CDAO posts computer vision CSO

On May 31, the Chief Digital and Artificial Intelligence Office (CDAO) posted the CENTCOM Computer Vision Defense Commercial Solutions Opening. Responses are due by 5:00 p.m. Eastern on June 30.

The CDAO is seeking demonstrations for commercial platforms to enable DoD personnel to rapidly create and maintain computer vision (CV) capabilities. Intended applications include automated detection and identification of features in imagery and video collected by spaceborne, airborne, and land/sea-based sensors. Desired solutions will support the following supervised learning functions:

IMAGERY/VIDEO INGEST – The capability must support manual and batch ingest of data from local or cloud data stores, inclusive of pertinent domain metadata (e.g., collection timestamp, sensor details, image processing details, geotags, content tags, synthetic content tags, etc).

BATCH DATA PREPARATION – Example functions include spatially-subdividing large images or temporally-subdividing videos for labelling purposes.

LABELED DATA SET INGEST – The capability must support ingesting existing labeled data sets, provided they conform to common labeled data set standards (e.g., COCO).

CV TASK AND ONTOLOGY GENERATION – CV tasks include classification, as well as single- and multi-class object detection and instance segmentation for imagery and video.

LABELING – The capability must manage distributed workforce labeling and secondary quality reviews. An optional, yet highly desirable, feature is incorporation of novel methods to enhance labeling efficiency. Examples include pre-labeling (i.e., machine-generated labels) and machine-assisted labeling (e.g., human-generated point automatically converted to polygon).

AUTOMATED LABEL AUGMENTATION – The capability must boost the size of a labeled dataset through intelligent batch transformations thereof (e.g., mitigate class imbalance).

AUTOMATED MODEL TRAINING – Model training must flexibly utilize cloud or on-prem GPU compute resources.

AUTOMATED MODEL SCORING – Performance scoring must be based on industry standard CV metrics (e.g., F1, precision, recall), contrasting findings from test and validation (i.e., withheld from training) sets.

MODEL CONTAINERIZATION – The capability must export a containerized model, inclusive of the model network, weights, and supporting metadata for traceable provenance. The containerized model must be fully viable as an air-gapped capability, with zero external system dependencies.

Review the CDAO computer vision CSO.

Source: SAM

The right opportunity can be worth millions. Don’t miss out on the latest IC-focused RFI, BAA, industry day, and RFP information – subscribe to IC News today.