Notate ML vs Orquesta

Preparing input image datasets is a critical and challenging task in any Machine Learning project.  Notate ML brings the power and usability of Apple's mobile devices to accelerate this task and deliver higher quality training data to your object detection model. FEATURES ・Create a dataset, type, scan, or speak your labels into it ・Snap new pictures, or import old ones from your photo library ・Crop images, draw bounding boxes, tag and label objects of interest ・Export images and annotations for training with YOLO, Apple Create ML, or Google Auto ML CREATING A DATASET ・Create a new data set and give it a unique name ・Add labels in one of several ways: ・One at a time, using the keyboard ・Copy and paste a comma-separated list from another document ・Scan from a physical text source ・Use the device voice feature to speak them in ・Labels can be edited or deleted by swiping left on each row ・Only unused labels can be deleted IMPORTING IMAGES ・Tap into the dataset and import photos (up to 50 at a time) from your Photo Library ・This app can only access photos that you select using Apple's photo picker feature CAPTURING IMAGES FROM THE CAMERA ・While browsing the images in a dataset, tap the "Capture" button to snap new images with the camera ・The app will require your permission to access the camera ・If you choose to decline, you can always grant access at a later date using "Settings" ANNOTATING THE IMAGE ・Select an image from the dataset image browser to annotate it ZOOMING and CROPPING ・Pinch with two fingers to zoom in or out of the image ・Drag with two fingers to move (pan) the image about ・Optionally crop the image to the visible square dimensions ・Cropping can only be done when the image has no bounding boxes. ・Double tap on the image to reset zoom and recenter the image ・Tap the "Reset" button in the bottom panel to restore image and annotations to the original state before editing ・The cropped image dimensions and visible area dimensions in pixels are displayed above the image at all times TAGGING ・Drag with one finger or stylus to draw bounding boxes around objects of interest ・Long press on a box to select an existing annotation ・Use the picker to assign a label to the selected bounding box ・Or tap the red Delete button to remove the selected box ・Tap anywhere outside the selected box to unselect it ・Tap the "Reset" button to restore image and annotations to the original state before editing ・Tap the "Done" button to complete annotating the image EXPORTING THE DATASET ・While browsing dataset images, tap the "Export" button, to select export options ・Datasets can be exported in one of three "schemas" for use with different object detector training frameworks ・The YOLO schema generates a set of files organized for use with the YOLO framework ・The Create ML schema generates files to use as input to Apple's Create ML application ・The Auto ML schema generates files to use as input to Google's Auto ML service ・ After reviewing the export options, tap the "Export" button on the Export options screen to generate data files for training ・ Choose one of the sharing options (e.g. Airdrop, Files) enabled by your device to export the data files DELETING ・Delete an image by swiping left in the dataset image browser ・Delete an entire dataset by swiping left in the dataset browser ・If a dataset contains images, you will be prompted for confirmation Some tips ・Crop the images before tagging them ・Use a stylus for more accurate bounding boxes ・Maintain different datasets for training and validation ・Limit datasets to under 1000 images each, for ease of use ・Export and delete unused datasets to conserve storage

Orquesta is the companion app for the Orquesta platform. It puts the projects you already run from getorquesta.com in your pocket, so you can see what your coding agents are doing without opening a laptop. WHAT YOU CAN DO • Projects — every project in your organization, with its agent's live online status and last heartbeat. • Prompts — the full history of what has been asked of each agent, with status, model, cost and the git branch and commit the work landed on. Queue a new one from the project screen. • Live logs — open any prompt and follow the agent's timeline as it runs: reasoning, tool calls, tool results and output. • Sessions — attach to an agent's interactive session and drive it from the phone: a live terminal with a key bar for esc, tab, arrows and ^C, so you can answer a prompt that is blocking the work instead of finding it stalled hours later. • Team chat — the same channels and threads as the dashboard. Mention a teammate, or address the agent by its @handle and it picks the message up. • Activity — one feed across every project, so you can see the whole organization at a glance. • Inbox — billing, team, deployment and system notifications, markable as read. • Star and tag prompts to keep the ones worth coming back to. SIGNING IN This app is for people who already use Orquesta. It needs an existing Orquesta organization — you cannot create an account from the app. Signing in opens getorquesta.com in your browser to authorize the device, and the resulting token is stored in the device's own secure credential store. You can revoke it at any time from Settings, which deletes the local copy and invalidates it on the server. Other devices keep working. Settings also links to account deletion, which permanently closes your Orquesta account and the data it owns. GOOD TO KNOW Agents run on your own machines, not on your phone. A prompt you queue is picked up by the agent on its next heartbeat, then runs for as long as the work takes — this is asynchronous by design, not a live chat with a model. If a machine is offline, the work waits for it to come back. Interactive sessions are the exception: those are live, and they show what happens from the moment you attach rather than what already scrolled past. The app reads your organization's data and can queue prompts, star and tag them, post to team chat, and mark notifications read. It does not have broad write access to your projects.