Loading video...

Video Failed to Load

There was a problem loading this video. This could be due to a temporary network issue or the video might be unavailable.

i trained this computer vision model just by asking for "tennis players", using roboflow rapid and SAM3 the model picks out Alcaraz and Sinner, while correctly avoiding the ball boys and fans then I exported the model to python and wrote a script to filter the data, track movement... show more

AA

26,189 subscribers

293,837 views • 8 months ago •via X (Twitter)

Gaming Science & Technology Sports

Anya Rossi• Live Now

Private livecam show

0 Comments

No comments available

Comments from the original post will appear here

Related Videos

typing "player with a red shirt" was all it took to train this computer vision model just a few minutes to train the model, export to python, and create this video i'll keep testing roboflow rapid and will report back with the results in the meantime, enjoy the magic of Zlatan

typing "player with a red shirt" was all it took to train this computer vision model just a few minutes to train the model, export to python, and create this video i'll keep testing roboflow rapid and will report back with the results in the meantime, enjoy the magic of Zlatan

AA

239,299 views • 8 months ago

People counting in zones using Ultralytics YOLO26 🧑‍🤝‍🧑 In this demo, people are moving on elevators. I used a custom-trained YOLO26 model to detect and track each person, then applied the object counting in zones solution to count the number of people within each zone. 🙂 #people #retail #MachineLearning

People counting in zones using Ultralytics YOLO26 🧑‍🤝‍🧑 In this demo, people are moving on elevators. I used a custom-trained YOLO26 model to detect and track each person, then applied the object counting in zones solution to count the number of people within each zone. 🙂 #people #retail #MachineLearning

Muhammad Rizwan Munawar

14,980 views • 2 months ago

i'm building an AI soccer coach to improve my shooting using: - roboflow RF-DETR to detect the ball / net / player - mediapipe to record body angles when striking the ball - python to analyze and annotate the video i'll also try adding a VLM into the pipeline for coaching feedback

i'm building an AI soccer coach to improve my shooting using: - roboflow RF-DETR to detect the ball / net / player - mediapipe to record body angles when striking the ball - python to analyze and annotate the video i'll also try adding a VLM into the pipeline for coaching feedback

AA

306,568 views • 2 months ago

another experiment with fruit, computer vision, and projection mapping live object detections (roboflow RF-DETR model) are used to control the animated scene (threejs) i also added mediapipe segmentation to the pipeline to create collisions with my silhouette, but i haven't gotten a hang of it yet more experiments are underway

another experiment with fruit, computer vision, and projection mapping live object detections (roboflow RF-DETR model) are used to control the animated scene (threejs) i also added mediapipe segmentation to the pipeline to create collisions with my silhouette, but i haven't gotten a hang of it yet more experiments are underway

AA

21,581 views • 2 months ago

My 3D Model of Malevola from Dispatch is finally finished! This was a super fun project, I learned how to use hair curves for the first time and also 3D printed the full model! Have a look at the full process video! #blender #b3d #dispatch

My 3D Model of Malevola from Dispatch is finally finished! This was a super fun project, I learned how to use hair curves for the first time and also 3D printed the full model! Have a look at the full process video! #blender #b3d #dispatch

Niall

15,127 views • 6 months ago

Football players detection + tracking using Ultralytics YOLO26 ⚽️ It took me 30 minutes to train the YOLO26 model on a custom dataset using Ultralytics Platform. Later, I used the object tracker (Bytetrack) for tracking the players; tracking could be improved further, and I am working on it this week. More info 👇 #soccer #football #MachineLearning

Football players detection + tracking using Ultralytics YOLO26 ⚽️ It took me 30 minutes to train the YOLO26 model on a custom dataset using Ultralytics Platform. Later, I used the object tracker (Bytetrack) for tracking the players; tracking could be improved further, and I am working on it this week. More info 👇 #soccer #football #MachineLearning

Muhammad Rizwan Munawar

42,586 views • 3 months ago

#VTuber #Live2DWIP For the first time I tried using blendshapes to tweak the head angles with this model. Yes, I went a bit overboard, but it works, and I'm happy about that 😸

#VTuber #Live2DWIP For the first time I tried using blendshapes to tweak the head angles with this model. Yes, I went a bit overboard, but it works, and I'm happy about that 😸

TRedCat 😺 赤猫 / Live2D Rigger CM: open

11,498 views • 1 month ago

Probably the first of a kind... a Substrate node executing a Plutus script to validate an UTxO transaction. This is a project by TxPipe called Griffin, a Substrate-based framework for building app-chains using the eUTxO model and the Plutus language. Cardano devs can use it to build partnerchains that leverage their exiting expertise around UTxO programmability.

