Jetson Nano review and Object Detection ft. TensorRT

Поділитися
Вставка
  • Опубліковано 16 лис 2024

КОМЕНТАРІ • 99

  • @ankursatpute9235
    @ankursatpute9235 5 років тому +86

    Can you do a tutorial on converting Tensorflow frozen graph into TensorRT graph?

  • @TheForbinone
    @TheForbinone 10 місяців тому

    As an ex teacher I must applaud your acumen. This is the stuff kids love to see and will inspire them to a career in electronics / AI / Computers ETC. I left in 2009 and now design multile seat VR simulators for theme parks. So no one is locked into any career.

  • @vinayentc
    @vinayentc 5 років тому +2

    Learnt tensorflow object detection from your tutorial...now learning it in action on nano....

  • @스머프-s7g
    @스머프-s7g 3 роки тому +1

    it should also be noted that you can get a smaller carrier board for nano. The carrier makes up the largest portion of it, the actual jetson nano chip is super small, like the size of laptop ram

  • @theappliedcoder9824
    @theappliedcoder9824 5 років тому +2

    sendex i followed you from my college days you make cool python videos and i learnt a lot from you

  • @shmarvdogg69420
    @shmarvdogg69420 5 років тому +34

    sentdex: you will need some sort of water cooling solution
    me: well there is a lake outside. *Throws board in lake*

    • @Shah_Khan
      @Shah_Khan 5 років тому

      This video released just now but you commented 2 days ago...how man?

    • @AnimatedHooman
      @AnimatedHooman 5 років тому

      @@Shah_Khan wtf, UA-cam is broken

    • @UD-tu1ue
      @UD-tu1ue 5 років тому +2

      @@Shah_Khan It's because of membership. People who pay get early access to his videos.

    • @shmarvdogg69420
      @shmarvdogg69420 5 років тому +1

      @@Shah_Khan Early access cause i joined the channel with the sponsor button

    • @Shah_Khan
      @Shah_Khan 5 років тому

      @@shmarvdogg69420Oo I get it. thanks

  • @rushirajparmar9602
    @rushirajparmar9602 5 років тому +3

    Thanks sentdex!!! A timely video when I needed some acquaintance with this!!!😅😅

  • @mockingbird3809
    @mockingbird3809 5 років тому +2

    Man......This is what I was waiting for. I bought this Jetson Nano just 6 Days Ago. OMG.....This is really cool. Can't wait for your upcoming videos

    • @sentdex
      @sentdex  5 років тому +1

      What would you like to see with it?

    • @matt-g-recovers
      @matt-g-recovers 5 років тому +1

      @@sentdex yes:)
      ... BTW loved the liquid joke. my custom liquid workstation is as big as an old 50"CRT 📺 ;)
      I really have no new ideas for this yet but love to watch

    • @gauravbansal1894
      @gauravbansal1894 5 років тому

      @@sentdex please please make videos on tensor_rt

  • @gabrielhhernandes
    @gabrielhhernandes 5 років тому +24

    "advanced temperature sensor" lol

  • @dr.jin.
    @dr.jin. 5 років тому +10

    Deepstream 4.0 with JetPack 4.2.1 enables 20 fps of uff converted ssd inception v2 in nano :)

    • @ikhtiyor
      @ikhtiyor 5 років тому

      how can I use deep stream with jetson nano? I heard that jetson nano support version is not out yet?

    • @dr.jin.
      @dr.jin. 5 років тому +2

      ​@@ikhtiyor You need to install JetPack 4.2.1 on Jetson Nano for use Deepstream 4.0.
      These were released last week.
      developer.nvidia.com/embedded/jetpack
      developer.nvidia.com/deepstream-sdk

  • @Hellhound_RedFox
    @Hellhound_RedFox 5 років тому +1

    Funny, I've been working with these at work just last week !

