11/
We also ship a benchmark dataset of Golgi morphologies and use it to compare image featurization tools: #ConvNeXt, #SubCell, #CellProfiler

Scale‑Space Modeling of Master Filters in Neural Networks

Researchers found eight recurring ‘master key’ filters in depthwise‑separable ConvNeXt layers and showed Gaussian‑based versions can replace them with comparable benchmark performance. Read more: getnews.me/scale-space-modeling-of-... #convnext #gaussianfilters

https://blog.stackademic.com/road-sign-recognition-using-convnextv2-96c8dd4c65c1?source=rss----d1baaa8417a4---4 ConvNeXt v2 is the latest evolution of the ConvNeXt family, blending the efficiency of classic convolutional networks with design…

Introducing ConvNeXt v2, the latest advancement in convolutional networks, combining classic efficiency with innovative design. #ConvNeXt #AIvolution blog.stackademic.com/road-sign-recognition-us...

F5-TTS

#F5TTS: New non-autoregressive #TTS system using flow matching with #DiT. Features #ConvNeXt for text representation & Sway Sampling for improved performance. Achieves 0.15 RTF, trained on 100K hours multilingual data. #AI #Speech 🗣️🔊

Demo: swivid.github.io/F5-TTS