ryanwold.net - Civic Tech, Bitcoin, and Open Government Innovation

Title

Text

I discovered Roboflow on Twitter,
where [skalskip92](https://x.com/skalskip92/status/1925611059257180425) was posting screenshots of NBA games with players being tracked using computer vision. Colored squares appeared around each player, and as the players moved, the bounding boxes move along with the identified player.

I looked into the company and was impressed to see the activity, product, and company offering computer vision as a service.

I've been a fan of motion graphic packages (like the animated graphics played on ESPN sports broadcasts).

I've thought about similarities between motion graphics and closed captions. These are both types of metadata overlays to video; and when the data becomes visible, the additional dimensions become explicit to see.

---

### Paths of exploration

* upload a public meeting video and train a model to identify board members, public speakers, and 3 different camera views.
* install the `inference-sdk` in python and run it locally to install a docker image. ran the Taylor Swift record example, and the bleacher comparison example. Queried the YOLO classifier. Queried the Microsoft SONO? classifier. Queried OpenAI.
* Use Elixir Livebook to call the Python `inference-sdk` successfully.

#### Playground: https://playground.roboflow.com/open-prompt

* When developing locally, `inference start` will create a Docker image. Once the Docker image is created, the Roboflow Inference web application runs locally and is available at http://localhost:9001.

### Ideas

Ideas about use cases I might apply computer vision to.

* Book inventorying
* Virtual item scanner -> inventory
* Screw counter
* Public meetings
  * Identify how much time a person is speaking
  * Count the number of public speakers
  * Determine how many decisions get made.
* Yolo tagging as a service

---

Based on https://blog.roboflow.com/inference-python/, 
I ran it locally and was able to see face detection running locally.

I ended up creating a couple](https://github.com/roboflow/inference/pull/1345) [tiny](https://github.com/roboflow/inference/pull/1346) PRs,
and reported a [bug](https://github.com/roboflow/inference/issues/1347) that somebody else seems to have experienced as well. I couldn't find the React code in the stacktrace, otherwise, I'd have looked into addressing it.

An entry