Have you seen this rather astonishing work by @citnaj ? (Quite a bit more than 100 lines - but well worth it considering the beautiful results :) )https://t.co/EInKW7b78W
— Jeremy Howard (@jeremyphoward) March 16, 2019
Have you seen this rather astonishing work by @citnaj ? (Quite a bit more than 100 lines - but well worth it considering the beautiful results :) )https://t.co/EInKW7b78W
— Jeremy Howard (@jeremyphoward) March 16, 2019
Nice work by Emil Wallner. ~100 lines of @TensorFlow to train model to color B&W images.
— Jeff Dean (@JeffDean) March 16, 2019
"Bananas are easy because they’re almost always yellow & have a unique shape. Moons & planets can be more confusing because of similarities they share w/each other"https://t.co/ZsOmZVryqu
Why ReLU networks yield high-confidence predictions far away from the training data and how to mitigate the problem #CVPR2019
— ML Review (@ml_review) March 14, 2019
w/ @maksym_andr
Proposes a training scheme to mitigate the problem.https://t.co/02IcXP3I1z pic.twitter.com/gymyMIWrzN
Code to reproduce the domain transfer experiment in “Deep Learning for Classical Japanese Literature” paper.
— hardmaru (@hardmaru) March 13, 2019
The model will take a pixel image of an old style Kuzushiji Kanji and try to predict how to write the modern version as a sequence of pen strokes. https://t.co/xqbd8g7SXH pic.twitter.com/ujFFOU0lim
Holistically-Nested Edge Detection with OpenCV and Deep Learning via @pyimagesearch https://t.co/VeRbri5Jnw #python #opencv #deeplearning pic.twitter.com/dFP6dIY3hz
— Python Weekly (@PythonWeekly) March 12, 2019
Imagenette too easy for you? Then try your hand at Imagewoof - just like Imagenette, but with more woof! Creating a doggie classifier from scratch with just 1000 images won't be easy...
— Jeremy Howard (@jeremyphoward) March 7, 2019
If you try it out, let me know! (But don't use a pretrained model.)https://t.co/joqArlFxvC pic.twitter.com/FsCv87qLjk
Here's how they knew a baseball might fool their shark detector:
— Janelle Shane (@JanelleCShane) March 6, 2019
"In this example we see another detector that seems to be playing two roles: detecting red stitching on a baseball and a sharks’s white teeth and pink inner mouth." pic.twitter.com/skov3wAkTt
Presenting Imagenette. A smaller subset of 10 easily classified classes from Imagenet, and a little more French.
— Jeremy Howard (@jeremyphoward) March 6, 2019
I've found this quite helpful in my research and will be using it in our teaching. I hope you find it helpful too! :)https://t.co/GFUJ9a5QTk pic.twitter.com/fKtPPhIbei
No GAN is complete without Ramen 🍜 https://t.co/w0M38XsAR5
— hardmaru (@hardmaru) March 5, 2019
I would like to announce that @albuslaev @creaf @cvtalks @viglovikov are releasing image augmentation library #albumentations v0.2.0.
— Vladimir Iglovikov (@viglovikov) March 4, 2019
The new features are key point transformations, multiple targets of the same type and many other improvements.https://t.co/jng9RljvPd pic.twitter.com/LsDOIbdbml
Classification / metric learning using @fastdotai!
— Radek Osmulski (@radekosmulski) March 1, 2019
The notebook features:
✅custom loss (combination of cross entropy and contrastive loss)
✅sampling progressively harder datasets
✅all the @fastdotai goodies (discriminative lrs, one cycle)https://t.co/bicTSuTU8x pic.twitter.com/kAdijrUUCy
This is a really great overview of many nice approaches to self-supervised learning for images https://t.co/aq9gbbIWYn
— Jeremy Howard (@jeremyphoward) February 28, 2019