New zero-shot model for object and scene recognition
Our latest innovation: a zero-shot model for object and scene recognition. This model includes an extensive collection of pre-trained labels and can now be customised using a separate dictionary.
All objects, including non-real objects such as cartoon characters, animals, cosmetics, etc., can be identified using a customised dictionary without the need for training data. Possible applications include the targeted placement of advertisements in specific locations.
New differentiation option for training data
Speech recognition transcript translation now available in API and user interface
All DeepVA users can now translate their speech recognition transcripts into 27 languages. Previously only available in the API, from 22 August it will also be available in the user interface. This opens up new possibilities for global communication and efficiency.
New face recognition model with class IDs to avoid name duplicates
What else is new? Bug Fixes & Co.
- Fixes a problem where Unicode decoding errors were not handled correctly for custom dictionaries. Note: Please use the default UTF-8 encoding for your dictionaries to avoid error messages.
- Usability improvements for subtitle detection and speech recognition results
- Correction of an issue where speaker dataset creation failed due to missing readout data
Try the latest version of DeepVA and experience the power of AI-driven analytics with advanced customization and improved performance. Discover the limitless possibilities and get deeper insights with DeepVA’s cutting-edge features.
All DeepVA changelog updates are available here: https://docs.deepva.com/changelog/