Train Otaku Update 2024 06 15

Current Location: Edge Esmeralda, Drafts and Demo weekend.

• Decided to work more on project, because there are some machine learning domain experts here.
• Used Roboflow auto labeler to begin labeling. 1 frame per second, for a total of 174 images.
• Only label applied is train-car. This allows further image processing and analysis to be done in separate pipeline, ideally using a VLM.

Goals for weekend

• Count cars, extract one image per train car. Output some stats.
• Create 1 minute demo video
• Write philosophy about of visual sampling.
• create video with song

Outcome

• Counted some train cars. It's not perfect, likely need more/better training data. Using a custom trained YOLOv8 model to recognize individual train-cars
• Created a remix of Bremer/McCoy – Op,