Yeah, I know, it has been a while I posted something. Somehow, lock downs tend to screw with your project schedules and blogging, which is usually at the bottom of the pile, gets pushed under the desk…
Anyhow, not posting was not because nothing was happening, but there was too much! So, here is a little overview of what I have been up to (in no particular order):
- simple-redis-helper – simple command-line tools to sending/receiving data to/from a redis backend, which I mainly use for testing other frameworks that use redis for exchanging data, like deep-learning pipelines.
- redis-docker-harness – this little library reduces the amount of boiler plate code that I need to write when adding redis support to our docker images (command-line handling, containers for passing through options, etc).
- requests4j – now has added support for making handling of JSON easier (sending and receiving).
- rsync4j – has been receiving regular updates, whenever cygwin released rsync or openssh updates; but it also added support for sshpass, for feeding in passwords into rsync/ssh commands.
- knn-undersampling (Weka package) – a user on the Weka mailing list didn’t know how to incorporate a code dump, so I created a Weka package (and, boy, did I need to fix that code first…).
- wai.tflite_model_maker – based on public example code from Tensorflow, I created command-line wrappers for the tflite model maker library (and, of course, some docker images as well).
- mmdetection – finally, finally got around to upgrading my docker images to use the 2.x branch of the mmdetection object detection framework.
- shallowflow – inspired by the ADAMS workflow engine, I started a simple Python-based workflow framework. It is only a side project, but being in Python-land, could make direct integration of Python-based deep-learning frameworks easier. Why? I hate copy/pasting code over and over again, something that seems accepted practice in data scientist circles. I prefer reusable operators that can be parametrized to duplicated code – simply because of only having to fix it in one location (the operator) rather who knows how many scripts. I’m just lazy… 😉
- video-frame-selector – is a little wrapper around our object-detection frameworks (i.e., run via docker images) that allows feeding in frames from videos, either generating a new video with the frames to keep (i.e., frames that had objects of interest in them) or just simply outputting the frames.
And that was just the outcome of the actual work (like data processing and modeling) that I was doing at the time. 🙂