I'm not sure about the rebranding of products, but some of the features like the learning path creation and coding challenges definitely look interesting.
If this leads to SO regaining some of its popularity, I'm all for it!
I've bought a bunch of Philips products over the years, and always hated the fact that a simple thing like a charger breaking would lead to throwing away the entire thing- especially grooming products like trimmers. This release of 3D printable models definitely feels like a step in the right direction.
I think being snarky about little things isn't really promoted on HN- it's the first guideline for commenting here. Be kind and try to answer the question if you do have an answer.
Cognitive Search is nowhere as good as a 'pure' vector DB. Behind the scenes, it's a managed elasticsearch/opensearch with some vector search capabilities. The 'AI' implementations I've done with Cognitive Search always boil down to hybrid(vector+fts) text search.
In context of RAG, the goal is not to have a pure vector DB but to have all the relevant data that we can gather for a user's prompt. This is where Cognitive Search and other existing DBs shine because they offer a combination of search strategies. Hybrid search on Cognitive Search performs both full text and vector queries in parallel and merges results which I find a better approach. Further, MS is rebranding Cognitive Search as Azure AI Search to bring it more in line with the overall Azure AI stack including Azure OpenAI.
Cognitive Search already contains hybrid search (vector + BM25 + custom ML reranking) and they use chunks of 2048 tokens with a custom tokenizer. So it should be now better than most vector DBs. One could probably make something better by using some version of SPLADE instead of BM25 but their secret sauce lies in their custom ML model for reranking that gives them the largest search performance boost.
You can pretty much solve this using modern DL models. There are options depending on how accurate you want your model and how much compute you have.
There is an entire spectrum of models, from something like Mask-RCNN, U-Net family upto something like Meta's SAM, which you can use without even training.
In this kind of cases I'd probably use the Hough-based algo as ground truth to see if you can indeed fine-tune a DNN on that regression task. If it does with reasonable accuracy, then you have a baseline that could be improved in multiple ways to surpass the original.
That said, there are not that many shapes of speedometer and wheels, and the view point is likely controlled, so your old school method is probably the better way ;)
For the purpose of learning, would you recommend some tutorials, articles or videos that help achieve that? Accuracy aside, this would make a great learning experience!
Is it better to look in the PyTorch community, or that's where some Tensorflow approaches shine? (CUDA is ok)
PyTorch is much nicer to play with in my opinion. Maybe start with their official tutorial, I've also heard good things about Karpathy's YouTube channel from beginners.
I'll recommend the Spotlight paper by Google[1]. There are very interesting datasets they created for this purpose. They mention they have a screen-action-screen dataset that is in-house and it doesn't look like they'll open it. Maybe owning Android has its advantages.
There's a recent paper by Huggingface called IDEFICS[2] that claims to be an open source implementation of Flamingo(an older paper about few-shot multi-modal task understanding) and I think this space will be heating up soon.
If this leads to SO regaining some of its popularity, I'm all for it!