Pinned
Amar Singh
792 posts
Amar Singh
@AmarSVS
In honor of the playoffs, I’d like to showcase what we’ve been working on here at Nexavision — a new way to generate basketball analytics through tracking with computer vision and AI: 🧵
00:00- Proud to announce winning the Sports Tech hackathon at @fdotinc! Came up with a novel way to track players on a NBA court with any video angle, excited to see its applications!
00:00 - Replying to @AmarSVSWe’ve built a system that can take in any basketball video feed to then track and identify all the players on the court. Most importantly, we calculate probabilities of making shots based on where the players are at, their movements, who the players are, and all the defenders.
00:00 - Replying to @AmarSVSCurious to check out all this? Come visit us at nexaodds.com to check out our stats! We are covering the rest of the NBA playoffs, WNBA coming soon, and we’ll get the NCAA back in the winter.
- Replying to @AmarSVSWe’ve taken this data and put it up on our site at NexaOdds so you can view all the shots taken after a game, their probabilities, and a whole bunch of other data on different possessions.
- Replying to @adithya_s_k1. KL-constrained SFT 2. If possible, fine tune using a reverse KL-based loss if you are doing a distillation 3. RL based methods are typically less efficient but better at preventing model collapse as well 4. Original data mixture like you said, nvidia released nemotrons
- Replying to @AmarSVSWe also have dashboards that show the expected results of games, how that would impact the game odds, player stats, and whole bunch more.
- Replying to @AmarSVSThis means we can look and grade each shot that a player makes on the court. Our shot probability model was trained with >2M shots.
00:00 - Are you a full stack dev interested in building player tracking software for basketball? Want to work with different collegiate+professional teams and help us build the front end for a new way of generating analytics? Send me a DM, we are looking for a cracked builder.
- I tested out the new Wan2.1 First Last Frame interpolation model and I compared it against Pikaframes (what I would consider current SOTA frame interpolation). I'm decently surprised - results in thread 🧵
- Looking for a new way to quantify player performance in basketball? Look no longer: nexavision.aiWe're excited to officially welcome Nexavision to the family 🚀 They're changing the way we track and understand basketball. Drop a 👋 to welcome @AmarSVS!
00:00 - It was a pleasure building this model -- beat top models like Claude 4 with only 12B on video annotation.Meet ClipTagger-12b. A new video annotation model built with Grass’s real-world video data, trained and deployed by @inference_net on their distributed compute network. It delivers high accuracy video labeling at a fraction of the cost, and it’s live today. Read more:
- This took a lot of trial and error to get right, particularly to train the long context summarizing models. The golden model ended up being hybrid attention, and actually unlocked the ability to process the 100M papers we will release soonWe're introducing Project AELLA, in partnership with @laion_ai & @Wyndlabs_ai AELLA is an open-science initiative to make scientific research accessible via structured summaries created by LLMs Available now: - Dataset of 100K summaries - 2 fine-tuned LLMs - 3d visualizer 👇
00:00
