FSOCO Stats for Nerds

This brief presentation should give you an overview of the underlying data.

Intentionally blank page¶

Loading BokehJS ...

WARNING:param.Parameterized: Use method 'warning' via param namespace 
WARNING:param.main: pandas could not register all extension types imports failed with the following error: cannot import name 'ABCIndexClass' from 'pandas.core.dtypes.generic' (/home/voedisch/anaconda3/envs/fsoco/lib/python3.8/site-packages/pandas/core/dtypes/generic.py)

Images

	Total	% of Total
yellow_cone	9,590	82.87 %
blue_cone	9,275	80.15 %
orange_cone	4,647	40.16 %
large_orange_cone	2,996	25.89 %
unknown_cone	911	7.87 %
Images in dataset	11,572	100.00 %

Cone Classes

	Total	% of Total
yellow_cone	93,257	42.22 %
blue_cone	86,116	38.99 %
orange_cone	25,991	11.77 %
large_orange_cone	8,217	3.72 %
unknown_cone	7,281	3.30 %
Cones in dataset	220,862	100.00 %

Cone Tags

	Total	% of Total
truncated	22,369	10.13 %
sticker_band_removed	10,130	4.59 %
knocked_over	2,596	1.18 %
Objects not tagged	186,809	84.58 %
Cones in dataset	220,862	100.00 %

Cone Class Distribution

BBox Aspect Ratios

Dataset Statistics

	global cosine 99%	global cosine 98%	global cosine 95%
fsoco_v1	8.97	28.85	163.65
fsoco_v2	0.97	8.13	169.69

FSOCOv1 was the first, rough iteration of this collaborative dataset.
You can read up on our reasons for doing a complete overhaul with the release of this joint project.
The values represent the mean amount of similar (other) images per image, with varying degrees of similarity.
For more details on the metrics, read up on the documentation for the Similarity Scorer.

FSOCOv1 within Team similarity

FSOCOv2 within Team similarity

What if I want to do my own data analysis?¶

The presentation was generated based on this Jupyter Notebook. You'll also need to download the pkl files to run it locally.