Vidsy.ai is a tool for indexing, searching and analyzing visual data. It uses an LLM to convert text searches into structured queries that are run against the results of various CNNs.

Key features:

• real-time camera analysis: multiple cameras can be analyzed in sync

• record video on multiple tracks in sync

• re-analyze recordings

• search through local images and videos

• share your findings with friends and colleagues

As this release is still a beta, there are some limitations:

• only supports a local file system based db, for smaller datasets

• the number of included models currently used for analysis is limited and is only related to faces.

• LLM usage is currently limited to 10 queries a day (a previously retrieved query can be re-used unlimited)

• max nr of bytes shared is limited at 20GB a month

• probably still a few bugs

What's coming next:

• more models for object search and such

  • stage 1 will be the more conventional model types (most likely Resnet based)

  • in a second stage I'm planning to work out another idea round CNNs in the hopes to make it easier to teach the system how to recognize things.

• support for external db systems (mongodb), which will also allow multiuser use-cases (share indexes)

• flesh out some more features that make things more convenient.

• cloud support for data processing so that things can scale with large camera pools.