btcmx

joined 10 months ago
sorted by: new top controversial old
[D] Resources for learning in c/machinelearning@academy.garden
[–] btcmx@alien.top 1 points 10 months ago

This post has a number of ML topics/skills you might (sooner or later) run into. In a sense it's kind of a roadmap/blueprint containing what you need to become an ML Engineer, (beyond learning a few topics/basics from a one(or more) courses.

1
[D] Is Computer Vision brighter than ever? Foundation Models are really reshaping CV (alien.top)
 

I keep diving and finding GPT-4V prototypes shared on X: e.g. narration for videos (source), posture correction (source), etc.

As foundation models in computer vision become even more accessible, will the field recover some attention (wrt to LLMs hype)?

[P] Model training bottlenecked by CPU. in c/machinelearning@academy.garden
[–] btcmx@alien.top 1 points 10 months ago

As other have rightly pointed out, verify you're using the Data Loader the right way. Ideally you need to create a custom dataset (in PyTorch terms) and apply all the transformations in this custom dataset. This might be helpful. Also, have you tried PyTorch Lightning?

[D] People who work on computer vision models on the edge, what devices do you deploy to? in c/machinelearning@academy.garden
[–] btcmx@alien.top 1 points 10 months ago

For CV, Jetson family! The following source has a high level overview of challenges and solutions when deploying NVIDIA TAO on Jetson devices.

1
[D] Can Diffusion Models really be Steered? Not really! Not even ControlNet, ICCV23 Best Paper (alien.top)
 

While diffusion models (e.g. Stable Diffusion) are all the rage, they don't seem to be prepared for downstream tasks. ControlNet looks great (on paper), but open-source implementations for mere mortals aren't ready for prime time.

Do you have examples that show the contrary? Will FAANGs and not-really-open Research Labs be the only ones capable of making it happen?