Skip to main content
. 2022 Jan 3;3:799067. doi: 10.3389/fdgth.2021.799067

Figure 2.

Figure 2

The proposed transfer learning framework :CovNet. (A) Parameters of the first n convolutional layers/blocks (convs1) of the current COUGHVID model are frozen and initialised by the corresponding first n convolutional layers/blocks (convs0) of the pre-trained FluSense model. (B) Embeddings are extracted after the n-th convs0 of the pre-trained FluSense model. The extracted embeddings are concatenated or added to the current embeddings generated after the n-th convs1 of the COUGHVID model.