Have something to say?
Optional priority flag on partner video uploads
Please add an optional priority parameter to the partner upload API (for example on add_video_by_url). The partner uses it to say how soon PB Vision should start to process a video. Suggested values, as a target for the start of processing: 4h: start within 4 hours 8h: start within 8 hours 12h: start within 12 hours If the field is absent, PB Vision uses its current default. Existing integrations do not change. We treat the value as a target, not a guarantee. Why we need it: partners such as clubs and coaches upload video with their own API key, and the urgency differs by video. A coach who records a lesson wants the analysis before the next lesson. A club that uploads a full day of games for a monthly review can wait. Today, PB Vision cannot tell these two cases apart. Why it helps PB Vision: you see the real urgency of each video. You can handle urgent videos at peak hours and run the rest when capacity is free. A partner that accepts a 12-hour start gives you more room to schedule. Items to consider: You may want to price faster start times higher, or give a discount for slower ones. We are open to either. Show the requested priority and the expected start time in the video status. Reject an unknown value with a clear error. Let a partner change the priority of a video that has not started yet.
Let partners tag and untag players on games in their Library after upload
Today, players are tagged on a game when it is uploaded. We ask for a way to add, change, and remove player tags on a game that is already in the Partners Library. Use cases A player wants to untag themselves. A player can ask to be removed from a game after upload. The player’s details were wrong or unknown at upload. The partner did not know the exact UID, name or email of a player at upload time. The partner needs to add or correct the player later, when the details are known. A player was tagged by mistake. The partner needs to move a tag from one player to another. Requested capability Add a player tag to a position in an existing game, by uid, name or email. Change which player is tagged at a position. Remove a player tag from a position. Details we ask you to consider Who can change a tag. The partner that uploaded the game. Ideally, a player can also remove their own tag. Webhook or API signal. Send an event when a tag changes, so that a partner system can update its own records or have a lightweight endpoint that gives some indicator like version or last_updated field to indicate when tagging, folder movement, video reprocessing occurred, etc. This would allow us to identify when we need to potentially refetch the details of the video. Why it matters Player consent and correct attribution both depend on this. Today, a wrong or unwanted tag stays on a game permanently. Tag edits after upload let partners respect a player’s request quickly and keep the data correct. Priority: High for partners that handle player privacy requests.
Add a paddle wearable
This would be a small, lightweight strap that sits on the throat of your paddle with a sensor that touches the face. Must be very light for minimal effect on your swing. The sensor contains a 6-axis IMU to measure paddle angle, acceleration and paddle rotation. Used to diagnose swing stroke issues.
Incorporate audio as positional input
This one’s a little out there, but it comes from a player who got feedback from his coach that he’s short-hopping basically all his shots. I immediately started thinking how to track his progress towards taking more balls at their apex or falling. Detecting a tight short-hop is too hard for the CV. I see it misclassify them as volleys all the time. But they have a distinct sound profile - that double-strike of the court then your paddle in close succession. I could track his progress by the time delta between those two amplitude spikes. Knowing bounce and strike times by sound could greatly improve those positional metrics. In practice, this won’t be easy. In my recordings, there’s a cacophony of strikes from nearby courts. You have to combine it with visual info, so you only pay attention to audio very near the visual strike point. And the amplitude spike of the court is barely detectable to my ear, and not at all on the far court. It might require additional frequency filters on that small window around the visual bounce point to extract ball noise. And there’s the speed of sound issue, but I imagine it’s either negligible, or we can measure distance from camera to the visual strike point to compensate for it. The tighter we get positional tolerances the better, IMO. Each level opens up a new class of stat we can track.
3D re-rendering of the rally
While the current video player is good to review from the camera angles for the real world, I would like a standardized 3d view, similar to the 2D coaching board that most coaches use. The 3d view can help illustrations so much better! The request is to re-render a match (or specific rally) in real time: with an illustrative court & net, and cartooned players and live balls (Hey PB vision can sell ads for the balls and the net as well, e.g Selkirk Net and Lifetime Balls). This can greatly help us to discuss shot selection and placement at the moment. Example of tennis in a game: 3rd person perspective is appreciated. Although 1st person can also be considered.
Keep ball tracking raw in cv.json
The ball trajectories in cv.json are fitted points on a parabola. I’d prefer just the raw frames with a confidence scale. This will allow me to fit whatever curve I want. Maybe I want to model bends and corkscrews. A parabola only exists in a vacuum. Right now, I’m modeling spin, drag and magnus. With 30 samples per second, I wouldn’t need to do that at all. It would be built into the observations. It would also be a cleaner boundary between the artifacts. CV is raw, insights is interpreted.
