, datasets
tide = TIDE()
tide.evaluate(datasets.COCO(), datasets.COCOResult('path/to/your/results/file'), mode=TIDE.BOX) # Use TIDE.MASK for masks
tide.summarize() # Summarize the results as tables in the console
tide.plot() # Show a summary figure. Specify a folder and it'll output a png to that folder.This prints evaluation summary tables to the console:
-- mask_rcnn_bbox --
bbox AP @ 50: 61.80
Main Errors
=============================================================
Type Cls Loc Both Dupe Bkg Miss
-------------------------------------------------------------
dAP 3.40 6.65 1.18 0.19 3.96 7.53
=============================================================
Special Error
=============================
Type FalsePos FalseNeg
-----------------------------
dAP 16.28 15.57
=============================
And a summary plot for your model's errors:
Jupyter Notebook
Check out the example notebook for more details.
Datasets
The currently supported datasets are COCO, LVIS, Pascal, and Cityscapes. More details and documentation on how to write your own database drivers coming soon!
Citation
If you use TIDE in your project, please cite
@inproceedings{tide-eccv2020,
author = {Daniel Bolya and Sean Foley and James Hays and Judy Hoffman},
title = {TIDE: A General Toolbox for Identifying Object Detection Errors},
booktitle = {ECCV},
year = {2020},
}
Contact
For questions about our paper or code, make an issue in this github or contact Daniel Bolya. Note that I may not respond to emails, so github issues are your best bet.