🤗 Transformers meets VISION 📸🖼️

v4.6.0 is the first CV dedicated release!

- CLIP @OpenAI, Image-Text similarity or Zero-Shot Image classification
- ViT @GoogleAI, and
- DeiT @facebookai, SOTA Image Classification

Try ViT/DeiT on the hub (Mobile too!):https://t.co/swIGmI6G0d pic.twitter.com/nOs8PEfmzo

— Hugging Face (@huggingface) May 12, 2021

Where Americans moved during the #COVID19 pandemic. Yellow = net loss, purple = net gain. #dataviz

Source: https://t.co/vgM8Tr1T2X pic.twitter.com/qpOFhYTuyZ

— Randy Olson (@randal_olson) May 12, 2021

Self-Supervised Learning with Swin Transformers
pdf: https://t.co/uNAcu1JViD
abs: https://t.co/J70bvOq6nc
github: https://t.co/5dktODVxa9

a self-supervised learning approach called MoBY, with Vision Transformers as its backbone architecture pic.twitter.com/4yJm1AWHkh

— AK (@ak92501) May 11, 2021

I was a bit overwhelmed by volume of datasets on https://t.co/8DOFywgAbp (thanks, @ramonaisonline!), so I built a quick @observablehq notebook to search through them! https://t.co/fbHsmY8fT7

— Mike Freeman (@mf_viz) May 11, 2021

This #dataviz overview table is used by the @FT team to decide which visualization to use. It's a handy starting point. HT @martinstabe pic.twitter.com/k9RSUBJqCh

— Simon Kuestenmacher (@simongerman600) May 10, 2021

Contrastive Learning for Unsupervised Image-to-Image Translation
pdf: https://t.co/tiRPgcY0ic
abs: https://t.co/OoyrZ6ERqQ

learn a discriminator that differentiates between distinctive styles and let the discriminator supervise a generator to transfer those styles across images pic.twitter.com/PigLyp2795

— AK (@ak92501) May 10, 2021

ResMLP: Feedforward networks for image classification with data-efficient training
pdf: https://t.co/0cTkeTwwne
abs: https://t.co/DA1zkiYq0S

an architecture built entirely upon multi-layer perceptrons for image classification pic.twitter.com/7VmEyovIkW

— AK (@ak92501) May 10, 2021

Are Pre-trained Convolutions Better than Pre-trained Transformers?
pdf: https://t.co/8L06XiPM1C
abs: https://t.co/gIAq2Od5GA

experimental results show that convolutions can outperform Transformers in both pretrain and non-pre-trained setups pic.twitter.com/IZtYlBXvkc

— AK (@ak92501) May 10, 2021

I just came across this underrated paper:

Continual Learning in Practice (@tommy_da_cat et al)https://t.co/XfyK26phKe

It talks about why this is hard and what a reference architecture could look like.

— Josh Tobin (@josh_tobin_) May 8, 2021

Over the last 20+ years I've learned a few things about creating tech startups - here's my attempt to summarize those learnings in one hour.https://t.co/MxmkYxX5CG

— Jeremy Howard (@jeremyphoward) May 7, 2021

Another chart for your collection of "ridiculous things that the US is divided on based on political lines". Sourced from @538politics: https://t.co/AEMmN7EEhy pic.twitter.com/MdW0cZmyLW

— Simon Kuestenmacher (@simongerman600) May 6, 2021

Important analysis on vaccine equity in CA, and how modelling choices can have a huge impact. Using ZIP codes rather than census tracts means >2 million Californians, living in neighborhoods w/ worst health outcomes, may not receive needed additional vaccine supply. https://t.co/eUfvkJIqpz

— Rachel Thomas (@math_rachel) May 6, 2021