  • @esysss
    @esysss 5 років тому +3

    Very impressive, I really like to see the tutorial about how to write the code by tf to make a object detection app, thank you any way, you're the best

  • @mbczxaydbzrkmszj
    @mbczxaydbzrkmszj 5 років тому +1

    I liked the humor throughout the video

  • @AmlanPanigrahi
    @AmlanPanigrahi 5 років тому +1

    Advanced temp. sensor called 'thumb'...rofl hahahaha, you nailed it there!

  • @MrAvanatiker
    @MrAvanatiker 5 років тому +1

    Are there already commercial aviable Tensor Processing Units wich use ASIC based tensor computing? Google already have them but there are hold secret

  • @cheesypufs
    @cheesypufs 5 років тому

    Would love to see a Coral Edge TPU USB added to this setup and benchmarked.

  • @amalnazirah5034
    @amalnazirah5034 5 років тому +2

    I am quite confused between TensorFlow and tf-lite. I am training SSD Mobilenet and exporting it into pb file. Is that considered a tf-lite model? Apologies if the question seem ridiculous. Currently a new explorer on Deep Learning.
    Ps: Have been watching Sentdex videos for a while now and honestly, highly recommended. It is easier to understand and my interest in Ai is piqued.

  • @ayushthakur736
    @ayushthakur736 5 років тому +1

    Thanks for this amazing review video. I am a big sentdex follower. I have quick questions. Is tflite reducing the size(Mb) of the actual model? Is running yolov3-tiny on rpi3 a good idea? I tried multi-threading to increase fps but the results were not that great.

  • @joshuam2341
    @joshuam2341 5 років тому +1

    Hi man! Your channel is amazing! What software are you using to screen record and record your webcam as well?

  • @AmitYadav-ig8yt
    @AmitYadav-ig8yt 3 роки тому

    Hello sir... Please help me.
    I want to train a object detection model on image. But As output I want object name, Object size and Objects information (If I give something).
    Let's say want to train on a wind turbine blade image. I want defect type in blade(there are 4 types of defects),which blade this defect belongs to.
    How to label image for YOLO in this situation?

  • @AnimatedHooman
    @AnimatedHooman 5 років тому +2

    This is more powerful that some laptop ls out there

  • @badejavuade6774
    @badejavuade6774 5 років тому

    Next time please give some information on how to set it all up. Code and workflow. Thanks for the video

  • @leiverandres
    @leiverandres 5 років тому +1

    Have you try to optimize Faster R-CNN with TensorRT?

  • @San_OO7
    @San_OO7 5 років тому +1

    Really Advanced Temperature Sensor My Thumb ; )

  • @ikhtiyor
    @ikhtiyor 5 років тому

    it would be a huge help if you can make a tutorial using pure tensorrt for object detection.

  • @cx3268
    @cx3268 5 років тому +1

    Of these very small but high powered boards (cards), so many have a thermal problems
    OK, big heatsink and/or fan and be careful how you enclose it
    Been looking at this one, Pi4 and some others to build a new desktop computer and get away from Windows

    • @sentdex
      @sentdex  5 років тому

      I would do nano over pi4 personally.

  • @nitinrai6093
    @nitinrai6093 5 років тому

    Deepstream works very well on nano

  • @brentgreeff1115
    @brentgreeff1115 5 років тому +3

    I am not interesting in making robots or whatever, but I would like a dedicated machine to learn ML. Could this be used as a desktop for learning / basic experiments?

    • @sentdex
      @sentdex  5 років тому

      It could, yes

    • @bediosoro7786
      @bediosoro7786 5 років тому +3

      I don't think this is good for desktop replacement. There are cheaper options available with at least gtx1050ti which can provide more performance than this stuff. If you need desktop buy with gtx or rtx graphics card. But if you need drone applications or low power applications testing you can used this.

  • @CodeAbstract
    @CodeAbstract 5 років тому

    Hi @Sentdex, do you have any memory concerns with this board? Awesome video btw!

  • @carterbabin6867
    @carterbabin6867 5 років тому +1

    Amazing!!! I am going to get one!

