Development Updates #
Short updates on VFRAME development progress, bug fixes, model releases, and news to provide transparency into project development and activity.
- Retrain the 9N235/9N210 detectors with new data to improve recall for damaged and partial objects and reduce false positives.
- Update 9N235 detector models to version 1D on demos.vframe.io
- Separate SHA256 media deduplication script to new command group
dedupand small usability improvements for handling added and removed media items (example:
vf dedup sha256 -i path/to/videos/)
- Moved modelzoo.vframe.io –> demos.vframe.io
- Updated the 9N235/9N210 nano and small architectures with version 1D. This update reduces false positives and improves recall on heavily damaged 9N210/9N235 submunitions based on reference photos from Ukraine.
- Submit research report to James Madison University journal on building the 9N235 detector and applicability of this technology to mine clearance and EOD work
- Remove verbose from
media-attrs-summarizefor simpler summaries
media-attrswith more manual binning interval and size options. The auto-binning yielded arbitrary intervals.
- Updated site to add work in progress for the Uragan, RBK-250, and RBK-500
- Improved clarity on 9N235 detector page
- Twint (for scraping Twitter) now seems to be fully broken
media-attrsout of development into public code. See research post on plotting media attributes
- Updated website, adding search and model development previews for current research
- Begin changelog for more project development transparency
- Released updated 9N235 detection models in n/s/m/l model architectures for YOLOV5 framework in .pt and .onnx formats