Add camera calibration to insights
CV.json has camera calibration to adjust for perspective. It also identifies near court from far court. That bit is gold. Can that migrate to insights? The issue is that near court positional data is far superior to far court. Understandable. Everything is very compressed over there. The best margin of error I’ve achieved on far court is 4.3 feet, which is fine for getting a general idea where you are on the court, but there are metrics I want to track that require more precision. You get that on near court. With a set of games I can say “only take positional data when the player is on the near court.” Then I can trust that data much more.
Add height of player to insights.json
We have player location in x,y. Can we get top of head in z-axis? I need this to determine if the volley is an overhead. Tall players may be taking this as a block volley at the chest and short players are taking an overhead. We know the ball position in x,y,z. An x,y,z player position opens up a lot of possibilities.
AI Match Commentary
Bring your pickleball matches to life with an optional AI voice commentator. Using shot and rally data, PB Vision could generate play-by-play commentary synced to your video—calling out big winners, key mistakes, and pivotal moments. Imagine watching your match with commentary like: “Alex spots the opening and puts the volley away!” “Jordan’s drop forces a low contact, and the reply catches the net.” “A patient dink exchange—and Sam breaks the pattern with a well-timed attack!” Choose from original commentator voices and styles, from energetic play-by-play to thoughtful analysis. Make match reviews more engaging and give shared highlights a broadcast feel.
Game Recording Interuptions
It seems like whenever I am recording a game and get a text or a call, the game recording is interupted\halted. Does anyone else experience this? Is there a way to prevent?
Dreambreaker support
Is there a way to tag multiple people in a singles dreambreaker? Right now the video is only registering 2 players and there were 8 in the dreambreaker.
Deepseek AI- 3rd shot drop vs 3rd shot drive.
Deepseek is having trouble differentiating between 3rd shot drop vs drive to give analysis. It does give feedback that it shouldnt be too hard. But the data labeling makes it unreliable. I think 3rd drop vs drive would be a very effect statistic. Also if there was a way to correlate 3rd shot drive when other team is not arriving at the kitchen could be informative as well.
List of AI prompts
Do you have a list of different AI prompts to ask for the best advice and analytics? I am unsure how to word all the different things that I might want to know. I figure you guys would have the best ideas for prompts.
Export recorded time to insights.json
It would be very useful to have the video recorded date/time in the insights.json. Then I could simply ask chat to compare the games I played on 6/8 with the ones on 7/12, for example.
Save Clips on Desktop
Option needed to save clips and download to desktop. Currently only available in the mobile app. Phone not always available, less storage space, and editing software is on desktop.
Right Side vs Left Side (Points Won / %) - Stronger Wing?
At the end of each point, track which side the player was on +1. Perhaps filter this further - plays won with forced errors caused by the player. Purpose: figure out which side is the “stronger side”. Helpful to know when comparing across multiple partners.
True KAP Computation
Many people like Anna Bright and RealClearStats take a much stricter view of what Kitchen Arrival Percentage is, at least for Serving KAP which is the most important. Roughly KA Success is both servers arriving at the line STABLE, or winning the point outright on 3/5/7 shot on the way in. PBV’s current Kitchen Arrival metric is far off - it often shows rallies where a serve team member doesn’t even get in. At high levels, this is one of the most important stats: in fact, Ben at RCS has data showing a clear correlation between DUPRs and KAP. What would it take to get a much more useful KAP metric? I’m happy to expound further and to even provide examples.
GoPro and Cloud Uploads
We need compatibility with other camera systems like using the GoPro livestream feature to directly record and upload games and at minimum have the ability to connect Cloud storage services to upload (GoPro cloud, Google Drive, Dropbox, etc). I run out of local storage way too often just trying to upload my games, And it would be even worse if I was trying to do 4K 60 FPS.
Additional Stat- Targeting/Icing- How many balls were hit to each player?
I would like to see stats showing how many times each player hits to each player or player area across the net. This would be helpful. I get annoyed when the other team accuses our team of targeting OR icing out a particular player. In rec play we should be more generous/forgiving. In tournament/league play we should absolutely go to the optimal shot placement regardless of who that player is. Most times, when accused, I really dont feel like targeting/icing was happening. IF it actually happened, it was completely subconscious. My subconscious mind just sees the court/players and suggests the better shot. The “right shot” just happened to be to a player that was out of position or just more vulnerable, regardless of who the player actually was. Conversely, am I the player being iced out or targeted? How can I support a player being targeted? How can I get more involved if I am being iced out? How can I ask my partner for support if I am being targeted? What are the partner strategies?
Using a back button
One of the most frustrating things I see (it's small I know) is when I click on Library and pick a game then click on either shot explorer, game stats, or team stats and I utilize the swipe back feature on my phone it brings me back to the main library screen instead of that game that k had clicked on previously. At the top where it shows Library > game name > > shot explorer (or any of the areas), I'm unable to click on game name and I always have to go back to the main library. Can there be a dedicated back button or be able to utilize the swipe back (Pixel 10) and go back one page instead of always back to the main library?