MTCNN to scan this packed room of builders and located {47} faces. Not that good, but not that bad.

A long time ago I worked at a small startup and we started having problems with our office WiFi. Every once in a while it would go out. I was the CTO and I did all kinds of network troubleshooting, resetting the wifi router to factory settings and trying all kinds of

Why don't competitive programmers use C? -> C is beautiful for systems programming, but in CP, time and reliability are everything.

Trying to segment noisy images without markers is kind of like Snoopy trying to save cookies with a firehose. A total mess. Switch to Adaptive Watershed. Mark the targets, ignore the noise and, eat those cookies. #LearnAiWithSnoopy @Snoopy

While my mask2former model was training, I built a multiplayer space-shooter game (with a little help from Gemini). Didn’t launch it... just exposed it through tunneling. The whole thing took under an hour.

Your dataset is imbalanced. You have plenty of normal images but almost no defects. What do you do? Anomaly detection has the answer. New blog on why flipping the problem works better: https://t.co/2KpFieDzK5

Let's learn about "Perspective Projection [PP]": When you take a picture of something like a tiled floor or a road, you’ll notice how the lines seem to come closer together and the tiles look smaller as they go farther. PP is the way a camera turns 3D world to a flat 2D image.

I beg you to try your best to fix your life just for once and lock in for 3 months. once you see a glimpse of your true potential, you will find it hard to go back to your current pathetic self. once you earn some money, taste some freedom and feel some recognition, you will

Om Saravana Bhava🙏

This film is not meant to hurt anyone’s feelings. It is a humble attempt to recreate what might have happened that tragic night. May all the departed souls rest in peace. ॐ शान्तिः (Om Shanti) #Kurnool #Bus #BusAccident

We need your help 🤍 The demo of our relaxing tower-building game launches in a few weeks. We’re a tiny indie duo just trying to get a bit more visibility on Steam before then. A like and retweet would go a very long way 🙏

Let’s build a simple neural network! With 4 inputs, 3 neurons in the hidden layer & 2 outputs. Here’s how it looks in PyTorch: (@grok is amazing!!)

Happy Diwali 🪔🎇

Eagerly waiting for Ai films!!

Hot storage keeps frequently accessed data ready for instant use (e.g., data in AWS S3 Standard or a database on SSDs), whereas cold storage preserves rarely used data safely for the long term at lower cost, such as AWS Glacier or tape backups.

Check out "Deep-ML". A LeetCode-style platform for hands-on ML & data science practice! link: https://t.co/aFCIsRlyHj

This is it. This is the stuff that actually matters. It’s one thing to make cool apps, but it’s another thing entirely to push science forward and potentially save lives. It's incredibly powerful to see that happening in real-time.

Some crazy people on Reddit managed to extract the "SynthID" watermark that Nano Banana applies to every image. It's possible to make the watermark visible by oversaturating the generated images. This is the Google SynthID watermark:

Story time: After I moved out of my home in 2014 to live in Japan, initially I used to call my parents daily. But as I became more "internationalized" I started calling my parents less and less. I used to tell my mom "I have nothing to talk about so we don't need to call daily".