Probably the first of a kind... a Substrate node executing a Plutus script to validate an UTxO transaction. This is a project by TxPipe called Griffin, a Substrate-based framework for building app-chains using the eUTxO model and the Plutus language. Cardano devs can use it to build partnerchains that leverage their exiting expertise around UTxO programmability.

Santiago Carmuega

10,249 views • 1 year ago

CLIMATE EMERGENCY - This man is lying! The spokesman for Just Stop Oil states that the planet was the hottest it’s been in 100, 000 years! But this wasn’t a measurement. It was a computed average from a computer model! (Think Covid) And the data only goes back to 1940 🤡

CLIMATE EMERGENCY - This man is lying! The spokesman for Just Stop Oil states that the planet was the hottest it’s been in 100, 000 years! But this wasn’t a measurement. It was a computed average from a computer model! (Think Covid) And the data only goes back to 1940 🤡

Bernie

189,714 views • 2 years ago

This is the vision im trying to achieve - I need to be able to change the name - not just the title and pretending its the name :( I COULD use after effects but I am dumb and would rather not have to

This is the vision im trying to achieve - I need to be able to change the name - not just the title and pretending its the name :( I COULD use after effects but I am dumb and would rather not have to

Tiraesia(s) TF TG

84,840 views • 7 months ago

this is the BEST vision language model I have ever tried! Aria is a new model by Rhymes.AI: a 25.3B multimodal model that can take image/video inputs 🤩 They release the model with Apache-2.0 license and fine-tuning scripts as well 👏 I tested it extensively, keep reading to learn more 🧶

this is the BEST vision language model I have ever tried! Aria is a new model by Rhymes.AI: a 25.3B multimodal model that can take image/video inputs 🤩 They release the model with Apache-2.0 license and fine-tuning scripts as well 👏 I tested it extensively, keep reading to learn more 🧶

merve

176,125 views • 1 year ago

🍦:Ever since I saw this song enter the riser music library, I knew it could belong to absolutely no one else but nutdan. This song was literally written for him. Let me tell you, I actually went and had a private chat with the songwriter myself. I even heard the track before Nut did And the songwriter told me "Man, I just really feel like this track has to be Nut and Nut only." #TastyNUTDAN_MVReaction #TastyNUTDAN #hongshihoshi #LYKN

🍦:Ever since I saw this song enter the riser music library, I knew it could belong to absolutely no one else but nutdan. This song was literally written for him. Let me tell you, I actually went and had a private chat with the songwriter myself. I even heard the track before Nut did And the songwriter told me "Man, I just really feel like this track has to be Nut and Nut only." #TastyNUTDAN_MVReaction #TastyNUTDAN #hongshihoshi #LYKN

𝑹𝒖𝒎𝒊 🎀

35,730 views • 1 month ago

Went out to eat with the boys tonight and when the waitress asked for some ID when I ordered a margarita, this video was all I could think about bro I’m COOKED

Sensitive content

Went out to eat with the boys tonight and when the waitress asked for some ID when I ordered a margarita, this video was all I could think about bro I’m COOKED

👑ℂℍ𝔸𝕆𝕊𝕂𝕀𝔻👑

1,570,600 views • 1 year ago

here's the model I made for Porcelain Maid during RIGGED! The one and only podcast where the guest leaves with a free scuff model! Was really fun having you on! This took 1 hr and 43 minutes to make

here's the model I made for Porcelain Maid during RIGGED! The one and only podcast where the guest leaves with a free scuff model! Was really fun having you on! This took 1 hr and 43 minutes to make

