Jeremy Blythe

@CurtisSzmania @honeycombio @evertzio @CurtisSzmania Glad you found it useful. Using the OpenTelemetry Semantic Conventions is great but making your own domain specific ones and generating code for your microservices to use is next level.

@honeycombio @jerbly @evertzio @jerbly Thanks for sharing your insights on Semantic Conventions. They are indeed very useful for microservices. I appreciate your efforts to make code instrumentation easier and more consistent.

Navigating complexities of namespaces in code instrumentation? 😵‍💫 Read @jerbly of @evertzio's strategies for implementing ✨ Semantic Conventions ✨ in your codebase! 🔓 Unlock the full potential of microservices with consistent attribute naming! go.hny.co/3U0xHnP

@timClicks We use it at evertz.io - here’s a blog post about our video observability which has Rust at it’s core: evertz.io/blog/2022-09-2…

Does anyone use Rust in production?

Anyone interested in working on this cool tech in my team, check out our careers page evertz.io/careers

@cyen Thanks so much for reading and taking the time on this thread. Feels like everyday someone says “I can’t imagine how we’d do this without Honeycomb” - Anyone interested in working on this cool tech check out our careers page evertz.io/careers

what a treat to get a glimpse into all that goes into ensuring streaming media is delivered reliably! ✨ my couch potato self thanks you :)

oooh - very cool blog post alert, from @jerbly of @evertzio on how their eng team ensures that a tv scheduler's *intent* is correctly playing out, the right way, on viewers' screens: evertz.io/blog/2022-09-2…

First few days off for the Holidays - made an On Air light using a @Raspberry_Pi, @home_assistant and a little bit of code github.com/jerbly/onair/b…. Whenever I use the microphone in @MicrosoftTeams, @zoom etc, the light comes on. Write-up here: jeremyblythe.blogspot.com/2021/12/raspbe…

@DynamicWebPaige @brettsky Slightly different playing cards, but here's an end-to-end project I did - labelling, training, inference, serverless deployment. Trained on AzureML and deployed as a static-web app backed by a function app on Azure. It's still up and running! jeremyblythe.blogspot.com/2021/02/practi…

@brettsky Oh!! There's a playing card dataset already on Kaggle: kaggle.com/luantm/playing… As well as four models doing precisely what you're after: kaggle.com/luantm/playing… Are you wanting to deploy this to something like a phone or a Raspberry Pi, or is laptop + webcam okay?

(1/4) evertz.io is excited to launch evertz.io-Stream, a revolutionary streaming and playout SaaS service delivered completely from the public cloud using the evertz.io platform. Learn more at evertz.io

IceVision v0.8 is out! - Support for latest @fastdotai version - Yolov5 by @ultralytics - Confusion metric for object detection - All models now support negative sampling (images used only as background) - Greatly improved autofixing speed Join us: discord.gg/JDBeZYK

Happy to announce 🎉IceVision 0.8 release🎉. We added some really cool features, and data auto-fixing speed is 🚀 Join the IceVisionaries and start building some cool AI/DL apps in your domain of expertise! Docs: airctic.com Forum: discord.gg/JDBeZYK

🎉Introducing Our 4 New Core-Developers🎉 We are so proud and lucky to have such talented contributors joining our IceVision CoreDevTeam. Very grateful for their help in shaping up our recent version 0.7 airctic.com @jerbly @rsomani95 @potipotrykus @adam_farquhar

@ai_fast_track @PyTorch @OpenMMLab @wightmanr Really cool to be a part of this: airctic.com Awesome work team!!

Super excited to announce 🎉IceVision 0.7.0🎉 We super-charged the new API: ✅Multi-Task Training to efficiently combine object detection, segmentation, classification models ✅MMDetection, Torchvision, and EfficientDet Support ✅Data auto-fixing @PyTorch @OpenMMLab @wightmanr

@svpino I did something very similar to this for @fastdotai v2 a year ago. Of course heavily inspired by @aureliengeron book Chapter 17. github.com/jerbly/fastai2… - results after only 30 epochs.

Your boy here pushed an example to the official Keras Code Examples repository: "Convolutional Autoencoder For Image Denoising." keras.io/examples/visio… It's an autoencoder to remove noise from pictures, and it does a fantastic job with the MNIST dataset.

@robmarkcole @ai_fast_track @PyImageSearch From that website this looks like a solved problem. 2500 lookups per month for free!

@jerbly @ai_fast_track @PyImageSearch For sure 2 stage. This is also necessary since there might be multiple plates in a single image. Several datasets on platerecognizer.com/number-plate-d…

I haven't blogged for a long time and this one makes up for it! jeremyblythe.blogspot.com/2021/02/practi… Practical ML end-to-end project: labelling, training, inference, serverless deployment. #Azure #azurecloud @ai_fast_track

@robmarkcole @ai_fast_track @PyImageSearch So maybe a two stage? Find the plate, crop and send that to the object detector. Know any good datasets?

@jerbly @ai_fast_track @PyImageSearch That said, many CCTV cameras are high def now (but not mine) so once you’ve got the number plate in a bounding box, the actual letters might not be that small

@robmarkcole @ai_fast_track @PyImageSearch I see, similar experience I had with my first attempt to solve my problem - hence why I switched to an object detection approach. I imagine in the number plate case though there may be the small-objects problem?

@jerbly @ai_fast_track @PyImageSearch Thanks for the link. I have tried out tesseract before but accuracy was poor for real number plates, which tend to be imaged on grainy CCTV footage

@robmarkcole @ai_fast_track Yeah, maybe. There's some pretty amazing stuff from @PyImageSearch here that you should look at if you want to read number plates: pyimagesearch.com/2020/09/21/ope…

@jerbly @ai_fast_track I guess it would only require a small tweak to do number plate reading?

I just added a small update to the post with this screenshot. It's pretty amazing what you can achieve with such a small dataset. This is the entire training set!