Brock Purdy had a perfect passer rating (158.3) in the 2nd half on 9/11 passing for 185 yds and 2 TD
Patrick Mahomes (2021 vs Steelers) and Purdy are the only QBs with a perfect passer rating in a half in the last 5 postseasons (min. 10 attempts)
On the Twelfth Day of #Shipmas, Roboflow gave to me: YOLOv8 train and deploy.
We added support for the brand-new SOTA #YOLOv8 model to Roboflow. You can export datasets in the right format, train with our notebook, and share/deploy with a single command! https://youtube.com/watch?v=wuZtUMEiKWYโฆ
On the Eleventh Day of #Shipmas, Roboflow gave to me: Image Tagging.
Organizing your images just got a lot easier: you can now add, sort, and filter your datasets by custom-defined metadata. Add image tags during upload, via API, from search results, or in the annotation editor.
The NFL owes such a big debt of gratitude to whoever invented fantasy football and incentivized people to watch so many games they otherwise wouldnโt have cared about.
82 NFL games in the top 100 most-watched TV shows of 2022
That football emoji at #35 is the 4-10 Broncos losing to the 4-10 Rams by 37 points on xmas... which got more viewers than any NBA game, any MLB game, any World Cup Game, the Oscars, or any scripted TV show
On the Tenth Day of #Shipmas, Roboflow gave to me: Label Assist for Segmentation.
You can now use segmentation models to pre-annotate your images (and then use Smart Polygon to refine the predicted shapes). This greatly speeds up the annotation process for active learning.
Can anyone explain to me in simple terms why the length of a sequence of right angles doesnโt converge as it approaches a straight line? (I get that it _doesnโt_, but intuitively this seems a lot like the geometric explanation for calculating the area under a curve.)
Nice read on reverse engineering of GitHub Copilot . Copilot has dramatically accelerated my coding, it's hard to imagine going back to "manual coding". Still learning to use it but it already writes ~80% of my code, ~80% accuracy. I don't even really code, I prompt. & edit. twitter.com/parth007_96/stโฆ
(2019 on)
โข the time I took in 2019 to learn ML (hacking in notebooks off GitHub)
โข that ramping the team enabled me to code more this year than the first 2.5 years of
I wrote a script that
- pulled @_akhaliq's last 7 days of tweets
- fished out the arxiv links
- downloaded raw paper .tex
- parsed out intros & conclusions
- automated a podcast dialogue about the papers w/ web automation & GPT
- generated a podcast
https://scribepod.substack.com/p/scribepod-1
Welcome to Roboflow's 12 Days of Shipmas!
We're ending 2022 strong and kicking off 2023 with a bang by shipping 12 features in 12 days.
#Shipmas starts now. Check back daily for updates.
I used the #StableDiffusion 2 Depth Guided model to create architecture photos from dollhouse furniture.
By using a depth-map you can create images with incredible spatial consistency without using any of the original RGB image.
See twitter.com/karenxcheng/stโฆ
OpenAI just dropped a prototype of 3D DALLE (called โPoint-Eโ) . It isnโt as good as Googleโs DreamFusion, but blazing fast! Like ~600x faster to generate . 2D DALLE has already turned the creative world upside down. How will 3D DALLE disrupt games, VR, metaverse, โฆ?
Twitter's signup captcha seems like an example of something that's hard/annoying for humans and easy for computers... You have to do 5 of these to pass.
Gave up on testing their onboarding flow after failing it twice.
Language models may โjustโ be string manipulation, but life is โjustโ chemical reactions, and chemistry is โjustโ physics. Surprising complexity emerges from interactions of simple primitives.
LMs are not likely sentient, but dismissing the possibility out of hand seems unwise. https://twitter.com/emilymbender/status/1535845774843473922โฆ
Most object detection papers benchmark against MS COCO; but better performance on a large dataset of common objects doesn't necessarily translate to better performance on small datasets of infrared images.
Today we're launching RF100, 100 real-world datasets across many domains.
NEW! Roboflow 100: A Multi-Domain Object Detection Benchmark
We compiled 100 datasets from our community spanning a wide range of domains, subjects, and sizes to be used for benchmarking state-of-the-art computer vision models.
Learn more: https://rf100.org
Continued..
8 hours into a hack and I'm still wrestling to get conda to properly install my dependencies in an arm64 lambda docker environment. This shit doesn't happen in node.
Hosting an event with friends to hack on LLMs and AI apis.
Attendees of the first event will get early private access to the DALL-E API on Tuesday 11/1/22 in SF.
Please share with SF builders and sign up to attend: http://hackingmodels.com
Roll up! It's time for the next #MadeWithTFJS show and tell this Friday at 8pm PT to see new #TensorFlow#WebML demos in action! Come join the fun, pre chat will start 20 mins before for early birds, so click sub / notify here so you don't miss out:
https://buff.ly/3yNuM6n
Found an optimization that took an operation in Roboflow from 1000 seconds down to 3 seconds.. Iโm both stoked about the 300x speed up and embarrassed it took 8 months of it living in prod to realize it shouldnโt be so darn slow!