What I see is the following processing of raw HTML when creating an EPUB:

1. Tags of the form:

    are processed to some extent insofar as the description of the file is added to the resulting content.opf manifest and the image is copied to the EPUB/media/ directory

2. <svg></svg> tabs are also partially processed: their presence in a file (book chapter) automatically causes pandoc to add "properties=svg" qualifier to the file's description in the manifest. On the other hand what lives inside the svg wrapper is ignored.

In case s/o runs into the same issue my best effort to workaround this difficulty:

1. add <img ... /> raw HTML referencing the images to the md source: causes pandoc to add them as 'media-type="image/png" (e.g.) to the manifest and copy the png/jpg files to the resulting epub.

2. add dummy <svg></svg> empty entries to all files/chapters that reference images so that pandoc marks the output files with the "properties=svg" attribute in the conten.opf manifest (stops epubcheck complain)

3. run pandoc

4. unzip the epub and batch edit the files in the EPUB/text/ directory... a lengthy but simple regex will replace all the <img /> by <svg></svg> wrappers + <images /> and zip the stuff back into place as if nothing happened.

Questionable... but easier than patching the code.

Not pandoc-related but for those who have a vested interest in epub's there is a lengthy not too technical article called Image handling in ePubs reaches new level of inanity by one M. A. Demers one her site: (mademers.com)