VideoLINCS BAA questions Round 2_26 JUL 24_C.pdf

VideoLINCS Q&A

Round 2, IARPA BAA 24-04

1. Question: Given the extensive information required in the Technical Volume limit of 30 pages, would you consider removing the Transmittal Letter and/or Cover Sheet from the Technical response page count? On the submission portal, the Transmittal Letter must be submitted as a separate document. Thank you.

Answer: No, IARPA will not be modifying the format, and the Transmittal Letter and/or Cover Sheet with remain within in the 30-page limit as stated in the BAA. Please reference again Section 4.B.1 for complete details on the formatting requirements.

2. Question: The answer to Q66 states that the pre-trained models that were trained outside of Video LINCS program do not need to be shared with other performers and training data for those models do not need to be shared across T&E and performer teams. Can you clarify if this data-sharing exception relates to the pre-trained foundation models for which training data may not be easily available or does this data-sharing exception also applies if the models are trained using data from other IARPA programs, such as DIVA/MEVA?

Answer: As stated in the prior Q&A, pre-trained models that were trained in the past outside of the Video LINCS program do not need to be shared with other performers and training data for those models do not need to be shared across T&E and performer teams. See that response for additional details. However, this data-sharing exemption does not include models trained using IARPA funded datasets such as DIVA/MEVA. If such models are used for Video LINCS, the IARPA funded data used for training would need to be shared across the program, even if those models were trained prior to Video LINCS.

3. Question: Will IARPA be sharing the related data generated through other IARPA programs (e.g., DIVA/MEVA) across the participating teams?

Answer: There should be no expectation of sharing of training data from IARPA. The MEVA dataset, as of 7/17/2024, is an open AWS dataset.

4. Question: The solicitation mentions that >50% of distinct truthed people will be <150 pixels high. Does the percentage refer to the maximum observed height of unique persons, i.e., 50%of unique people will never be observed with >150 pixels high resolution. Or does the percentage refer to the total number of observations across all unique persons?

Answer: As stated in Section 1.2, "Nominal test set conditions" are provided and "The exact conditions are subject to potential update during program execution". The T&E team will determine the conditions under which these percentages apply, with the objective, as stated in Section 1.A.1.a that "recognizable faces will predominantly be unavailable throughout the video collection."

5. Question: What are expected sizes for vehicles and generic objects?

Answer: As the BAA states in Section 1.A.1, "The objective is for Video LINCS to be effective in the open-world setting where there is no advance knowledge of the sensor types, sensor collection geometries, scenes, objects to be re-identified, etc." As a result, there are no constraints on expected sizes.