My
#TensorFlow
Lite on the
#RaspberryPi
video is out! It shows how to use TFLite models for real-time object detection on the Pi. I've been getting a TON of requests for this video, and I'm thrilled to finally have it done 🥳! Here's a preview! Full video:
@NickPetraccione
Took racquetball class from this guy at
@montana_tech
! Some interview advice he gave us during Career Fair: "Tell the interviewer they can fire their two best guys and you'll cover for them no problem!"
Just had a major "HELL YES!! 🤘👍🙌" jump-for-joy moment where I finally got a difficult piece of
#Python
#ComputerVision
code I'm developing to work. These are the moments that drive my passion for tackling challenging projects.
#Programming
can be really fun sometimes!
I made a
@Raspberry_Pi
HAT layout template for
@kicad_pcb
! Use this template if you're interested in making your own
#RaspberryPi
add-on. These pictures show the template and how it fits on a Pi 3 (using a perfect purple
@oshpark
PCB). Link to template:
It's been awesome working with
@tryolabs
on this project! One thing that really impressed me was the multi-threaded
#Python
application coded by their rockstar engineer. Makes for a great example of how to implement multiple processes for parallelization:
#MaskCam
: Real-time AI on the edge 📹
As part of the
@edgeaivision
, we partnered with
@bdti
@nvidia
&
@Jabil
to create a product capable of running a real-time face mask detection model on an embedded device.
Check this open-source project 👇
Oh man. That lovely feeling when you write 50+ lines of code without testing it, and then it runs without errors on the first go! (This almost never happens to me 😅)
#Python
Hey
@dansitu
, I'm researching applications of tiny edge-based machine learning models for a project I'm working on. What are some of the most interesting, practical, non-gimmicky MCU-based ML application you've seen so far? 😎👍
#MachineLearning
#AIoT
Very proud of my mom, who has officially won the primary election as candidate for Lieutenant Governor of Montana on
@GregForMontana
’s ticket. She’s one of the most intelligent, hardworking, and caring people I know, and I support her 100%!
New
#deeplearning
video is out! It shows how to train a
#TensorFlow
Lite detection model inside Google Colab:
✅ Import data
✅ Train model using free GPUs
✅ Test on images inside your web browser
✅ Calculate model mAP
✅ Export model to
#RaspberryPi
How do y'all feel about being explicit with boolean variables in
#Python
to make code more readable?
if (flag == True):
print(message)
versus
if flag:
print(message)
Both will do the same thing, but the first option is more readable for newcomers. Agree/disagree?
I always smile when I search for an OpenCV-related topic and see a
@pyimagesearch
article as one of the top results. I think, "learning this is going to be 10x easier now!" Thanks Adrian for making so many useful guides on
#OpenCV
and
#ComputerVision
!
MaskCam is now a featured project on
@Hacksterio
! If you've got a spare Jetson Nano lying around, it only takes about an hour to set up. Go try it out!
#ComputerVision
#AIoT
@ryandbusse
Hi Ryan this sounds interesting! Can you please point me to the specific law that Greg signed that caused this tax increase? Or
@raphgraybill
Didn't think I'd be spending all Saturday trying to build YOLOv4 on a client's PC, but after 6 hours of bashing my head against build issues, here we are. At least I finally got it working and can walk away victorious! 👍
#deeplearning
#YOLO
#consulting
New video is out! It's a
#RaspberryPi
3 vs Pi 4 performance comparison that shows framerate when running realtime
#MachineLearning
detection models with
#TensorFlow
, TensorFlow Lite, and the Coral USB Accelerator.
@Raspberry_Pi
4 + USB Accelerator = FAST!
Very proud of my mom, the new Lieutenant Governor-elect of Montana! She will do a great job guiding Montana, increasing its prosperity, and protecting its status as the Last Best Place.
#mtpol
#mtgov
I made a tool to help speed up image labeling for
#MachineLearning
! I call it "Supervised Automated Labeling". 1) Train a model using only 100 images or so, 2) Use it to automatically label the rest of the images, 3) Go through images and accept or decline each label.
I've been working with
@bdti
and
@tryolabs
on a reference design for a smart mask detection camera based around the
#NVIDIA
#JetsonNano
, and we just open-sourced it today! On the Nano, we can run YOLOv4-tiny at up to 14FPS. Learn more in this video:
I just crossed 10K subscribers on
#YouTube
! Thank you so much to everyone who has subscribed, liked, left comments, and otherwise supported my content. Creating tutorials that democratize access to technology has been a very rewarding undertaking. Next stop: 100K subscribers! 😺
Hey
@petewarden
, when will
@TensorFlow
v2.1 be released on
@piwheels
? The latest update to libedgetpu breaks compatibility with TF v2.0. libedgetpu is needed for using the Coral Accelerator. TF v2.1 isn't (easily) available on the Pi.
I made a simple
#Python
video player using Tkinter and
#OpenCV
! It gives the user control to move to individual frames and pause on them. Working on this for a small portion of a client project. Hoping to make a tutorial video on using Tkinter at some point!