Pipi Slayer🦇💀

39,264 views • 6 months ago

This is my "feel the AGI" moment: I used GPT-5.6 Sol to train my own autocorrect model that outperforms GPT-5.6 Sol (wtf??) I have no ML background. I have no idea what I'm doing. I just kept pushing Sol until it spat out a SOTA model. And I spent $0. The motivation: Years of talking to AI have made me terrible at typing. Rather than fix my skill issue, I decided to throw more AI at it. My idea was: instead of autocorrect that interrupts my flow, I want to type fast with mistakes and have AI clean it up after. I wanted the smallest local model possible, for speed, for battery life, for science! So I decided to train my own. Inspired by Andrej Karpathy’s autoresearch, I ran Codex /goal with this setup: pick an experiment, try it, record the results to a doc, throw it out if it fails, and plan the next experiment without repeating failures. I gave a few examples that had to pass, tight latency targets, and let it run. Sol did some amazing things. First, it scanned benchmarks and shortlisted base models: Qwen 3.5, Gemma 4, Liquid LFM 2.5. It found a dataset on HuggingFace for typed text. Then it built a simulator for fingers striking a Mac keyboard, modeling the physical layout with a Gaussian distribution around each key. It simulated striking the wrong key, wrong order, fat-fingering, etc. With the models + data + simulator, it fine-tuned using MLX right on my MacBook. It had a working prototype within an hour! But accuracy was pretty poor. — Problem 1: Tokenization Sol read papers, ran tests, and identified that the tokenizer was the bottleneck. Tokenization makes typos hard for the model to see, so it memorizes mappings instead of using its language priors. Sol tried ByT5, Google’s tokenizer-free byte-level LLM. This made a big improvement, but the model is old and lacked the knowledge needed to reach Sol performance. Sol dug deeper and realized a tokenizer-free model isn’t needed; instead, it used T5Gemma, an encoder-decoder model. This can understand the input deeply before producing output, and furthermore, Sol could post-train the encoder to improve performance. This gave a much higher ceiling. — Problem 2: Loss function Now the model was correcting some typos perfectly, but ignoring most. Sol realized that standard cross-entropy loss was teaching the model to avoid edits, because the vast majority of characters in the training data were left unmodified. The fix was wild: Sol wrote a custom loss function that byte-aligns the source and target strings, uses a dynamic programming algorithm to compute the minimum edits between the two, then weights correct edits much higher than copies. After a lot of tuning, this dramatically improved accuracy. — Problem 3: Autoregression One failure mode remained: if the model made a mistake, it couldn’t backtrack. It could only predict the next token. Teaching it to “think” like a reasoning model would solve this, but would be far too slow. Sol found a beautiful solution: instead of greedily predicting the next token, beam search over all possibilities. This parallelizes the exploration instead of one linear chain-of-thought. At the end, choose the path with highest cumulative log probability. This worked great, but made the experience worse, since the user wouldn’t see progress until the whole search was done. To fix this, Sol made a clever observation: after each search step, the longest common prefix among surviving branches is guaranteed to appear in the final result, so it can be displayed immediately. As the search progresses, weaker paths are dropped and the prefix grows, so the user sees continuous progress. Sol built all this as a custom MLX pipeline that does the parallel decoding on the MacBook GPU, with just ~40ms TTFT. It’s crazy fast and entirely local. — Final eval (error reduction rate, higher is better): - Apple autocorrect: 49.66% - GPT-5.6 Luna: 82.47% - GPT-5.6 Terra: 87.64% - GPT-5.6 Sol: 90.56% - Our model (1.7B): 91.02% Final cost: - 1 quota reset (thanks Tibo) - $0 (And yes, I verified there's no cheating. In fact, we test words scrubbed from the training data to prove the model isn’t memorizing) There were a ton more details and tangents I could write about: contrastive learning, GRPO, DPO, dynamic masking, and more. Sol is a fascinating and creative model. It blew my mind so many times. Don’t let a lack of experience stop you: Sol makes AI experiments accessible to anyone!

