๐ฑ New iPhone, Who Dis? At @photoroom_app , we kicked off the iPhone year with our, now famous, annual on device ML benchmark by @Olotiar Thread ๐
OMG. Mojo AI is ๐คฏ
What @jean_patry and the team are building is SO COOL ๐
Here is a video I created, with 0 edits, using Mojo AI
Stay tuned โ or DM me if you can't wait a few weeks ๐
Iโm super excited to share the first video Iโve recorded with the team at @photoroom_app ๐
Watch my colleague @Olotiar explain how you can make any object property animatable ๐ฅ
(even when that object is not part of the view hierarchy!)
https://t.co/d0RWODF5mI
@v_pradeilles Had to check. In the end, no difference in the resulting assembly besides the presence of the symbol corresponding to the setter for `age`.
@pocketpixels Yeah I agree, it is really interesting; We are leveraging the fact that iOS apps run natively on M1 Macs, so it's the exact same code that is compiled in both cases. We feed directly the MLModel instance by creating a MLFeatureProvider.
Today is new-iPhone day! At @photoroom_app it means it's CoreML benchmark day. We have made it a habit to benchmark new hardware for on-device machine learning.
TLDR: OS version has little impact. Consistent incremental improvements generation after generation of iPhones in all cases, but the iPad Pro is an incredible device for on-device ML.
This systematic speed increase will allow us iOS developers to think of cool new use cases.
@pveugen @matthieurouif We have a benchmark app we've been refining over the years; it generates a performance report of some of the models we use in the iOS app, but we trigger it manually
A few years ago, after struggling to install CUDA and OpenCV dependencies for the 53th time, I moved all my projects to Docker.
I just wrote a short illustrated article to explain how to easily package a deep learning project in Docker โคต๏ธ https://t.co/s3YwV62oPH
@marcoarment@parrots Modern Bluetooth will struggle to reach those speeds. Theoretical asymmetric data rate is 2.1Mbps (so only 48kHz 16 bit would fit) and even that is without taking interferences into account.