I posted a YouTube video today, after 364 days of no videos! Couldn't let it get to one full year 😅.
I'm finally getting my legs underneath me in self employment and having more time to make videos. Planning to post two more this month!
Working on a little demo for an upcoming YouTube video. Trained a "change counter" model on a custom dataset of 750 coin images. Runs pretty dang fast!
#MachineLearning
Framework:
#TensorFlow
Lite
Model: ssd-mobilenet-v2 (quantized)
Hardware: Raspberry Pi 4 8GB + Coral Edge TPU
Working from the one coffee shop in little White Sulphur Springs, Montana. Doing
#MachineLearning
development and training with 1) my laptop, 2) remotely connected desktop PC, 3) an
#AWS
cloud computer. All you need is an internet connection 🤓
YOLOv4 has been released! Paper says it has 10% better accuracy and 12% better speed than YOLOv3. I'm excited to try it out!
#ComputerVision
#DeepLearning
#YOLO
EJ Technology Consultants has a logo! 😎👍
I'm ecstatic with how it turned out. It feels great to finally have a logo I like after spending so much time and energy on approaches that didn't work. On to the website next!
#logo
#branding
I learned how to use laser cutters recently and I'm kind of obsessed! Here are some
#edgeAI
camera enclosures I made that contain:
📷 IR camera
💻
#RaspberryPi
🔋 8000mAh Battery pack
Never going back to
#3Dprinting
again, sorry!
I made a card-counting
#AI
called RAIN MAN 2.0! It uses a YOLO v3 model trained on 50,000 synthetically generated images of playing cards, and it can count its way through a deck of cards very accurately. Check out my new video!
#MachineLearning
#blackjack
Here's a side-by-side comparison of my YOLOv3 card detection model vs a
@TensorFlow
Faster-RCNN model. Trained on the same 50,000 image dataset, running on the same PC. YOLO is faster and more accurate! From my new video:
#YOLO
#MachineLearning
#TensorFlow
Dang,
@adafruit
is cool! I Google something like “
#RaspberryPi
compass” and the first result is an Adafruit module that has clear setup instructions and code libraries. Found like 3 other random modules from them this week. They make prototyping too easy!
The first part of my
#TensorFlow
Lite object detection guide is finished! It gives step-by-step instructions on how to train, convert, and run a TF Lite model. Video series coming soon. Check it out!
#MachineLearning
#DeepLearning
#RaspberryPi
Hey
@petewarden
, not sure how involved you are in
#TensorFlow
Lite object detection anymore, but is the TensorFlow team working on making TFLite compatible with EfficientDet? I want to try it out on my
#RaspberryPi
!
#MachineLearning
I just published my (free) 81-page guide on learning
#ComputerVision
,
#DeepLearning
, and
#OpenCV
! 🚀🔥
Includes step-by-step instructions on:
- Getting Started
- Face Applications
- Object Detection
- OCR
- Embedded/IoT
- ...and more!
Check it out here: 👍
I'll be giving another talk this year at Embedded Vision Summit 2021! Creating a training dataset is quickly becoming the most time-intensive task in training a
#MachineLearning
model. In this talk, I'll present ways to save time building image datasets.
Learn strategies for quickly building a new dataset for training an object detection model and review tools and methods for speeding up the process of curating and labeling images with Evan Juras! Save 10% using SUMMITSOCIAL21 to see him and 80+ others.
Just donated to
@montana_tech
's wonderful Electrical Engineering program! A small repayment for landing me a well-paying engineering job right out of school with zero dollars in student debt.
#BestValueSchool
Want to use your
#RaspberryPi
to run
#TensorFlow
#DeepLearning
models at BLAZING FAST speeds? I just finished my video showing how to set up the Coral USB Accelerator with the Pi to speed up TensorFlow Lite detection models. Check it out here!🐯🐯
Here’s a closeup of my Supercon Super Speaker Badge Hack for the
@hackaday
badge hacking competition! This 5-watt puppy blasted its way into winning the Music Hack competition category.
Just finished recording my
#YouTube
video on setting up
#TensorFlow
Lite for object detection on the
#RaspberryPi
! Will be editing it tomorrow and posting it on Monday! Woohoo 🎉🥳😺
Whew! Finally finished the second draft of my slides for my Embedded Vision Summit presentation. I have been entirely too busy the past couple weeks, so it's nice to have that off my plate for a bit!
@estefanniegg
@RobThez
Very true... I feel the same way. But I think it would have a net positive impact. So many of us are clueless as to what women are experiencing. Seeing real examples is eye-opening, and it might help more people realize how extreme the harassment is.
This work-from-home isolation has taught me two things:
1. I'm much less of an introvert than I thought I was 🙀
2. I'm not nearly disciplined enough to be self-employed😅
Item 1 is a pleasant surprise, item 2 is something I need to work on!
