Thomas Wolf

Go to conferences or hang out on some community slack or discord

The best is to explore the possibilities with your team and infrastructure or DevOps teams. In many case there are a lot of aspects that you may not have a visibility on and which will be essential to assess the best tool: which compute provider are your using, which maintainance tools are used, how is the cost-benefit of the migration looking, what are the next step on the horizon/roadmap, what are the price and budget constrains and do they need to be updated.

In many case you can learn more in a 10 minute discussion with a colleagues in another company or related industry than by reading 1 hour of news on social media. Use your network to stay up-to-date with newer trends in your field and don't settle for your current skills. The world of tech and research is an ever changing landscape that you need to keep exploring with the joy of learning new things.

Collaborative aspect is essential in annotation. In most case you'll want to be able to work and scale your annotation task on-demand and this requires a tool which is well able to handle team work as well as the related authorization management and possible race-conditions and more complex dynamical team size extension. Don't forget to have clear procedure to evaluate the tool and some target metrics before-hand or a clear comparable product in order for your trail to be as effective as possible.

Some of the most important aspects for a good annotation tools are: - good UX: intuitive and fast, limiting the number of operation per annotation as much as possible - covering well your use-case - output data should be very easy to extract and as customizable as possible An ideal annotation tool should be as invisible as possible: fast, easily integrated in your workflow and transparent. An open-source annotation tool has the advantage that you can more easily customize it if needed as well as inspect the underlying internals.

Always start with simple and well understood method before you dive in the latest research. Also remember that research publication follow strange laws that have nothing to do with real use-case: scientist want to publish complicated methods and stories because they are more impressive to conference reviewer. Very often a much simpler method can give similar or even better results. The same hold with negative results: it's very hard to publish and share what has not been working, try to keep it in mind to fill the gap when reading a recent technical report.