Hye thanks for all these videos. But can you please make a video about mili csi virtual channel and how can we use them explaining everything required

What would be relevant for creators is if there were a way to feed in, say, six Midjourney images of the same object, showing it from four sides plus top and bottom, and then have a quality 3D mesh generated from that. So far, Midjourney and others fail to provide six views of the same object. Any ideas?

paligemma же умеет детектить

@AntonMaltsev I have RK3576 2NPU and 8CPU and my camera fps is 30. I used that one but still not improved. Is it possible to combine cpu+npu? how?

internvl2 уже имела функционал детекции.

Thank you for video, can you please create video how to run Qwen2.5-VL on rockchip.

Very good video! Been watching you for a while now :)

Здраствуйте, извиняюсь что не по теме видео, просто только недавно наткнулся на ваш канал. Я начал интересоваться комптютерным зрением. Эта область мне нравится. Однако заметил, что возможно, может быть лучше войти в эту область со смежной специальности, скажем, обычного ПО-разработчика и тд? Мне интересно, как лучше поступить, чтобы наиболее спокойно войти в эту область на рынке, не оставшись без работы. Просто я подумал, что пороговый вход тут сожет быть большим. Что вы можете посоветовать мне по этому поводу?

LLM based AI is far from diagnosing my water filter issues. I need a water filter repairmen for that.

livebench ai benchmark is quite good.

AGI is already here. I'm not an influencer. I'm an engineer, working on agentic systems built on top of small (below 14B) LLMs. My systems are capable of general problem solving - simply by offloading all the reasoning to external tools (such as Prolog and SMT solvers). I do not believe we need LLMs to be able to reason on their own, we already have a much better and more precise reasoning engine.

чувак, продумай заранее что хочешь сказать, мысль скачет туда-сюда - трудно понят что ты вообще хочешь сказать.

Yes, the solution is leaking into datasets, but deepseek v2,5 in my local offline test was the only one which repaired very unusual code, all others qwen or mistral Large can't repair that logical problem. I'm testing locally now V3 (next their r1) and it's quite good, the only i need to raise from Q5 to Q6 because quality lacks on low quants. But even on Q5 it's trying hard to write me a Bach symphony in sound programming language. I've noticed how it's self tuned from dumping the all code in every answer, which is sign of dumb models like qwen or mistral, but now it's writing only parts to edit or replace, which saves tokens.

Anton, what do you think about AI camera "reCamera 2002" from seed studio?

how about performance?

Произносится "Тексас" а не техас )

Did you use rhe AI Pro 20 TOPs with 24gb or 8gb 8TOPs

Доброго времени суток Какое время предсказания для этой платы ( Inference) для yolo? Если это есть в видео сложно услышать

It might be worth looking into this board again! The documentation is a lot better, but still isn't the best... Community documentation being split among two websites is the biggest thing for me. I've been able to get some amazing stuff working, like running small LLMs, Arm and RiscV concurrency, etc. Cross compiling is also a lot easier, still needs a little setup but now I can just run a custom GCC command and use it just like normal GCC. Both for building custom images, and for building code to run on the RiscV processor. Planning on getting Yolo running on here too for some drone footage processing, found this video while trying to find _anyone_ who's posted a video of its performance

AI is applied effectively in natural language, performs even better in coding, and excels further in games. A finite set of human rules enables refined statistics and probabilistic methods to function with reasonable accuracy. Reality is philosophically deterministic in terms of causality, but not in a strictly mathematical sense. The challenge for AIs in dealing with reality lies in its non-deterministic nature, a given state does not always lead to the same outcome. There is a lack of abstraction and symbolism, persistent issues in current models that cannot be solved merely through scaling or refinement.

Good video, hard to follow your speaking pattern. Feel free to shoot multiple takes man. Especially for a 3 minute video. Recommend getting a better microphone for these things or moving closer to your microphone

One of the few AI related videos not overly dramatic or critical, just showing the current limits. i like it

The real ones know computer scientist promised "AGI anyday now" in 1956... 1956! This isn't a dystopia...

Zed2i is the best

I'm interested for now in videogen models, text-video is always random, but image-video is more interesting. LTXVideo almost there, but finding perfect setting takes hours (in ComfyUi), Ruyi kinda works but only short 5 sec gifs and they can't make it faster (40 minutes to generate on 4070 Ti Su), Hunyan praised as best chinese video gen but i haven't tried images and it's most demanding in processing power & VRAM.

Может быть имеет смысл сделать два отдельных канала, один на русском, другой на английском? 😶

Спасибо! Жду новых выпусков!

Крутяк, спасибо! Послушал от корки до корки с интересом!

AI is overrated

Thank you man for the great tuto 🙏

This is true, everyone is stating that Yolo11 isn't a big deal. I'm performing many tests and will conclude how better it is. I have a RTX 3080 TI. There is only one reason why I'm working with YOLO11, it has extremely well documentation on how to get things running.

Огонь !

Интересно, а обработка сигнала с выделением каких-то паттернов и признаков тоже вероятно область lvm?.. и стоит ли применять существующие большие модели для этого, скажем запихнув эти модели куда нибудь в пайплайн обёртку вроде langchain? Или мб свою попробовать лучше запилить, если есть датасеты?..

Народ, с Новым годом!! Вы в Београде, судя по крышам?..☺️

Хорошо говориш на русском. Как будто родно ;))

Очень интересно!

Все таки лучше с микрофоном

Что вы думаете про китайский Проект Генезис?

Interesting video! I watch almost every video on your chanel. I wanted to ask for some advice: I’m planning to create a KZbin channel and wanted to know how to create timestamps for videos. Is it done manually or is there a way to automate the process?

Hi Anton. I tried running my converted RKNN model on my rk3588 board but am getting an error that the .RKNN model is either formatted wrong or is or corrupt. Can you plz send me your converted model ' yolov8n.rknn' so that I can check my inference.

Anton, have you tried to run inference on different NPU cores? (Im having an issue with it, inference works only while using core 0). Also, is there an option to run inference on all 3 cores?

what challanges we can achive with rk3588??

Thank you for sharing you helped me a lot

awesome)

Anton, living in the future :)

Two remarks. 1) In the absence of the GPU, the Rock5C-Lite is targeted for CLI applications. With the official Debian KDE-OS GUI, your CPU load can be up to 6x 85% when showing videos or RSTP streams on the screen. (Not too handy when it comes to vision tasks) 2) The popular Ubuntu OS of Johua-Riek doesn't support the RK3582 (yet).

where to buy can anyone please send some links to buy theser products

how can rk3582(4 tops) is faster than rk3588(6 tops)?

awesome)

КРАСАВА