  • @Argoo41
    @Argoo41 3 роки тому

    Can you share which model did you use here? smth like faster rcnn?

  • @maikdean8490
    @maikdean8490 4 роки тому

    What’s the Limit of megapixel camera on this board or it does not matter plz help ?

  • @ronvasserman
    @ronvasserman 5 років тому

    Even though it's insane how it packs so much power in such a small size, theres no point of wasting such a big portion of it for usb ports.. they could've fit more on it and you can buy hubs for more ports.

    • @sentdex
      @sentdex  5 років тому +1

      Well this is a dev board. If you want you can buy just the chip and implement exactly what you need

  •  5 років тому +2

    How can we convert own tensorflow keras model to optimize performance for jetson nano?It's important to be fast.

    • @sentdex
      @sentdex  5 років тому +5

      I will likely cover this at some point

  • @jar93jar
    @jar93jar 5 років тому

    Hello sentdex I was wonder if at any point you will have an updated version of the gta5ai with a tutorial on how to implement the pip through pycharm. I love the ai and hoping to implement into a story mode speedrun ai that learns from a speed run play to get better.

  • @girish7914
    @girish7914 5 років тому +1

    can you please create an AI app using Bert, voice to voice interface simple AI app (integrating all the components).

  • @minkangchew6693
    @minkangchew6693 4 роки тому

    Hi, do you have any tutorial for the tensorrt conversion? Thanksss

  • @da_bs94
    @da_bs94 5 років тому +1

    What OS did you use for the nano? Thanks!!

  • @jeremiahblessing3891
    @jeremiahblessing3891 5 років тому +1

    Loved it❤️❤️

  • @howardhughes7491
    @howardhughes7491 5 років тому

    Thank you!

  • @SandwichMitGurke
    @SandwichMitGurke 5 років тому +3

    Did I just hear gta 5 self driving? :)))

  • @shalsteven
    @shalsteven 5 років тому

    Where is the tutorial?

  • @mattizzle81
    @mattizzle81 5 років тому

    I bought two of these, a Google Edge TPU and a PI 4.
    Honestly they are just collecting dust. I suppose it is my personal preferences but I find developing ML on smartphones much more interesting and useful.
    Sure, my high end smartphone capable of 30fps object detection does not cost $99, but it has everything in it, screen, battery, more powerful, etc. It is what I have with me at all times, and I can build an app for the masses if I want to.
    I found these a waste of money. Maybe if I was into robotics etc it would be different

  • @chenjunsern5440
    @chenjunsern5440 5 років тому

    Does the Jetson support Faster-RCNN v2?

  • @riteshpatel1460
    @riteshpatel1460 4 роки тому

    you can improve fps by parallel processing on gpu

  • @EndermanAPM
    @EndermanAPM 5 років тому

    But... What is this used for? Just iot general stuff?

  • @dawah7556
    @dawah7556 5 років тому

    congrats for 700K subs..

  • @josephfouad3561
    @josephfouad3561 5 років тому

    Isn't TensorRt is made for TensorCore based cards? Maxwell GPU does not contain TensoreCores I think

  • @selfawaredevices
    @selfawaredevices 5 років тому

    it also has a #selfblow feature ;)

  • @sanjaybhatikar
    @sanjaybhatikar 2 роки тому

    Love it

  • @hzlin6655
    @hzlin6655 5 років тому

    It is faster rcnn inception v2? With this model i can just getting 3 FPS with my gtx 1060...using tensor rt is not helpful

  • @HarshGupta-de8ju
    @HarshGupta-de8ju 4 роки тому

    How can I run TensorRT on Jetson Nano, can you show a tutorial for the steps you took to get 9FPS

  • @knabbagluon
    @knabbagluon 5 років тому

    What is better the Nano or the Pi4? I don't know. Nano has the better GPU but the PI4 is better overall? I would love a comparison for the use cases of a small web server, wiki and Maschine lerning.

