Fast Scale-able Video Understanding for Video Infrastructure

dashboard password: geostl

Inspiration

wanting to use marengo embeddings and pegasus understanding on full videos but faster with more custom pre and postprocessing.

What it does

any video uploaded is sent through 5 neural network pipelines. 1 for marengo clip search, 1 for marengo embeddings to find the frame in the video most relevant. one for pegasus understanding. a custom yolo for powerline detection and a custom yolo for insulator and powerpole segmenetation

How we built it

We built it on AWS using terraform for IaC

Challenges we ran into

Very little data, needing custom labels for training.

Accomplishments that we're proud of

Full AWS deployment. All features working. 5 deployed neural network pipelines with automation.

What we learned

video embeddings

What's next for Fast Scale-able Video Understanding for Video Infrastructure

productionalization

Built With

• javascript
• python
• terraform