←
Duffy Digest463 posts
Duffy Digest avatar

Duffy Digest

@DigestDuffy

My research, opinions, and data viz: (main @duffydrew1)

◷ Joined January 2023▧ 147 Photos and videos
Duffy Digest avatar
Duffy Digest@DigestDuffy·Nov 12

Not sure why I thought you might enjoy this @danarestia and @TerenceFoy -- but @DukeMLAX has a column in their roster for 5in5 eligibility year (3,1,5) seen below: Seems like they're on the forefront of this and other schools will soon follow. Pretty cool.

Post attachment
Duffy Digest avatar
Duffy Digest@DigestDuffy·Oct 19

Ryan Croddick had an incredible 2026 Final Four run to cap of his first team All-American season. Here is a plot of his saves in the semifinal against Duke using CV from the broadcast highlights. Most shooters are pretty good from 12yds and in...not against Croddick!

Post attachment
Duffy Digest avatar
Duffy Digest@DigestDuffy·Sep 29

Broadcast tracking is able to give us unprecedented insight into shooter tendencies and evaluation. Here, one of the best in the college game, Benn Johnston vs Princeton in the semifinal 2026. Distances are still approximate, but seem relatively close based on the video.

Post attachment
Duffy Digest avatar
Duffy Digest@DigestDuffy·Sep 28

Princeton/Duke in 2026 FF to map shot angle and both estimated shot distance and defender positioning. Croddick makes a save on a shot from ~3 yards out with no defender within stick length (~3 yards). I’d classify as: “uncontested”. So many derivations to create w this

Post attachmentPost attachment
0:16
Duffy Digest avatar
Duffy Digest@DigestDuffy·Sep 25

Next CV step would be creating heatmaps for offensive/defensive locations. Here we look at ND offense (again frames drop for individuals). I evaluated this based on absolute time. Further derivatives would include scoring value area, dodging angles and initiation, speed, etc.

0:37
Duffy Digest avatar
Duffy Digest@DigestDuffy·Sep 21

The next step that can be built off broadcast feed in my eyes would be "role annotations". On ball defender, primary and secondary slide guys, and distance that each player is away from assignment. Astra is very good at estimating distances relatively accurately from video too.

0:37
Duffy Digest avatar
Duffy Digest@DigestDuffy·Sep 17

This is really where I think lacrosse computer vision tracking will become highly valuable and advanced in the coming months: - O/D shapes being so fluid at different points on field - Slide timing - Pick and Roll effects - Sub advantages All done through GPT-6 Astra.

0:37
Duffy Digest avatar
Duffy Digest@DigestDuffy·Sep 17

@laxreference found some of your opinions from 2017. Hoping there’s more opportunities in this space in the coming months/years! (Would help with a PLL investment into the tech too). Thanks for some of the inspo!

Duffy Digest avatar
Duffy Digest@DigestDuffy·Jul 26

Here are: 1) zones I charted a full game into 2) the resulting xT values of those zones 3) how zones begin to change with build up and more look ahead 4) where the ball actually flows from the course of a game Hope you enjoy the read! Would love to hear your feedback!

Post attachmentPost attachmentPost attachmentPost attachment