AI Suisse is Switzerland's largest group of professionals with focus on artificial intelligence. In our regular meetups and online-only conferences, we invite top speakers to share insights into how AI can be used in various industries, what works, and whatnot. We are also Switzerland's first ambassador for deeplearning.ai, which was founded by Andrew Ng, a renowned expert in the field. Join our community at meetup.com/de-DE/Artificial-Intelligence-Suisse to stay informed about upcoming events and conferences. Hope to see you soon! - Steffen Konrath Founder, AI Suisse www.linkedin.com/in/steffenkonrath/
I was running NER using flair and genia datasets (18MB texts data train contain 18.000 sentences) , and it stopped when training the datsaets in google clab free (8GB RAM + GPU), any suggestions?
Thanks for sharing. A note to Michael. This is indeed potentially useful application. I like you implemented the approach end to end, all steps from detection to distance calculation for a (life) video stream. The downside to me is the use of proprietary SAS software in between, that does not really add much value to what you could do with only open source. As a matter of fact, most of the community works on open source only, and if you really want this to spread people need open source code. I believe that implementing the thing fully open source could be a nice project!
@iuoi oiuu Thank you for already pointing out most of the things I would have covered now. Of course you can always build any kind of solution with open source software. In fact, many functions from professional software companies also build on open source. The difference is that professional vendors also thought about how to provide an infrastructure that is reliable and scalable - an important factor if you're in the professional world. The way I've built the demo allows you to easily scale up to any number of cameras, you don't have to worry about data source connection. You can also dynamically change models (newer version, day/night, different weather conditions...) and most importantly transfer the results to any kind of analytical solution. Don't forget: Analyzing the images is only a step to derive information from images. The more important part is the integration into your (business)-process and combination with other data, e.g. to create reports, generate alerts, perform modelling and forecasting. Nevertheless, I think you'd like to play around with the demo by yourself. If you carefully go through my code, you'll notice that it is very easy to replace the SAS parts. In less than 30 minutes I managed to implement your project idea utilizing Yolo in Darknet via it's Python API. If you're interested in trying this out, you can check my GitHub repository where I've uploaded a 100% open source version. github.com/Mentos05/OpenSource_DeepLearning/tree/master/Social%20Distancing%20Demo But keep in mind: You loose everything that makes this usable from a professional standpoint. The solution doesn't scale up, it only connects via OpenCV to local files, it doesn't save the output. It is nothing more than an interesting demo - but definitely nothing you would use in your production system. Additionally, I am so kind to share my GPU-enabled docker container for my open source developments. That way you don't have to fiddle with the difficult setup of all the requirements you need which can easily cost you hours if not even days. (Another thing that professional software vendors will manage for you) Just pull it from dockerhub. Note: You'll need an NVIDIA GPU, latest drivers and CUDA 10.2 github.com/Mentos05/OpenSource_DeepLearning/tree/master/GPU%20accelerated%20docker%20container I hope you now understand why I've used SAS for some parts. It's not like I didn't want to produce the demo with open source - it's just a matter of who your target audience is. And mine are usually bigger professional companies and organizations that need reliable products. Best Regards Michael