    • @sentdex
      @sentdex  5 років тому +2

      Nano is better or equal in all ways as far as I know.

    • @mattizzle81
      @mattizzle81 5 років тому

      @@sentdex I don't like the power supply on the Nano though. I had so many issues with random shutdowns etc. They should have went with usb-c

  • @JustThomas1
    @JustThomas1 5 років тому +1

    Can I train GPT-2 on it though?

  • @SajanKumar-mx2jg
    @SajanKumar-mx2jg 5 років тому

    Can we use use full fledge desktop version Linux system.

  • @sagarbhatnagar1047
    @sagarbhatnagar1047 5 років тому

    can you do a video on semantic segmentation or object detection

  • @OAelkadi
    @OAelkadi 5 років тому

    Can you do the same for Nvidia AGX Xavier , would it as fast as 30 frames per second with faster Rcnn . Also I should thank you a lot for your channel, you made me learn python and deep learning application by TensorFlow API in an easy way , keep going man ;)

  • @ahmedyossef9944
    @ahmedyossef9944 11 місяців тому

    Can i run carla simulator on nano ??

  • @sohaibarif2835
    @sohaibarif2835 5 років тому

    Looking forward to TensorRT. About the heat and FPS, you need to use a 4 V 5A powersupply for this. I used this model github.com/SteveMacenski/jetson_nano_detection_and_tracking with mobilenet v110 and also with a kalman to do tracking. I got 10 fps and it ran for about 3 hours outdoors(Detroit Maker Faire).

  • @tylersnard
    @tylersnard 5 років тому

    Where can I buy this "My Thumb?"

  • @bediosoro7786
    @bediosoro7786 5 років тому

    This board policy is bad. They claim 99$ however, you don't have sd card for storage and no 5vdc power supply accessories. You need to find the right by yourself separately. It could be better if they used on board storage like emmc and power supply included.

    • @sentdex
      @sentdex  5 років тому +1

      This is all commonplace when you buy a board like this.

    • @bediosoro7786
      @bediosoro7786 5 років тому

      @P A aha really. I did not know that

    • @aidos9753
      @aidos9753 5 років тому +1

      It's even worse when you don't live in America..

  • @rockyrivermushrooms529
    @rockyrivermushrooms529 5 років тому +3

    No ADC just like a pi? booo but thats okay because theres still arduino and adc chips.

  • @MaebhsUrbanity
    @MaebhsUrbanity 5 років тому +1

    but when cooled rasberypi can be overclocked.

  • @girish7914
    @girish7914 5 років тому +1

    can you please create model from scratch in DCGAN!!!!!!

  • @shmarvdogg69420
    @shmarvdogg69420 5 років тому

    gigabit internet is all i need

  • @girish7914
    @girish7914 5 років тому

    can you please create tutorials on pytorch!!!!

  • @pepehimovic3135
    @pepehimovic3135 5 років тому +1

    I like you.

  • @watcher7351
    @watcher7351 5 років тому

    hı. ı want to continue this series. nvidia most popular and lots of app. jetson nano both cheap and amazing. but little bit slow

  • @littlenewton6
    @littlenewton6 4 роки тому

    你的声音真性感。

  • @ayush0477
    @ayush0477 5 років тому

    compare it with your pc.

  • @Stinosko
    @Stinosko 5 років тому

    Cool 🧠! 😄

  • @matt-g-recovers
    @matt-g-recovers 5 років тому

    How does an object detection program like Darknet AI pjreddie.com/darknet/
    stack up to what your using here?
    I think he mentioned in his talk further optimization for mobile, which may be really well suited for this lil guy? I am really new to this too so please excuse any ignorance in my question...I've recently started your ML course via Bucky...who incidentally played a large role in my early language agnostic development.
    cheers!

  • @webaruhazberles
    @webaruhazberles 5 років тому

    Shall I start worrying about you, what is going on? where are my daily base videos, are tired? don't be! you are to young for that :)) We want more line of codes, projects from planing to finish.