Releases: Terkwood/vision
🌳 Upgrade yew
Set coco-ssd version to 0.1.1 📓
coco-ssd lib is now on version 1.0, which is presumably compatible with only TFJS 1.0+. We fixed the version of coco-ssd to 0.1.1.
Fix text placement for large bounding boxes
Fixes the placement of the captions when drawing large bounding boxes, so that the text is visible to the user.
Add image download button
Adds a button which provides image downloads of the object detection results.
Shrink WASM size, add privacy notice, add bounding box demo
This release includes modifications which shrink the size of the generated WebAssembly binary, adds a notice about the app's privacy policy, and adds a small bounding box demo based on Paul Irish's multi-touch work. (See the README for attribution.)
The privacy policy is stated when you land at the app's front page. Put simply, all images and video handled by this app will stay local to the device, and will never be transmitted to third party services.
This is an educational demo app, and we have no desire to engage in marketing pursuits. The backbone of this app is tensorflow.js, a product released by Google. Our understanding is that tensorflow.js is limited in scope and won't spy on anyone. Most of the work contained within this project consists of simple user interface controls which allow the object recognition algorithm to be run on pics captured by the user.
Fix behavior in Chrome, add startup instructions
Run object detection using device camera
This release predicts bounding boxes for images captured using the device's camera and the javascript getUserMedia API.
Adds camera button
Adds a clickable camera button attached to a trivial greeting.
Fixes initialization
The app now initializes cleanly with the example image. The code has been cleaned up a bit.
Proof of concept
This simple example invokes coco SSD to infer bounding boxes for a provided image.