Ghost Premier - New England

Why New England Premier? A D1-level facility in Massachusetts. Full strength & conditioning. Trackman, HitTrax, force plates & more. Two self-hosted showcases + real college coach connections. No eye wash, just real development.

Working on building our own motion capture automated pipeline with @theiamarkerless

@PremierNewEng Top notch facility with top notch coaching!

@KostekMikey up to 86-88 in his second bullpen back with a wipeout slider! Sidearmer throwing from a 2ft9 release height!

Extremely grateful for @PremierNewEng getting me built up this offseason. 2nd bullpen of the offseason FB at 86-88! 2ft 9 - 2ft 11 release height. @CoachWelchRBI @ScitBaseball

Some front toss swings, getting ready for high school season! Uncommitted 2027 MIF 5’10 150 4.3 GPA @LakersBaseball9

First higher intent throws of the off season with @PremierNewEng, up to 86 multiple times from the flatground on Trackman. Excited for my first bullpen next week! @CoachWelchRBI

Recent live at-bats @PremierNewEng @SuffolkU_BSB @WesBaseball @AmherstBaseball @BabsonBaseball @EphsCollBB @HamCollBaseball @SalveBaseball

Receiving and blocking from recent live at-bats @PremierNewEng @TuftsBaseball @BabsonBaseball @SuffolkU_BSB @BentleyBaseball @AmherstBaseball

High Follow| 2028 Grad | UP TO 87 | GOOD FEEL FOR 4 PITCHES | CHANGE UP WITH GREAT DEPTH

LIGHT TOWER POWER OUT OF THIS FRESHMAN IN HS | HIGH FOLLOW

2029 3B/1B Giovanni Naples Dexter Southfield || @PB_Uncommitted EV: 102 #BeSeen

2028 grad up to 87

Tune up live abs @CoastalBaseball @UConnBSB @BCBirdBall @AlabamaBSB @GoNUbaseball

Live at-bat @PremierNewEng

@John_Burke_28 @_Candela12 @xbhsbaseball 2028 grad

Lives! @PremierNewEng @_Candela12 @xbhsbaseball Fb: 78-81 Sl: 70-73

Indy ball pitcher Jean Calderon throws cut ride fastballs 95-97 with a power slider and changeup that consistently gets below the zero line

@drivelinekyle Keep the technical stuff coming!

Some footnotes: - Extrinsic calibration leading to world coordinate work requires utilizing Epipolar geometry, something that no one actually understands - Yes, this is a custom chessboard I had commissioned by my friend at FastSigns who does great work - The lenses used for the Edgertronic tend to have much lower distortion due to sensor size and lens type compared to the more common C/CS lens mounts, so calibration is less taxing - How a single camera's intrinsic calibration measurements are useful for a 2d reprojection on to 3d positions GIVEN A KNOWN OBJECT SIZE is super cool and beyond the scope of this tweet thread (you have to make assumptions on the object for a single camera) I am making it a thing in 2026 to write longer technical stuff on social media, the Driveline blog, and my personal website. Sadly, I don't think it will hold much value for our business and I unfortunately believe engagement will be much lower than it used to be back in the day when we had more independent analysts/coaches looking into stuff like this. The majority of people who will find value in these writings are people working in professional baseball or other organizations where employees are forbidden from commenting and discussing technical work openly, which is annoying. I get plenty of DMs and texts from people privately when I write stuff like this, but that isn't helping newcomers to the field see open discussion - to say nothing of the fact that poor engagement on platforms like Twitter/X limit out of sample distribution of my content. Still, I want to write more to simply continue to write more, so I'll be doing that for personal enjoyment and improvement - keep an eye out for more things like "Pitch Design as an Optimization Problem" that has a ton of bullshit graphics about Pareto Frontier and boring stuff from your ECN 101-302 courses you slept through. Also before @clayton_t22 can get me with the Pangram meme:

My user icon is me holding a giant chessboard with lines drawn on it. What's that all about? 🤔 You need the chessboard to help teach a computer one or two things: - How much lens distortion there is - Where each camera is relative to the other The first is called Intrinsic Calibration and the second is Extrinsic Calibration. For Pitch Design, you only have one camera, so Extrinsic Calibration / World Coordinate Modeling is not necessary - that's a whole different kettle of fish. Look at the attached images: Both frames detected all 77 corners on our calibration board perfectly. Same board, same camera, same lighting. But one scored a 98 and the other scored a 5 per OpenCV's grading. The major differences are: - Board tilt / alignment - Image clarity 📊 MORE ON CALIBRATION 📊 Intrinsic Calibration is how we teach the software exactly how this specific lens distorts the image. So when you hold the board flat and straight at the camera, OpenCV can find the corners fine, but it's basically looking at a flat grid with no depth information - and making things worse, the image is slightly blurry. You are making it easy for the computer vision model to detect the board, but the tradeoff is that you aren't giving it much novel information! Varying the angle, tilt, "waggle," depth, and position of the board is important to help the robustness of the calibration model. So the poorly graded frame doesn't help the model much, but we also wouldn't want to do without it. Straight-on frames are still useful, it's just that we need diversity in our sample along all axes - the poor grade is more of an indicator that this frame was both overrepresented in the sample and doesn't help the model on its own, so we don't need any more of this type of frame shot. 🪃WHY IT MATTERS 🪃 These calibrations feed directly into how we measure seam orientation using our fine-tuned CV models. We know exactly how big a baseball is - and the fact that it is a sphere, which is important for single-camera reprojection - so when we detect seam pixels in the image, we can accurately reproject them. But that ONLY works if we've corrected for how the lens bends light. Barrel distortion pushes edge pixels out of position (think skateboarding fisheye lens videos for extreme examples!), and lenses tend to have worse distortion at the corners. Intrinsic Calibration gives us the correction map. Without it, those seam pixels land in the wrong spot on the sphere, our reprojection is worse, and the latitude/longitude coordinates we use for Mollweide projections drift. That's the whole basis of seam-shifted wake analysis - if the seam orientation is off by a few degrees, measurement noise dominates our calculations. --- So if you're ever running a calibration session: Wave the board around. Tilt it, angle it, rotate it. Variability is important for a good calibration!

@drivelinekyle Multiple Claude’s open in multiple vs codes 🤣

Sadly:

Fallball! Morrisette Legion vs Leominster. Tuning up some pitches.