Month: February 2025

AI/ML

Google Coral TPU and TensorRT (Frigate + NVIDIA GPU/TensorRT)

These are two majors which allow to run object detection models. Google Coral TPU is a physical module which can be in a form of USB stick. TensorRT is a feature of GPU runtime. Both allows to run detection models on them. Coral TPU: And TensorRT: Compute Capabilities requirements CC 5.0 is required to run DeepStack and TensorRT, but 7.0 to run Ollama moondream:1.8b. Even having GPU with CC 5.0 which is minimum required to run for instance TensorRT might be not enough due to some minor differences in implementation. It is better to run on GPU with higher CC.

AI/ML

AI video surveillance with DeepStack, Python and ZoneMinder

For those using ZoneMinder and trying to figure out how to detect objects, there is deepquestai/deepstack AI model and builtin HTTP server. You can grab video frames by using ZoneMinder API or UI API: You need to specify address, monitor ID, user, password. You can also specify single frame (mode=single) or motion (mode=jpeg). Zoneminder uses internally /usr/lib/zoneminder/cgi-bin/nph-zms program binary to grab frame from configured IP ONVIF RTSP camera. It is probably to the most quickest option, but it is convenient one. Using OpenCV in Python you could also access RTSP stream and grab frames manually. However for sake of simplicity

Technology

Hetzner Cloud with pfSense gateway

Instead of using dedicated Hetzner servers you can use their cloud solutions for less demanding use case scenarios. The idea is to have one server for pfSense and the other one for regular private use, hidden behind gateway. It will be little bit cheaper than renting dedicated server as CX22 costs around 3 – 4 Euro per month comparing to 50 Euro at least for dedicated server. You can have up to 10 virtual cloud server in a price of single dedicated server. Choose cloud for “simple” (or rather straightforward) solutions, dedicated servers for much complicated setups with much more

Hobby Technology

Design and 3D-print wall switch in OpenSCAD

So we decided to try Zigbee wall switches and many of available brands work with Fibaro. But not all of them and not in all configurations. Even if some switch or module works it may only work with Tuya gateway and application and not with Fibaro gateway. So there are some blanks in this concept. We left with double frame with electronic switch and one blank for shutters as there is no compatible switch or module at this time. So I thought: maybe I just 3D-print it… OpenSCAD code: Few test desigs: Let’s print them: And here we go! Now

Technology

NextCloud 26 update failed

I use NextCloud as alternative to Google Photos. My smartphone sends pictures over OpenVPN to my NextCloud instance. However for over a year Android application has been showing notification that server version is outdated, and from time to time automatic upload do not work. So I decided that it is the time to do an upgrade. It did not work well: I tried to run upgrade using UI. It got stuck in various places, and finally I got this: But there is much simpler option. Install fresh copy of NextCloud. I prefer using VM ISO (https://www.turnkeylinux.org/nextcloud) instead of LXC containers.

Technology

Elasticsearch & Kibana: version 8.x installation

At Indatify we use PostgreSQL for most of the time when dealing with data. We tried to use Cassandra, but for now it is too much of a constraint in such dynamic data environment. We put it on shelf. However, for textual data we choose Elasticsearch, because we know it and it provides full text search out of the box. Later, we will come back to Cassandra, but with more specified use case, as it requires precise data model to be predefined by query and not by structure. So, to install Elasticsearch (if running without sudo, then run from root):