I just migrated a React Native app to @expo that has been in production for 7 years
So nice so see 186k (!) lines deleted from the project!
This is mostly because of CNG which auto-generates the ios and android folder, so that I could remove them from git
The model predicts the angle of the hour, minute and second hands, which I again translate to a timestamp.
The model is exported as .tflite and used in my app through the amazing react-native-fast-tflite by @margelo_com!
Detection of time from analog watch achieved🤖⌚️
1. Live on-device watch detection
2. Take image and crop around the watch
3. On-device custom made ViT model for detecting the time shown on the analog watch
Now users will have a better default value when measuring their watch!
The time detection model is based on a pre trained timm model from @huggingface, and fine tuned on a large image dataset of watches I have created with my app
https://t.co/VK7CXHVL9K
And here it is implemented in the app! 📱
Object detection model (trained on self-labeled, anonymized data from users) detects the watch ➡️ camera focuses on the center of the detection.
After the image is taken, a final detection is done to crop around the watch 👌
🔎Auto focus on a object
Today I trained an object detection model to find wrist watches with @ultralytics YOLO, imported it in a fresh @expo app, an hooked it up with react-native-vision-camera and react-native-fast-tflite from @margelo_com
Auto-focus on watches achieved👌📸
I wanted to create this setup to improve the experience in my app, https://t.co/uf76mv5Q2V, so that the camera always focuses the watch.
This will give my users the clearest possible image for checking the accuracy of their watches ⌚️
🔎Auto focus on a object
Today I trained an object detection model to find wrist watches with @ultralytics YOLO, imported it in a fresh @expo app, an hooked it up with react-native-vision-camera and react-native-fast-tflite from @margelo_com
Auto-focus on watches achieved👌📸
We just released react-native-runtimes! 🥳
👉 Render React components on separate JS Runtimes
👉 Share state (Zustand) across Runtimes, synced in C++
👉 Dual-Thread-approach for rendering/business logic
We plugged this into MetaMask which made the app start substantially faster:
Pro tip for all Codex users:
use the built in imagegen skill in Codex to have it create images fitting you product based on the color scheme and the style in you codebase!🖼️
Just used it for my new pay wall:
https://t.co/U15oni7yGo
Doing an experiment where I upgrade my pay wall with these changes:
1. Feature aware: first shown premium feature is the on the user actually pressed
2. Generated feature image: showcase feature better
3. Less legal writing: use links instead
What would you change?
Doing an experiment where I upgrade my pay wall with these changes:
1. Feature aware: first shown premium feature is the on the user actually pressed
2. Generated feature image: showcase feature better
3. Less legal writing: use links instead
What would you change?
Doing an experiment where I upgrade my pay wall with these changes:
1. Feature aware: first shown premium feature is the on the user actually pressed
2. Generated feature image: showcase feature better
3. Less legal writing: use links instead
What would you change?
First ASO experiment: translate to 15 languages🌐
People more often search in their native language, so localizing the app makes sense for better reach.