The purpose of this document is to provide descriptions of six core experiments on MPEG Compact Descriptors for Video Analysis (CDVA). The previous three CEs have been reviewed, and based on the responses received and the open issues they pose, one additional CE on matching and retrieval has been added, and the CEs related to deep-learning based descriptors have been split. The description of all CEs has been improved, adding more details on the experiments and expected results.
The results of experiments will be discussed on the reflector before the 119th MPEG meeting.
The report of each CE should include; (1) a comparison between the tested solutions, (2) recommendation to the AhG based on the results of the CE.
The results according to the evaluation criteria described below shall be reported (in particular those defined by the evaluation framework), but results based on other evaluation criteria may be included.
If learning-based methods are used, then the CDVA data set must not be included for training. The data sets used for training shall be reported. The details of the network configuration being used shall be included in the CE response (or as a reference to an input document that contains these details).
