Kidus's Chronicles

9,000+ images. Fewer than 30 annotations per image. 2 straight days training an image detection model. And I somehow burned through my NVIDIA credits in 5 hours 💀

I shared one of my covers. She have insane voice guys

What do you mean such voice exists?? Unearthly

Here is my first ever official vlog, hope you enjoy it❤️ Ill be doing this everyday so be ready 😇, its kinda scary to vlog in public tbh gn remember kids, the best things in life comes out of discomfort

What would you guys want me to talk about? I have a lot of interest and a lot of things to talk about. I really need your help in this

now on this, the detector picks up the hand in real time, which overlays the 21 mediapipe landmarks, classifies the gesture and shows the confidence score instantly. but there is a lot to work on.

this is how the training goes, accuracy is 98.36 percent, and the confusion matrix shows most misclassifications happen between visually similar hand shapes. which is expected and honestly the interesting problem to solve as more letters are added. only 10 letters for now 😂 since i am teaching myself sign language letters to even train it

Something I have been working on. I am building a sign language detection model for the Amharic Alphabet, for now Fingerspelling. Still in training, but the pipeline is taking shape and I wanted to share what's going on under the hood. The approach basically: Mediapipe for hand landmark extraction, feeding 21 key points per hand (x, y and z coordinates) into a classifier. Instead of raw image frames, the model works on the landmark vectors. which means it's lighter, faster, and more robust to background noises and especially lightning conditions. Ethiopian Sign Language is massively underrepresented in ML, Even the normal language itself, I was actually involved in a project called Jigglypuff which is basically working on the Data side in training Amharic NLP Link and I understood Most sign language datasets that exist are for ASL or BSL. Building something that actually works for our alphabet means collecting data, labeling it, and training from scratch and no transfer learning shortcut here. Still ironing the accuracy on similar-looking hand shapes, which is honestly the hardest part 😭. Some letters are just brutal to distinguish at inference time.

Goooood morning

Lesson Of The Day V18.0 For the past year, I’ve sat with myself every single night to ask: What did today teach me? I’ve shared those whispers with you here and as my 18th year comes to an end I wanted to wrap them all up into a gift for you. I’ve collected every single lesson including the ones from my breaks in April and May into this book. I hope you can open it to any page and find a little bit of light or comfort for your own day. Make sure to listen to the custom YouTube music playlist I made for you while reading through the pages. If these lessons have been a companion to you and you’d like to keep the tea warm while I write the next book, you can support my journey here ☕️ I’m so incredibly grateful to have you all in my world. The book is officially yours😊 consider it birthday gift i gave y'all https://lesson-of-the-day-book.vercel.app/ @lidsverse

If you're a Telegram Creator, please help this message reach everyone 🤪 The plan is to get rejected for all the reasons App Store Guidelines say are non-compliant features or UX flaws as much as possible in the external beta review. That helps me avoid friction on the actual App Store submission. And it means it may take me a week or so to deliver the external beta (first version and build) for those of you who sent me email addresses. Meanwhile, you can fill in this form to be one of the Ledgerly Early Adopters: https://forms.gle/Q6TjQHfjm1xqf6Ux5

Gooooood morning beautiful people And Eid mubarek to all of my Muslim friends, may this holiday bring prosperity and happiness to your homes. Have a wonderful day y'all

#DailyReminder Your death will come on an ordinary day, in the middle of unfinished plans, and the world will continue without you. So live a little.

142 people... wow. 🖤 I just noticed the number, and I wanted to say thank you to every single one of you for joining. It might seem small to some, but to me it means a lot. Thank you for being here and for supporting this little journey. More to come. ✨

I was on the road and people were looking at me a lot when i took these 😂

Photos i took today, I am spamming you guys

You know the world is so beautiful, we sometimes forget that. I am taking a walk and everything around me is beautiful, the sun, flowers, the smell of grass and rain, people each existing in their own little world, I love it.

Every woman deserves a Safe Place to run to