Evaluate accuracy of mask2former segmentation model

[jayqi]

We're using a pretrained mask2former model (see #55). We should run it on our images specifically (probably both Mapillary and @danbjoseph's own photos) and do some kind of accuracy evaluation.

[ericfrench2015]

@jayqi @danbjoseph I wonder if it's time for us to formally meet. I've been doing a little of this in the background - specifically, comparing mask2former-swin-large-cityscapes-semantic used in the NatureVisibility project to whatever Mapillary uses, as well as looking at outlier results in this MVP, and would love to compare notes and talk about how I can contribute. Dan, you're the center of gravity here. As usual my late afternoons are generally pretty flexible.

[danbjoseph]

maybe a virtual join option for the group at the next project night? https://www.meetup.com/Civic-Tech-DC/events/300644779 @ericfrench2015 you can also join the Slack via https://www.civictechdc.org/slack and look for the #street-level-imagery channel.

[ericfrench2015]

done.... Enjoy(?) Trinidad