New version: v3.0.0-beta.2 , need help with testing

Releasing v3.0.0-beta.2 🎉️ , featuring:

• Support for YOLOv4 (set "NEURAL_NETWORK": "yolov4" in config.json)
• Upgraded docker setup and build workflow, see https://github.com/opendatacam/opendatacam/issues/178 , now you can run the docker container from jetson without needing to hack the support for CUDA… It also simplified the Dockerfile file and made the build of OpenCV happening
• Support for RaspberryPi cam v2 (or any CSI cam compatible with nvarguscamerasrc) when running from the docker container for Jetson nano. Just set “VIDEO_INPUT”: “raspberrycam”
• Improved performance by 6 FPS for Linux CUDA desktop machine by using cuDNN for the docker build
• Made ports configurable from config.json
• Add a configurable confidence threshold in config.json to filtrer out detections that are below this threshold. (default keep > 20% confidence)
• Add favicon
• Add “confirmation” modal when deleting recordings
• Add version number in side menu
• Updated all javascript dependencies: Next.js , Redux, React etc etc…
• Revert back to using the main ./darknet API as memory leak has been also fixed

How to install

I did the build for Jetson nano, ubuntu machines, if you have a TX2 or Xavier, sorry but you should be able to install it without docker. . 🚨️ Make sure to reinstall darknet from the odc branch as specified in the doc

For the docker installation, still the same procedure (Make sure to remove your remove your previous config.json, install-opendatacam.sh, run-docker.sh, run-opendatacam.sh before installing new version)

# Download install script
wget -N https://raw.githubusercontent.com/opendatacam/opendatacam/v3.0.0-beta.2/docker/install-opendatacam.sh

# Give exec permission
chmod 777 install-opendatacam.sh

# NB: You will be asked for sudo password when installing the docker container

# Install command for Jetson Nano
./install-opendatacam.sh --platform nano

# Install command for a Nvidia-docker machine
# NB: Will run from demo file, you can change this after install, see "5. Customize OpenDataCam"
./install-opendatacam.sh --platform nvidiadocker

Next up

We aim to release v3.0.0 in ~ 1 month or before as we are waiting that Jetpack 4.4 final lands along with the Jetson Xavier NX… And maybe we will also be able to integrate tiny-yolov4 if it is released by then.

• Update darknet fork once again using the new -dontdraw_bbox flag AlexeyAB kindly added for us : https://github.com/AlexeyAB/darknet/commit/da050367a8b174f751a2c9720195d51a369c4a4a
• Expose tracker config IOU and UNMATCHED FRAME tolerance
• Check if config.json version matches Opendatacam version
• Clean up code for drag& drop feature… maybe add some UI tweaks like a drop indicator and a specific progress message
• Update API docs with new counters attributes
• Improve installation process, maybe ask people if they want to run from camera / CSI cam or run from file at install time… or maybe add something to change this in the UI
• Update documentation for Jetson nano with Raspberrypi CSI cam
• Document how to run from custom weights + maybe add a way to use custom weights with docker build
• Document gstreamer use and pipelines
• Documentation for yolov3-tiny-prn and yolov4
• Document tweaking width height size in yolo cfg
• Add Thresshold to filter out large detections
• Update OpenMojis
• Add a “ethics” message
• Change picture of OpenDataCam with something with the jetson nano and not the tx2 ?

Complete changelog from v2.x to v3.x

• Memory leak fixed on jetson’s, expected running time is weeks now instead of hours ( https://github.com/opendatacam/opendatacam/issues/182 )
• Drag & drop support for file, just drop a file on the UI and it will restart to process it
• Directional counters, see: https://github.com/opendatacam/opendatacam/issues/122#issuecomment-603733234
• Save / Reload counters config: see https://github.com/opendatacam/opendatacam/issues/122#issuecomment-604952995
• Support raspberrypi cam (CSI cam) for Jetson nano
• Support for YOLOv4
• Made ports configurable from config.json
• Add favicon
• Add delete confirmation modal for recordings
• Add version number in side menu
• Support raspberrypi cam (CSI cam) for Jetson nano
• More reliable counting
• Add a configurable confidence threshold in config.json to filtrer out detections that are below this threshold. (default keep > 20% confidence)
• Updated all javascript dependencies: Next.js , Redux, React etc etc…
• Several smaller bugs fixes, clean up logs etc etc…

Thanks for your help testing : @shams3049 @rantgithub @Denizzje @b-g

[tdurand]

@b-g quick note about openmojis, I just updated all the icons of OpenMojis to 12.3.0 but I didn’t use the npm package (sorry !) … as it just adds unecessary complexity from my opinion… I would need to add some script that copy the svgs from the node_modules to the public folder of the next.js app… (or maybe I didn’t understand how to use the package 😉 ) . see https://github.com/opendatacam/opendatacam/commit/cfb5cfb47b438421f0ace130e962be1fd505e5c1

[tdurand]

FYI AlexeyAB gently included a fix into darknet for us that we were maintaining on our branch: https://github.com/AlexeyAB/darknet/commit/da050367a8b174f751a2c9720195d51a369c4a4a , will include in next version OpenDataCam