Each month we bring you the high level bullet-points of improvements and additions to Roboflow. If you missed it, last month's changelog is here.
In March we were focused on expanding Roboflow's inference capabilities with the release of on-device NVIDIA Jetson support, our open source video inference library, a Python webcam inference demo, and example code snippets in 8 programming languages.
- Updated onboarding flow to customize user experience based on user persona.
- Improved chat-bot triggers to be less intrusive.
- Added ability to revoke API Keys.
- HEVC/H.265 video detection & alert.
- Detect image upload issues (eg firewall block) and surfaced helpful message.
- Added Upload API examples in 8 programming languages.
- Updated our GPU infrastructure and achieved 3.4x faster training times.
- Backend work to support upcoming deployment targets.
- Improved early stopping heuristics.
- Launched support for deploying models to NVIDIA Jetson.
- Decreased size of on-prem inference-server Docker by 40%.
- Added video inference.
- Released webcam inference example.
- Refactored to support an arbitrarily large number of models with no performance penalty.
- Improved cold start times for model inference.
- Better error handling.
- Added new public model to Label Assist.
- Improved default zoom value.
- Enabled "Show Source" to see transformed images' original source.
- Fixed 9 bugs.
- Welcomed Thomas to the team as our new Full Stack Developer (he has already become our most-prolific developer as measured by number of commits)!
- Joint webinar with Paperspace.
- Added utilities and telecommunications industry pages.
- 22 bug fixes.
- 14 blog posts.
- 8 YouTube videos.
- Featured on/at: ODSC, General Assemb.ly, The Darius Gant Show, 1 Million Cups, The Current State, General Assembly, FIRST Robotics.
- Fun user projects: recycling sorter, autopilot tester, Raspberry Pi cyclist monitor, smart hospital supply closet, Bangalore pothole reporter, PPE detector.