#Quarantine
After a full day of struggling, I finally figured out how to use PHP to upload Picamera pictures from my
#RaspberryPi
to a local Apache server... and I only had 3 or 4 mental breakdowns in the process! 😫 Added to list of "things I would make a
#YouTube
vid about if I had time"😅
Check out this cool Kickstarter for a Raspberry Pi based AI Camera called Vizy, developed by
@CharmedLabs
! It implements many of the object detection and classification features I've discussed in my videos, plus lots more.
Currently working on a very cool computer vision project for a client. I’m still in disbelief that I’m getting paid to work on something so FUN! I think my day job is going to be in the rear-view mirror soon. 😁😁
#ComputerVision
#Python
Yay!! My
#youtube
channel has officially hit 5,000 subs! 🥳🍾🎉 I'm so grateful to all my viewers and subscribers. Thanks for coming along with me on this journey! I've got lots more
#raspberrypi
and
#machinelearning
videos in the works, so stay tuned for more!
Was talking to a coworker today about how you can claim a tax write-off for the aging of expensive capital equipment. Except I kept calling it “deprecation” instead of “depreciation”. So that’s how I know I’ve been working with
#TensorFlow
too much lately 😬
#programming
Man. Trying to future-proof my
#TensorFlow
guides so they're resistant to changes in future versions is tough. Once the video hits YouTube, I can't change any instructions in the video. But I'm going to try future-proof them anyway, dammit!
Having a couple weeks off means I get to make good progress on my blackjack robot! Currently trying to gauge just how far the camera needs to be above the table for it to see the dealer's cards and the player's cards.
The
#Python
script automatically saves an XML file with label data if accepted. Declined images are moved to an "unlabeled" folder to be manually labeled later.
Still a fairly manual process, but it cuts labeling time by 5X! Huge time savings if you have 1000s of images.
Hi
@NVIDIAEmbedded
&
@balena_io
, here's the super pleasant user experience I go through every time I plug my
#JetsonNano
SD card into my Windows PC to re-flash JetPack onto it. Any way to prevent all these windows from popping up? This doesn't happen with my
#RaspberryPi
SD card!
@TensorFlow
@raspberrypi
@petewarden
But- but- but I spent a whole month working on this 20-minute video showing you how to install
#TensorFlow
on the
#RaspberryPi
😢😢😢. Glad to see you guys have made it much easier! Next, you should make it easier to install protoc on the Pi!
interviewer: do you have any experience with version dependencies?
me: well, I've built
#TensorFlow
from source on Windows...
interview: say no more fam
Has anyone successfully trained a SSD-MobileNet object detection model and converted it to TensorFlow Lite format using TF v2.x and the TF Object Detection API? What TF version did you use?
#TensorFlow
#DeepLearning
#IoT
Here's the result from my latest attempt using TF v2.7 🙃
Hey
@ShawnHymel
! I want to try training
@EdgeImpulse
's new FOMO model with a custom dataset and running it on my OpenMV Cam. Where should I go to get started?🙂
Happy to announce that I'm leaving
#Seattle
and moving back to my home state of
#Montana
! I will be starting as a Senior Electronics Engineer at the
@mksinstruments
Bozeman office this April. I'm excited to
@MakeMontanaHome
and be a part of the state's growing tech industry!
My brother embarked today on a journey to row all the way from the headwaters of the
#Missouri
River, down the
#Mississippi
, and into
#NewOrleans
. I’m very proud of him! (And jealous.) I’m sure it will be an amazing adventure.
#Montana
A
#GitHub
user has quite the nefarious
#MachineLearning
-powered
#IoT
application idea. A
#RaspberryPi
smart camera that warns you if you're about to get parking ticket so you can move your car in time! Love it! 👿
@CyrChristina
Wow! Can't tell you how much his Electromagnetic Compatibility Engineering textbook helped me in my career. Pro tip: Before any HWE interview, read that textbook cover to cover... I guarantee 80% of the questions you'll be asked are covered in that book!
A weird combination of sleep deprivation, coffee, and wine has put me into a 2AM hyper-productive mood for some reason. Need to see if I can replicate this at the office tomorrow! 🍷☕️🥱
A wonderful
#GitHub
user put in the effort to make a full Korean translation of my
@TensorFlow
#ObjectDetection
guide! This is why
#opensource
is awesome 👍👍!! I will be adding a help request for translations in other languages too.
@chelscore
That's just your inner perfectionist trying to push you to be more perfect. Understand it, accept it as part of your being, and learn to co-exist with it.
At least.... that's how it is for me! 🤷♂️
One thing I've learned about writing
#RaspberryPi
guides: lots of people don't actually read the text (or dont understand English) and just go through copy+pasting the commands in their terminal. Could complain about it, but instead will focus on making my guides usable that way!
I've been wondering... how the hell does a tiny $75 Edge TPU run object detection models faster than my $300 GPU? It turns out
@googlecloud
has an excellent article explaining why the TPU is so FAST! It's all about the custom silicon, baby!
#DeepLearning