This is my "feel the AGI" moment: I used GPT-5.6 Sol to train my own autocorrect model that outperforms GPT-5.6 Sol (wtf??) I have no ML background. I have no idea what I'm doing. I just kept pushing Sol until it spat out a SOTA model. And I spent $0. The motivation: Years of talking to AI have made me terrible at typing. Rather than fix my skill issue, I decided to throw more AI at it. My idea was: instead of autocorrect that interrupts my flow, I want to type fast with mistakes and have AI clean it up after. I wanted the smallest local model possible, for speed, for battery life, for science! So I decided to train my own. Inspired by Andrej Karpathy’s autoresearch, I ran Codex /goal with this setup: pick an experiment, try it, record the results to a doc, throw it out if it fails, and plan the next experiment without repeating failures. I gave a few examples that had to pass, tight latency targets, and let it run. Sol did some amazing things. First, it scanned benchmarks and shortlisted base models: Qwen 3.5, Gemma 4, Liquid LFM 2.5. It found a dataset on HuggingFace for typed text. Then it built a simulator for fingers striking a Mac keyboard, modeling the physical layout with a Gaussian distribution around each key. It simulated striking the wrong key, wrong order, fat-fingering, etc. With the models + data + simulator, it fine-tuned using MLX right on my MacBook. It had a working prototype within an hour! But accuracy was pretty poor. — Problem 1: Tokenization Sol read papers, ran tests, and identified that the tokenizer was the bottleneck. Tokenization makes typos hard for the model to see, so it memorizes mappings instead of using its language priors. Sol tried ByT5, Google’s tokenizer-free byte-level LLM. This made a big improvement, but the model is old and lacked the knowledge needed to reach Sol performance. Sol dug deeper and realized a tokenizer-free model isn’t needed; instead, it used T5Gemma, an encoder-decoder model. This can understand the input deeply before producing output, and furthermore, Sol could post-train the encoder to improve performance. This gave a much higher ceiling. — Problem 2: Loss function Now the model was correcting some typos perfectly, but ignoring most. Sol realized that standard cross-entropy loss was teaching the model to avoid edits, because the vast majority of characters in the training data were left unmodified. The fix was wild: Sol wrote a custom loss function that byte-aligns the source and target strings, uses a dynamic programming algorithm to compute the minimum edits between the two, then weights correct edits much higher than copies. After a lot of tuning, this dramatically improved accuracy. — Problem 3: Autoregression One failure mode remained: if the model made a mistake, it couldn’t backtrack. It could only predict the next token. Teaching it to “think” like a reasoning model would solve this, but would be far too slow. Sol found a beautiful solution: instead of greedily predicting the next token, beam search over all possibilities. This parallelizes the exploration instead of one linear chain-of-thought. At the end, choose the path with highest cumulative log probability. This worked great, but made the experience worse, since the user wouldn’t see progress until the whole search was done. To fix this, Sol made a clever observation: after each search step, the longest common prefix among surviving branches is guaranteed to appear in the final result, so it can be displayed immediately. As the search progresses, weaker paths are dropped and the prefix grows, so the user sees continuous progress. Sol built all this as a custom MLX pipeline that does the parallel decoding on the MacBook GPU, with just ~40ms TTFT. It’s crazy fast and entirely local. — Final eval (error reduction rate, higher is better): - Apple autocorrect: 49.66% - GPT-5.6 Luna: 82.47% - GPT-5.6 Terra: 87.64% - GPT-5.6 Sol: 90.56% - Our model (1.7B): 91.02% Final cost: - 1 quota reset (thanks Tibo) - $0 (And yes, I verified there's no cheating. In fact, we test words scrubbed from the training data to prove the model isn’t memorizing) There were a ton more details and tangents I could write about: contrastive learning, GRPO, DPO, dynamic masking, and more. Sol is a fascinating and creative model. It blew my mind so many times. Don’t let a lack of experience stop you: Sol makes AI experiments accessible to anyone!

Anshu

178,934 views • 1 month ago

And here's the actual model, just wanted to test what I could get done from scratch in a couple days

And here's the actual model, just wanted to test what I could get done from scratch in a couple days

kkoa but 3d?

11,442 views • 1 year ago

⚽️👇 Activity Instructions Below 🎨 Animation Created Using The Tactics App In this unopposed finishing drill, four players are positioned in a diamond shape on the edge of the penalty area. The drill focuses on quick passing combinations and movement to create shooting opportunities. Players work through various combinations, ensuring that all four players are involved in the build-up before a shot on goal. The ball is played between the players in different sequences, such as one-touch passes, wall passes, or quick give-and-go’s. After completing the combination, the player receiving the final ball takes a shot at goal. This drill emphasizes precise passing, movement off the ball, and clinical finishing.

⚽️👇 Activity Instructions Below 🎨 Animation Created Using The Tactics App In this unopposed finishing drill, four players are positioned in a diamond shape on the edge of the penalty area. The drill focuses on quick passing combinations and movement to create shooting opportunities. Players work through various combinations, ensuring that all four players are involved in the build-up before a shot on goal. The ball is played between the players in different sequences, such as one-touch passes, wall passes, or quick give-and-go’s. After completing the combination, the player receiving the final ball takes a shot at goal. This drill emphasizes precise passing, movement off the ball, and clinical finishing.

The Coaches Zone

21,707 views • 1 month ago

I really think that this smoke could be giga useful to help save your B anchors utility for longer and I think the A player could reliably throw it and not be too late if you have a highway player by default(he'd hear the rush)

I really think that this smoke could be giga useful to help save your B anchors utility for longer and I think the A player could reliably throw it and not be too late if you have a highway player by default(he'd hear the rush)

mind

29,048 views • 3 months ago